0% found this document useful (0 votes)

137 views

Building Skills in Object-Oriented Design 2.1.1 Python

Uploaded by

Tr Tzara

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

137 views

Building Skills in Object-Oriented Design 2.1.1 Python

Uploaded by

Tr Tzara

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 317

Building Skills in Object-Oriented Design

Release 2.1.1-Python

Steven F. Lott

December 30, 2009

CONTENTS

Front Matter
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3
5 5 6 7 9 9 10 11 13 13 15 15 16 17 22

1 Preface 1.1 Why Read This Book? . . . . . 1.2 Audience . . . . . . . . . . . . . 1.3 Organization of This Book . . . 1.4 Why This Subject? . . . . . . . 1.5 Programming Style . . . . . . . 1.6 Conventions Used in This Book 1.7 Acknowledgements . . . . . . . .

2 Foundations 2.1 Problem Statement . . . . . . . . . . 2.2 Our Simulation Application . . . . . . 2.3 Soapbox on Use Cases . . . . . . . . . 2.4 Solution Approach . . . . . . . . . . . 2.5 Methodology, Technique and Process 2.6 Deliverables . . . . . . . . . . . . . .

Roulette

3 Roulette Details 29 3.1 Roulette Game . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.2 Available Bets in Roulette . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.3 Some Betting Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 4 Roulette Solution Overview 4.1 Preliminary Survey of Classes . . . . . . 4.2 Preliminary Roulette Class Structure . . 4.3 A Walkthrough of Roulette . . . . . . . . 4.4 Roulette Solution Questions and Answers 5 Outcome Class 5.1 Outcome Overview . . . . . . . . 5.2 Design Decision Object Identity 5.3 Outcome Design . . . . . . . . . . 5.4 Outcome Deliverables . . . . . . . 5.5 Message Formatting . . . . . . . . 6 Bin Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 35 36 37 38 41 41 41 44 45 45 47 i

6.1 6.2 6.3 6.4 6.5

Bin Overview . . . . . . . . Design Decision Choosing Bin Questions and Answers Bin Design . . . . . . . . . Bin Deliverables . . . . . . . . . . . . . . . . . .

. . . . . . . . A Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

47 47 48 49 50 51 51 52 53 54 55 55 56 59 59 60 61 61 61 62 63 64 65 67 67 68 69 69 71 71 72 73 74 74 75 77 77 78 79 79 80 81 83 83 85 86 87

7 Wheel Class 7.1 Wheel Overview . . . 7.2 Wheel Design . . . . 7.3 Non-Random Design . 7.4 Wheel Deliverables . 8 Bin 8.1 8.2 8.3 8.4 8.5

Builder Class Bin Builder Overview . . . . . . . . . Bin Builder Algorithms . . . . . . . . BinBuilder Design . . . . . . . . . . . Bin Builder Deliverables . . . . . . . Internationalization and Localization

9 Roulette Bet Class 9.1 Roulette Bet Overview . . . . . . . . . . . . . . . . 9.2 Design Decision Create or Locate an Outcome . . 9.3 Design Decision Where to Keep the Outcome Map 9.4 Roulette Bet Questions and Answers . . . . . . . . 9.5 Roulette Bet Design . . . . . . . . . . . . . . . . . . 9.6 Roulette Bet Deliverables . . . . . . . . . . . . . . . 10 Roulette Table Class 10.1 Roulette Table Overview . . 10.2 InvalidBet Exception Design 10.3 Roulette Table Design . . . . 10.4 Roulette Table Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11 Roulette Game Class 11.1 Roulette Game Overview . . . . . . . . . . 11.2 Passenger57 Design . . . . . . . . . . . . . 11.3 Roulette Game Design . . . . . . . . . . . . 11.4 Roulette Game Questions and Answers . . 11.5 Roulette Game Deliverables . . . . . . . . . 11.6 Additional Roulette Design Considerations 12 Review of Testability 12.1 Testability Overview . . . . . . . . 12.2 Test-Driven Design . . . . . . . . 12.3 Capturing Pseudo-Radom Data . 12.4 Testability Questions and Answers 12.5 Testable Random Events Design . 12.6 Testability Deliverables . . . . . . 13 Player Class 13.1 Roulette Player Overview 13.2 Player Design . . . . . . 13.3 Martingale Player Design 13.4 Player Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14 Overall Simulation Control 14.1 Simulation Control Overview . . 14.2 Simulation Terms . . . . . . . . 14.3 Simulator Design . . . . . . . . . 14.4 Player Rework . . . . . . . . . . 14.5 Simulation Control Deliverables 15 SevenReds Player Class 15.1 SevenReds Player Overview . . 15.2 SevenReds Design . . . . . . . 15.3 Player Rework . . . . . . . . . 15.4 Game Rework . . . . . . . . . 15.5 SevenReds Player Deliverables 16 Statistical Measures 16.1 Statistics Overview . . 16.2 Some Foundations . . . 16.3 Statistical Algorithms . 16.4 IntegerStatistics Design 16.5 Statistics Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

89 89 89 91 93 93 95 95 96 97 97 97 99 99 100 101 102 103 105 105 106 106 107 107 108 109 110 111 111 111 112 112 113 113 113

17 Random Player Class 17.1 Random Player Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.2 Random Player Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17.3 Random Player Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Player 1-3-2-6 Class 18.1 Player 1-3-2-6 Overview . . . . . . . . . . . 18.2 On Polymorphism . . . . . . . . . . . . . . 18.3 Player 1-3-2-6 Questions and Answers . . . 18.4 Player1326 State Design . . . . . . . . . . . 18.5 Player1326 No Wins Design . . . . . . . . . 18.6 Player1326 One Win Design . . . . . . . . 18.7 Player1326 Two Wins Design . . . . . . . . 18.8 Player1326 Three Wins . . . . . . . . . . . 18.9 Player1326 Design . . . . . . . . . . . . . . 18.10 Player 1-3-2-6 Deliverables . . . . . . . . . 18.11 Advanced Exercise Refactoring . . . . . . 18.12 Advanced Exercise Less Object Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19 Cancellation Player Class 115 19.1 Cancellation Player Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 19.2 PlayerCancellation Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 19.3 Cancellation Player Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 20 Fibonacci Player Class 20.1 Fibonacci Player Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.2 PlayerFibonacci Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20.3 Fibonacci Player Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Conclusion 117 117 118 118 121

iii

III

Craps
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

123
127 127 128 128 131 131 133 133 134 134 135 137 137 139 139 140 140 141 141 142 143 144 144 145 146 147 149 149 150 150 151 152 152 155 155 156 157 158 159 161 161 163 163 164 166

22 Craps Details 22.1 Craps Game . . . . . . . . . . . . . . . 22.2 Creating A Dice Frequency Distribution 22.3 Available Bets . . . . . . . . . . . . . . 22.4 Some Betting Strategies . . . . . . . . . 22.5 Wrong Betting . . . . . . . . . . . . . . 23 Craps Solution Overview 23.1 Preliminary Survey of Classes . . . . . 23.2 Preliminary Class Structure . . . . . . . 23.3 A Walkthrough of Craps . . . . . . . . 23.4 Craps Solution Questions and Answers 24 Outcome Class 24.1 Outcome Overview . 24.2 Overloaded Methods . 24.3 Outcome Rework . . 24.4 Outcome Deliverables 24.5 Advanced Exercise . . 25 Throw Class 25.1 Throw Overview . . . 25.2 Throw Design . . . . 25.3 Natural Throw Design 25.4 Craps Throw Design . 25.5 Eleven Throw Design 25.6 Point Throw Design . 25.7 Craps Game Design . 25.8 Throw Deliverables . 26 Dice 26.1 26.2 26.3 26.4 26.5 26.6 Class Dice Overview . . . Throw Rework . . . NumberPair Design Dice Design . . . . . Dice Deliverables . . Dice Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27 Throw Builder Class 27.1 Throw Builder Overview . . . . . . . . 27.2 Outcomes with Variable Odds . . . . . 27.3 Refactoring The Outcome Hierarchy . . 27.4 Soapbox on Subclasses . . . . . . . . . 27.5 Soapbox on Architecture . . . . . . . . 27.6 Throw Builder Questions and Answers 27.7 Soapbox on Justication . . . . . . . . 27.8 Design Light . . . . . . . . . . . . . . . 27.9 Design Heavy . . . . . . . . . . . . . . . 27.10 Common Design . . . . . . . . . . . . . 27.11 Throw-Builder Deliverables . . . . . . .

28 Bet Class 167 28.1 Bet Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

28.2 Bet Rework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 28.3 CommissionBet Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 28.4 Bet Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 29 Craps Table Class 29.1 Craps Table Overview . . . . . . . . . . . . . . . . 29.2 Design Decision Table vs. Game Responsibility . 29.3 Design Decision Allowable Outcomes . . . . . . 29.4 CrapsGame Stub . . . . . . . . . . . . . . . . . . . 29.5 CrapsTable Design . . . . . . . . . . . . . . . . . . 29.6 Craps Table Deliverables . . . . . . . . . . . . . . 30 CrapsGame Class 30.1 Game State . . . . . . . . . . . . . . . . 30.2 Game State Class Hierarchy . . . . . . 30.3 Resolving Bets . . . . . . . . . . . . . . 30.4 Moveable Bets . . . . . . . . . . . . . . 30.5 Design Decision Win, Lose, Wait . . . 30.6 Additional Craps Design . . . . . . . . 30.7 Craps Game Implementation Overview 30.8 Throw Rework . . . . . . . . . . . . . . 30.9 ThrowBuilder Rework . . . . . . . . . . 30.10 Bet Rework . . . . . . . . . . . . . . . . 30.11 CrapsPlayer Class Stub . . . . . . . . . 30.12 CrapsGameState Class . . . . . . . . . 30.13 CrapsGamePointO Class . . . . . . . . 30.14 CrapsGamePointOn Class . . . . . . . . 30.15 CrapsGame Class . . . . . . . . . . . . 30.16 Craps Game Deliverables . . . . . . . . 30.17 Optional Working Bets . . . . . . . . . 31 CrapsPlayer Class 31.1 Craps Player Overview . . . . 31.2 CrapsPlayer Superclass Design 31.3 CrapsPlayerPass Subclass . . . 31.4 Craps Martingale Subclass . . 31.5 Craps Player Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 169 170 170 171 171 172 173 173 175 176 177 178 179 180 180 182 183 183 184 185 186 188 190 190 191 191 192 193 193 194 195 195 199 200 201 203 204 204 205 207 207 208 208 209 210 211 v

32 Design Cleanup and Refactoring 32.1 Design Review . . . . . . . . . . 32.2 RandomEventFactory Design . . 32.3 Wheel Class Design . . . . . . . 32.4 Table Class . . . . . . . . . . . . 32.5 Game Class . . . . . . . . . . . . 32.6 RouletteGame Class . . . . . . . 32.7 CrapsGame Class . . . . . . . . 32.8 Refactoring Deliverables . . . . . 33 Simple Craps Players 33.1 Simple Craps Players Overview . 33.2 CrapsPlayer Design . . . . . . . 33.3 CrapsSimplePlayer superclass . . 33.4 Craps Martingale Player . . . . 33.5 Player1326 State . . . . . . . . . 33.6 Craps1326 Player . . . . . . . .

33.7 CrapsCancellation Player . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 33.8 Simple Craps Players Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 34 Roll-Counting Player Class 34.1 Roll-Counting Overview . . . 34.2 BettingStrategy Design . . . 34.3 NoChangeBetting Class . . . 34.4 MartingaleBetting Class . . . 34.5 Bet1326Betting Class . . . . 34.6 CrapsOneBetPlayer class . . 34.7 CrapsTwoBetPlayer class . . 34.8 CrapsSevenCountPlayer class 34.9 Roll-Counting Deliverables . 35 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 215 217 217 218 219 220 221 222 223 225

Blackjack

227

36 Blackjack Details 231 36.1 Blackjack Game . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 36.2 Available Bets and Choices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 36.3 Betting Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 37 Blackjack Solution Overview 37.1 Preliminary Survey of Classes . . . . . . . 37.2 Preliminary Class Structure . . . . . . . . . 37.3 A Walkthrough . . . . . . . . . . . . . . . . 37.4 Blackjack Solution Questions and Answers 38 Card, Deck and Shoe Classes 38.1 Card, Deck and Shoe Overview . . . . . 38.2 Card-Deck-Shoe Questions and Answers 38.3 Card Superclass . . . . . . . . . . . . . 38.4 FaceCard Class . . . . . . . . . . . . . . 38.5 AceCard Class . . . . . . . . . . . . . . 38.6 Deck class . . . . . . . . . . . . . . . . 38.7 Shoe class . . . . . . . . . . . . . . . . . 38.8 Card-Deck-Shoe Deliverables . . . . . . 39 Hand and Outcome Classes 39.1 Hand Overview . . . . . . . . . 39.2 Hand Total Class Design . . . 39.3 Hand Hard Total Class Design 39.4 Hand Soft Total Class Design . 39.5 Card Class Updates . . . . . . 39.6 Hand Class Design . . . . . . . 39.7 Hand Deliverables . . . . . . . 40 Blackjack Table Class 40.1 Blackjack Table Overview . . 40.2 BlackjackTable Class . . . . 40.3 Hand Rework . . . . . . . . . 40.4 Blackjack Table Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 235 236 236 238 239 239 240 241 243 243 243 244 245 247 247 250 250 250 251 251 253 255 255 255 256 256

41 Blackjack Game Class 41.1 Blackjack Game Overview . 41.2 Blackjack Collaboration . . . 41.3 Dealer Rules . . . . . . . . . 41.4 BlackjackPlayer Class . . . . 41.5 Card Rework . . . . . . . . . 41.6 Hand Rework . . . . . . . . . 41.7 BlackjackGame Class . . . . 41.8 Blackjack Game Deliverables

. . . . . . . .

259 259 260 262 262 264 264 264 266

42 Simple Blackjack Player Class 269 42.1 Blackjack Player Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 42.2 SimpleBlackjackPlayer Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 42.3 Blackjack Player Deliverables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 43 Variant Game Rules 43.1 Variant Game Overview . . 43.2 BlackjackGame Rework . . 43.3 OneDeckGame Class . . . . 43.4 Variant Game Deliverables 44 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 273 274 274 274 275

Fit and Finish

277

45 Python unittest Testing 281 45.1 Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 45.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 46 Python doctest Testing 285 46.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286 46.2 Add the Test Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 46.3 Mixed unittest and doctest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 47 Python Documentation 291 47.1 Basic RST Markup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292 47.2 RST Field Markup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294 47.3 Class Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296

Back Matter
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

299
301 301 301 301 301 301 301 301 303 305

48 Bibliography 48.1 Use Cases . . . . . 48.2 Computer Science 48.3 Design Patterns . 48.4 Statistics . . . . . 48.5 Python . . . . . . 48.6 Java . . . . . . . . 48.7 Casino Games . . 49 Toolset 50 Indices and tables

vii

Bibliography

307

viii

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Step-by-Step Construction of A Complete Application

Legal Notice This work is licensed under a Creative Commons License. You are free to copy, distribute, display, and perform the work under the following conditions: Attribution. You must give the original author, Steven F. Lott, credit. Noncommercial. You may not use this work for commercial purposes. No Derivative Works. You may not alter, transform, or build upon this work. For any reuse or distribution, you must make clear to others the license terms of this work.

CONTENTS

Building Skills in Object-Oriented Design, Release 2.1.1-Python

CONTENTS

Part I

Front Matter

CHAPTER

ONE

PREFACE
Penss, The Provincial Letters: Provincial letter 16, p. 571. The present letter is a very long one, simply because I had no leisure to make it shorter. BLAISE PASCAL

1.1 Why Read This Book?

The coee-shop reason for reading this book is to provide the beginning designer with a sequence of interesting and moderately complex exercises in OO design. If thats all you needed to know, skip to the next chapter. Its okay. We dont mind. The Problem. Some software developers nd themselves stalled when trying to do object-oriented (OO) design. As programmers, theyve understood the syntax of a programming language, and pieced together small examples. However, it is often diicult to take the next step to becoming a designer. The transition from guided learning of language features to self-directed design work is often ignored. Programmers are left to struggle through their rst design projects without appropriate skills or support. This may be you. Youve learned the language, but you cant take the next step. While it is critically important to read examples of good design, a nished product doesnt reveal the authors decision-making process that created the design. Theres little support that helps a programmer come to understand the design process that leads to a nal product. The most notable consequence of this skills gap is some n00b programmers will create of software that is far more complex than necessary to eectively solve a given problem. This, in turn, leads to software with high maintenance costs stemming from the low quality. It also leads to an unfair indictment of OO technology; this is usually voiced as we tried OO programming and it failed. Unrealistic Expectations. As programming team leaders, educators and consultants, we nd that software development training is focused on the programming tools, but does not expose the process of creating a design. We all start out building software designed by someone else. Whats involved in design? In the building trades, we would neither expect nor allow apprentice plumbers to design the sanitary sewage system for an urban oice building. Yet, in too many Information Technology (IT) departments, software developers are expected to leap from basic training in their tools to application design. To continue this rant, we also nd that some managers are entrusted with signicant projects, but are uncomfortable with OO design on modern high-performance hardware. They tend to focus their design energies on the kinds of software architectures that were appropriate when the enterprise owned a single computer, when 64 megabytes of memory was all the enterprise would ever need, and centralized disk storage was charged back to end user departments at a rate of pennies per track per month. In some organizations,

Building Skills in Object-Oriented Design, Release 2.1.1-Python

there are a few enduring symptoms of this mind set in some of the ways that end-user computing is separated from enterprise computing; we relegate everything non-mainframe to second class status. Management discomfort with OO technology surfaces in many ways. One shocking comment was that no application needs more than six classes. A consequence of this management attitude is an unrealistic expectation for schedule and the evolution of the deliverables. Closing the Skills Gap. The deepeer answer on the intent of this book is to help you, the beginning designer, by giving you a sequence of interesting and moderately complex exercises in OO design. The exercises are not focused on a language, but on a design process. The exercises are not hypothetical, but must lead directly to working programs. The long answer is that this book will make you work. This book can also help managers develop a level of comfort with the process of OO software development. The applications we will build are a step above trivial, and will require some careful thought and design. Further, because the applications are largely recreational in nature, they are interesting and engaging. This book allows the reader to explore the processes and artifacts of OO design before project deadlines make good design seem impossible. We hope to prevent managers from saying the following: We had a good design, but were forced to compromise it to meet our schedule. As consultants, we nd this to be a sad statement of managements emphasis of one near-term goal over long-term value. In one case, this was the result of a series of poorly-informed management decisions compounded on weak design skills. One of the root causes was the inability of the designers and managers to agree to a suitable course of action when a new kind of requirement made devastating changes to an existing design. We believe that more informed managers would have made a decision that created better long-term value.

1.2 Audience
Our primary audience you includes programmers who are new to OO programming. You need to have an exposure to the language, but you need more time to understand objects and objectorientation. We will provide exercises that have four key features: just complex enough to require careful design work, just fun enough to be engaging, easy enough that results are available immediately, and can be built in simple stages. In our eort to support you, well provide a few additional details on language features. Well mark these as Tips. For more advanced students, these tips will be review material. We will not provide a thorough background in any programming language. The student is expected to know the basics of the language and tools. Helpful additional skills include using one of the various unit test and documentation frameworks available. Weve included information in the appendices. OO Misdirection. Have you found your rst exposure to objects to be more distasteful than empowering? Why does this happen? Some instructors preface a course on a language (like Python or Java) with an extensive presentations on object orientation. Its not easy to grasp OO design concepts before getting to the language fundamentals. In some cases, this leaves students lost as to how they will accomplish the noble and lofty goals of OO. Other instructors leave OO for last, exposing the procedural side of the language rst, and treating objects as a kind of add-on. This leaves students feeling that objects are optional. Additionally, some very skilled instructors are not skilled developers, and will often show examples that dont reect currently accepted best practices.

Chapter 1. Preface

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Classroom Use. Instructors are always looking for classroom projects that are engaging, comprehensible, and focus on perfecting language skills. Many real-world applications require considerable explanation of the problem domain; the time spent reviewing background information detracts from the time available to do the relevant programming. While all application programming requires some domain knowledge, the idea behind these exercises is to pick a domain that many people know a little bit about. This allows an instructor to use some or all of these exercises without wasting precious classroom time on incidental details required to understand the problem. Skills. This book assumes an introductory level of skill in an OO programming language. We provide specic examples in Python (at least version 2.5). Student skills we expect include the following. If you cant do these things, this book is too advanced. Close it now. Create source les, compile and run application programs. While this may seem obvious, we dont discuss any integrated development environment (IDE). We have to assume these basic skills are present. Use of the core procedural programming constructs: variables, statements, exceptions, functions. We will not, for example, spend any time on design of loops that terminate properly. Some exposure to class denitions and subclasses. This includes managing the basic features of inheritance, as well as overloaded method names. We will avoid Python-unique features like multiple inheritance and callable objects, and focus on that subset of Python features that map directly to Java. For the Python equivalent of overloaded methods, we will assume that Python programmers can make use of default parameter values and named parameters. Some exposure to the various collections frameworks. For Java programmers, this means the classes in the java.util package. For Python programmers, this means the built-in sequence and mapping types. Optionally, some experience with a unit testing framework. See the appendices for supplemental exercises if you arent familiar with Pythons unittest or doctest or Javas JUnit. Optionally, some experience writing formal documentation. For Java programmers, this means javadoc comments. For Python programmers, this often means Epydoc or a similar documentation package. See the appendices for supplemental exerises if you arent familiar with formal, deliverable documentation. Language Details. There are two editions of this book, one emphasizing Python syntax and the other emphasizing Java syntax. Both books come from a common source, so that the OO Design issues are central and the language nuances are segregated.

1.3 Organization of This Book

This book presents a series of exercises to build simulations of the common, popular casino table games: Roulette, Craps and Blackjack. Each simulation can be extended to include variations on the players betting system. With a simple statistical approach, we can show the realistic expectations for any betting system. Each of these games has a separate part in this book. Each part consists of a number of individual exercises to build the entire simulation. The completed project results in an application that can provide simple tabular results that shows the average losses expected from each betting strategy. The interesting degree of freedom in each of the simulations is the players betting strategy. The design will permit easy adaptation and maintenance of the players strategies. The resulting application program can be extended by inserting additional betting systems, which allows exploration of what (if any) player actions can minimize the losses.

1.3. Organization of This Book

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Roulette. For those whove never been in a casino, or seen movies that have casinos in them, Roulette is the game with the big wheel. They spin the wheel and toss in a marble. When the wheel stops spinning, the bin in which the marble rests denes the winning outcomes. People who bet on the right things get money. People who bet on the wrong things lose money. Starting in Roulette, we proceed slowly, building up the necessary application one class at a time. Since this is the simplest game, the individual classes reect that simplicity. We focus on isolation of responsibilities, creating a considerable number of classes. The idea is to build skills in object design by applying those skills to a number of classes. The rst chapter of the part provides details on the game of Roulette and the problem that the simulation solves. The second chapter is an overview of the solution, setting out the highest-level design for the application software. This chapter includes a technique for doing a walk-through of the design to be condent that the design will actually solve the problem. Each of the remaining sixteen chapters is a design and programming exercise to be completed by the student. Plus or minus a Frequently Asked Questions (FAQ) section, each chapter has the same basic structure: an overview of the components being designed, some design details, and a summary of the deliverables to be built. The overview section presents some justication and rationale for the design. This material should help the student understand why the particular design was chosen. The design section provides a more detailed specication of the class or classes to be built. This will include some technical information on Java or Python implementation techniques. Craps. For those whove never been in a casino, or seen the play Guys and Dolls, Craps is the game with the dice. A player shoots the dice. Sometimes theres a great deal of shouting and clapping. A throw of the dice may or may not resolve bets. Additionally, a throw of the dice may also change the state of the game. A casino provides a number of visual cues as to the state of the game and the various bets. In Craps, we build on the design patterns from Roulette. Craps, however, is a stateful game, so there is a more sophisticated design to handle the interactions between dice, game state and player. We exploit the State design pattern to show how the design pattern can be applied to this simple situation. The rst chapter is background information on the game of Craps, and the problem that the simulation solves. The second chapter is an overview of the solution, setting out the highest-level design for the application software. This chapter also provides a walk-through of the design. Each of the remaining eleven chapters is an exercise to be completed by the student. Each chapter has the same basic structure: an overview of the component being designed, some design details, and a summary of the deliverables to be built. Blackjack. For those whove never been in a casino, or seen a movie with Blackjack, Blackjack is a game with cards. The dealer deals two cards to themselves and each player. One of the dealers card is up and one is down, providing a little bit of information on the dealers hand. The players may ask for additional cards, or keep the hand theyve got. The idea is to build a hand thats close to 21 points, but not more than 21. In Craps and Roulette there are a lot of bets, but few player decisions. In Blackhjack, there are few bets, but really complex player decisions. In Blackjack, the game states are more sophisticated than Craps or Roulette. In casino gift shops, you can buy small summary cards that enumerate all possible game states and responses. The more advanced student can tackle these sophisticated playing strategies. For the less advanced student we will simplify the strategies down to a few key conditions. The rst two chapters are background information on the game of Blackjack, the problem that the simulation solves, and an overview of the solution, setting out the highest-level design for the application software. Each of the remaining six chapters is an exercise to be completed by the student. Since this is more advanced material, and builds on previous work, this part has many simple deliverables compressed into the individual chapters.

Chapter 1. Preface

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Fit and Finish. We include several t-and-nish issues in Fit and Finish. This includes more information and examples on unit testing and documentation. Additionally, we cover some main program issues required to knit all of the software components together into a nished whole.

1.4 Why This Subject?

Casino table games may seem like an odd choice of subject matter for programming exercises. We nd that casino games have a number of advantages for teaching OO design and OO programming. Casino games have an almost ideal level of complexity. If they were too simple, the house edge would be too obvious and people would not play them. If they were too complex, people would not enjoy them as simple recreation. Years (centuries?) of experience in the gaming industry has ne-tuned the table games to t nicely with the limits of our human intellect. Simulation of discrete phenomena lies at the origin of OO programming. We have found it easier to motivate, explain and justify OO design when solving simulation problems. The student can then leverage this insight into other applications of OO programming for more common transactional applications. The results are sophisticated but easy to interpret. Probability theory has been applied by others to develop precise expectations for each game. These simulations should produce results consistent with the known probabilities. This book will skim over the probability theory in order to focus on the programming. For a few exercises, the theoretical results will be provided to serve as checks on the correctness of the students work. Theyre more fun than most other programming problems. This book does not endorse casino gaming. Indeed, one of the messages of this book is that all casino games are biased against the player. Even the most casual study of the results of the exercises will allow the student to see the magnitude of the house edge in each of the games presented.

1.5 Programming Style

We have to adopt a style for each of the languages were presenting. We wont present a complete set of coding standards; we will omit a number of issues that should be standardized. Some IT shops have documents they call coding standards, but are little more than descriptive style guides. What follows is not this kind of style guide; instead, it is some justication of the style we use for the examples in this book. Just to continune this rant, we nd that source code examples speak louder than any gratuitously detailed specication of the desired style. We nd that some IT organizations waste time trying to write denitions of the preferred style. A good example trumps the description of the example. In particular, as consultants, we are often asked to provide standards to an inexperienced team of programmers. While the programmers only look at the examples (often cutting and pasting them), some managers prefer to spend money on empty verbiage peripheral to the useful example. Terminology. We tend to prefer Java-centric terminology: eld and method. Occasionally, we will emphasize the dierences between Java and Python by using the Python terms attribute, instance variable or method function . Naming Style. We avoid using complex prexes for variable names. In particular, we nd prexes to be little more than visual clutter. For example, an integer parameter with the amount of a bet might be called pi_amount where the prex indicates the scope (p for a parameter) and type (i for an integer).

1.4. Why This Subject?

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This style of name is only appropriate for primitive types, and doesnt address complex data structures well at all. How does one name a parameter that is a LinkedList of Sets of Outcomes? In Java programs, the variables are formally declared, therefore, we nd that we dont need additional cues for their data type. In some cases, prexes are used to denote the scope of an instance variables. Variable names might include a cryptic one-letter prex like f to denote an instance variable; sometimes programmers will use my or the as an English-like prex. We prefer to reduce clutter. In Python, instance variables are always qualied, typically by self., making the scope very clear. Generally, method functions (and classes) should be short enough and simple enough that complex navigational aids are not necessary.

1.6 Conventions Used in This Book

Here is a Python example.

Typical Python Example

from collections import defaultdict combo = defaultdict(int) for i in range(1,7): for j in range(1,7): roll= i+j combo[roll] += 1 for n in range(2,13): print "%d %.2f%% " % ( n, combo[n]/36.0 )

1. We create a Python dictionary, a map from key to value. We use the collections.defaultdict so that missing keys are created in the dictionary with an initial value created by the int() function. 2. We iterate through all combinations of two dice, using variables i and j to represent each die. 3. We sum the dice to create a roll. We increment the value in the dictionary based on the roll. 4. Finally, we print each member of the resulting dictionary. The output from the above program will be shown as follows:
2 0.03% 3 0.06% 4 0.08% 5 0.11% 6 0.14% 7 0.17% 8 0.14% 9 0.11% 10 0.08% 11 0.06% 12 0.03%

We will use the following type styles for references to a specic Class, method(), or variable. Most of the design specications will provide Java-style method and variable descriptions. Python doesnt use type specications, and Python programmers will have to translate the Java specications into Python by removing the type names.

Chapter 1. Preface

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Sidebars When we do have a signicant digression, it will appear in a sidebar, like this. Tip: Tip There will be design tips, and warnings, in the material for each exercise. These reect considerations and lessons learned that arent typically clear to starting OO designers.

1.7 Acknowledgements
We would like to thank Chuck Pyrak for putting us up to this. His idea of a One Room Schoolhouse to teach Java to an audience at multiple skill levels was a great idea. Additionally, our colleagues who collaborated through BLOKI brought innte wisdom and insight to a complex and diicult project. Thanks to Dion Dock and Robert Lucente for comments and corrections.

1.7. Acknowledgements

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 1. Preface

CHAPTER

TWO

FOUNDATIONS
Well set our goal by presenting several elements that make up a complete problem statement : a context in which the problem arises, the problem, the forces that inuence the choice of solution, the solution that balances the forces, and some consequences of the chosen solution. Based on the problem statement, well present the high-level use case that this software implements. The use case is almost too trivial to bother dening. However, we have seen many projects run aground because they lacked even the most rudimentary description of the actor, the system and how the system helps the actor create value. We will summarize the approach to the solution, describing the overall strategy that we will follow. This is a kind of overall design pattern that well use to establish some areas of responsibility. We will also describe the technical foundations. In this case, they are not terribly complex, but this is an important part of describing any software solution, no matter how simple. We will dance around the methodology issue. Our intent is not to sell a particular methodology, but to provide some perspective on how we broke the work into manageable pieces. Finally, well present some important parts of getting started on the solution. These are more specic, technical considerations that dene common aspects of our approach.

2.1 Problem Statement

Well start with a big-picture overview of our problem. Well present the context in which the problem arises, a summary of the problem, and a business use case. This will show how our application is used. We can then dig into the details of our application. Important: Fools Rush In Its important not to rush in to programming. Heres the rule: The rst person to cut code loses. Be sure you understand the problem being solved and how software solves that problem. Context. Our context is the classic casino table games played against the house, including Roulette, Craps and Blackjack. We want to explore the consequences of various betting strategies for these casino games. Questions include How well does the Cancellation strategy work? How well does the Martingale strategy works for the Come Line odds bet in Craps? How well does this Blackjack strategy I found on the Internet compare with the strategy card I bought in the gift shop?

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A close parallel to this is exploring variations in rules and how these dierent rules have an inuence on outcomes. Questions include What should we do with the 2x and 10x odds oers in Craps? How should we modify our play for a single-deck Blackjack game with 6:5 blackjack odds? Our context does not include exploring or designing new casino games. Our context also excludes multiplayer games like poker. We would like to be able to include additional against-the-house games like Pai Gow Poker, Caribbean Stud Poker, and Baccarat. Problem. Our problem is to answer the following question: For a given game, what player strategies produce the best results? Forces. There are a number of forces that inuence our choice of solution. First, we want an application that is relatively simple to build. Instead of producing an interactive user interface, we will produce raw data and statistical summaries. If we have little interaction, a command-line interface will work perfectly. We can have the user specify a player strategy and the application respond with a presentation of the results. If the results are tab-delimited, they can be pasted into a spreadsheet for further analysis. Another force that inuences our choice of solution is the need to be platform and language agnostic. In this case, we have selected an approach that works well on POSIX-compliant operating systems (i.e., Linux, MacOS, and all of the proprietary UNIX variants), and also works on non-compliant operating systems (i.e., all of the Windows versions). We have chosen two OO languages that work identically on both platform families: Java and Python. We also need to strike a balance between interesting programming, probability theory and statistics. On one hand, the simplicity of these games means that complete analyses have been done using probability theory. However, thats not a very interesting programming exercise, so we will ignore the pure probability theory route in favor of learning OO design and programming. Another force is the desire to reect actual game play. While a long-running simulation of thousands of invidual cycles of play will approach the theoretical results, people typically dont spend more than a few hours at a table game. If, for example, a Roulette wheel is spun once each minute, a player is unlikely to see more that 480 spins in an eight-hour evening at a casino. Additionally, many players have a xed budget, and the betting is conned by table limits. Finally, we need to address the subject of money management: a player may elect to stop playing when they are ahead. This structures our statistical analysis: we must simulate sessions of play that are limited in time, the amount lost and the amount won. Use Case. The high-level use case is an overall cycle of investigation . From this overall view, the actors goal is to nd an optimal strategy for a given game. Heres the scenario were imagining.

Business Use Case

1. Actor. Researches alternative strategies. Uses IDE to build new classes for a simulator. 2. IDE. Creates new classes for the simulator. 3. Actor. Runs the simulator with selection of game and strategy. 4. Simulator. Responds with statistical results. 5. Actor. Evaluates the results. Uses a spreadsheet or other tool for analysis and visualization. Consequences. Were going build the simulator application that supports this high-level (or business) use case. Were not going to build the IDE to build the new classes. Any IDE should work. Additionally, we wont address how to analyze the results.

Chapter 2. Foundations

Building Skills in Object-Oriented Design, Release 2.1.1-Python

One of the most important consequences of our solution is that we will build an application into which new player betting strategies can be inserted. Clever gamblers invent new strategies all the time. We will not know all of the available strategies in advance, so we will not be able to fully specify all of the various design details in advance. Instead, we will nd ourselves reworking some parts of the solution, to support a new player betting strategy. This forces us to take an Agile approach to the design and implementation.

2.2 Our Simulation Application

The previous section was a uy overview of what were trying to accomplish. It sets some goals and provides a detailed context for whos using this application and why. Armed with that informatin, we can look at the simulation application were going to write. Our simulation application will allow a programmer to experiment with dierent casino game betting strategies. Well build a simple, command-line simulator that provides a reliable, accurate model of the game. We need to be able to easily pick one of a variety of player betting strategies, play a number of simulated rounds of the game, and produce a statistical summary of the results of that betting strategy. This leads us to a small essential use case. There is a single actor, the investigator. The actors goal is to see the expected results of using a particular strategy for a particular game. The typical scenario is the following.

Essential Use Case

1. Actor. Species which game and betting strategy to test. The game may require additional parameters, like betting limits. The strategy may need additional parameters, like an initial budget, or stake. 2. System. Responds with a statistical summary of the outcomes after a xed number of cycles (spins, or throws or hands). The number of cycles needs to be small (on the order of 200, to reect only a few hours of play). On Simplicity. Yes, this use case is very simple. Its a command-line application: its supposed to be simple. The point is to explore OO design, not development of a fancy GUI or web application. Simplicity is a virtue. You can add a fancy GUI or web presentation of the results later. First, create some results.

2.3 Soapbox on Use Cases

We feel that the use case technique is badly abused by some IT organizations. Quoting from [Jacobson95]. A use case is a sequence of transactions in a system whose task is to yield a result of measurable value to an individual actor of the system. A use case will clearly identify an actor, dene the value created, and dene a sequence of transactions. A use case will be a kind of system test specication. A use case will dene the systems behavior, and dene why an actor bothers to interact with it.

2.2. Our Simulation Application

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A use case is not a specication, and does not replace ordinary design. We have had experiences with customers who simply retitle their traditional procedural programming specications as use cases. We hypothesize that this comes from an unwillingness to separate problem denition from solution denition. The consequence is a conation of use case, technical background, design and programming specications into gargantuan documents that defy the ability of programmers or users to comprehend them. There are a number of common problems with use cases that will make the design job more diicult. Each of these defects should lead to review of the use case with the authors to see what, if anything, they can do to rework the use case to be more complete. No Actor. Without an actor, its impossible to tell who is getting value from the interaction. A catch-all title like the user indicates a use case written from the point of view of the database or the application software, not an actual person. An actor can be an interface with other software, in which case, the actual software needs to be named. Without knowing the actor, you will have trouble deciding which classes are clients and which classes provide the lower-level services of the application. No Value Proposition. There are two basic kinds of value: information for decisionmaking or actions taken as the result of decision-making. People interact with software because there are decisions the software cannot make or there are actions the actor cannot make. Some use cases include value-less activities like logging in, or committing a transaction, or clicking Okay to continue. These are parts of operating scenarios, not statements of value that show how the actor is happier or more successful. Without a value proposition, you will have no clue as to what problem the software solves, or what it eventually does for the actor. No Interactions. If the entire body of the use case is a series of steps the application performs, we are suspicious of the focus. We prefer a use case to emphasize interaction with the actor. Complex algorithms or interface specications should be part of an appendix or supplemental document. Without any interaction, it isnt clear how the actor uses the software. We also try to make a distinction between detailed operating scenarios and use cases. We have seen customers write documents they call detailed use cases that describe the behavior of individual graphical user interface widgets or panels. We prefer to call these scenarios, since they dont describe measuable business value, but instead describe technical interactions.

2.4 Solution Approach

From reading the problem and use case information, we can identify at least the following four general elements to our application. The game being simulated. This includes the various elements of the game: the wheel, the dice, the cards, the table and the bets. The player being simulated. This includes the various decisions the player makes based on the state of the game, and the various rules of the betting system the player is following. The statistics being collected. An overall control component which processes the game, collects the statistics and writes the details or the nal summary. When we look at common design patterns, the Model-View-Control pattern often helps to structure applications. A more sophisticated, transactional application may require a more complex structure. However, in this case, the game, the player, and the statistics are the model. The command line selection of player

Chapter 2. Foundations

Building Skills in Object-Oriented Design, Release 2.1.1-Python

and the reporting of raw data is the view. The overall control component creates the various objects to start the simulation. While interesting, we will not pursue the design of a general-purpose simulation framework. Nor will we use any of the available general frameworks. While these are handy and powerful tools, we want to focus on developing application software from scratch (or de novo) as a learning exercise. Our solution will depend heavily on desktop integration: the actor will use their IDE to create a strategy and build a new version of the application program. Once the application is built, the actor can run the application from the command line, collecting the output le. The statistical results le can be analyzed using a spreadsheet application. There are at least three separate application programs involved: the IDE (including editor and compiler), the simulator, the spreadsheet used for analysis. A typical execution of the simulator will look like the following example.

Sample Execution
python -m casino.craps --Dplayer.name="Player1326" >details.log

1. We select the main simulator control using the package casino and the module craps. 2. We dene the player to use, player.name="Player1326". The main method will use this parameter to create objects and execute the simulation. 3. We collect the raw data in a le named details.log. We are intentionally limiting our approach to a simple command-line application using the default language libraries. Avoiding additional libraries assures a lowest-common denominator multi-platform application. For Java, this standard is the J2SE set of libraries; we wont use any J2EE extensions. For Python, it is the base installation. There are a number of more technical considerations that we will expand in Deliverables. These include the use of an overall simulation framework and an approach for unit testing. Among the topics this book deals with in a casual possibly misleading manner are probability and statitics. Experts will spot a number of gaps in our exposition. For example, there isnt a compelling need for simulation of the simpler games of Craps and Roulette, since they can be completely analyzed. However, our primary objective is to study programming, not casino games, therefore we dont mind solving known problems again. We are aware that our statistical analysis has a number of deciencies. We will avoid any deeper investigation into statistics.

2.5 Methodology, Technique and Process

We want to focus on technical skills; we wont follow any particular software development methodology too closely. We hesitate to endorse a specic methodology; doing so inevitably alienates readers who embrace a dierent methodology. To continue this rant, we nd that almost everyone has an existing notion of the proper way to organize software development work. This leads to the common practice of customizing methodologies, in most cases without a deep background in the methodology or the changes being made. We prefer to lift up a few techniques which have a great deal of benet. Incremental Development. Each chapter is a sprint that produces some collection of deliverables. Each part is a complete release. Unit Testing. We dont dwell on test-driven development, but each chapter explicitly requires unit tests for the classes built.

2.5. Methodology, Technique and Process

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Embedded Documentation. We provide appendices on how to use Epydoc or javadoc to create usable API documents. The exercises are presented as if we are doing a kind of iterative design with very, very small deliverables. We present the exercises like this for a number of reasons. First, we nd that beginning designers work best with immediate feedback on their design decisions. While we present the design in considerable detail, we do not present the nal code. Programmers new to OO design will benet from repeated exposure to the transformation of problem statement through design to code. Second, for iterative or agile methodologies, this presentation parallels the way software is developed. A project manager may use larger collections of deliverables. However, the actual creation of functional source eventually decomposes into classes, elds and methods. For project managers, this exposition will help them see where and how rework can occur; giving them a chance to plan for the kind of learning that occur in most projects. Third, for project teams using a strict waterfall methodology with all design work completed before any programming work the book can be read in a slightly dierent order. From each exercise chapter, read only the overview and design sections. From that information, integrate the complete design. Then proceed through the deliverables sections of each chapter, removing duplicates and building only the nal form of the deliverables based on the complete design. This will show how design rework arises as part of a waterfall methodology. This section addresses a number of methodology or process topics: Quality, in general Rework Technical Decision-Making Reuse Design Patterns

2.5.1 On Quality
Our approach to overall quality assurance is relatively simple. We feel that a focus on unit testing and documetation covers most of the generally accepted quality factors. The Software Engineering Institute (SEI) published a quality measures taxonomy. While oicially legacy, it still provides an exhaustive list of quality attributes. These are broadly grouped into ve categories. Our approach covers most of those ve categories reasonably well. Need Satisfaction. Does the software meet the need? We start with a problem statement, dene the use case, and then write software which is narrowly focused on the actors needs. By developing our application in small increments, we can ask ourself at each step, Does this meet the actors needs? Its fairly easy to keep a software development project focused when we have use cases to describe our goals. Performance. We dont address this specically in this book. However, the presence of extensive unit tests allows us to alter the implemention of classes to change the overall performance of our application. As long as the resulting class still passes the unit tests, we can develop numerous alternative implementations to optimize speed, memory use, input/output, or any other resource. Maintenance. Software is something that is frequently changed. It changes when we uncover bugs. More commonly, it changes when our understanding of the problem, the actor or the use case changes. In many cases, our initial solution merely claries the actors thinking, and we have to alter the software to reect a deeper understanding of the problem.

Chapter 2. Foundations

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Maintenance is just another cycle of the iterative approach weve chosen in this book. We pick a feature, create or modify classes, and then create or modify the unit tests. In the case of bug xing, we often add unit tests to demonstrate the bug, and then x our classes to pass the revised unit tests. Adaptation. Adaptation refers to our need to adapt our software to changes in the environment. The environment includes interfaces, the operating system or platform, even the number of users is part of the environment. When we address issues of interoperability with other software, portability to new operating systems, scalability for more users, we are addressing adaptation issues. We chose Python and Java to avoid having interoperability and portability issues; these platforms give admirable support for many scalability issues. Generally, a well-written piece of software can be reused. While this book doesnt focus on reuse, Java and Python are biased toward writing reusable software. Organizational. There are some organizational quality factors: cost of ownership and productivity of the developers creating it. We dont address these directly. Our approach, however, of developing software incrementally often leads to good developer productivity. Our approach (Incremental, Unit Testing, Embedded Documentation) assures high quality in four of the ve quality areas. Incremental development is a way to focus on need satisfaction. Unit testing helps us optimize resource use, and do maintenance well. Our choices of tools and platforms help us address adaptation. The organizational impact of these techniques isnt so clear. It is easy to mis-manage a team and turn incremental development into a quagmire of too much planning for too little delivered software. It is all too common to declare that the eort spent writing unit test code is wasted. Ultimately, this is a book on OO design. How people organize themselves to build software is beyond our scope.

2.5.2 On Rework
In Problem Statement , we described the problem. In Solution Approach, we provided an overview of the solution. The following parts will guide you through an incremental design process; a process that involves learning and exploring. This means that we will coach you to build classes and then modify those classes based on lessons learned during later steps in the design process. See our Soapbox on Rework for an opinion on the absolute necessity for design rework. We dont simply present a completed design. We feel that it is very important follow a realistic problemsolving trajectory so that beginning designers are exposed to the decisions involved in creating a complete design. In our experience, all problems involve a considerable amount of learn as you go. We want to reect this in our series of exercises. In many respects, a successful OO design is one that respects the degrees of ignorance that people have when starting to build software. We will try to present the exercises in a way that teaches the reader how to manage ignorance and still develop valuable software. Soapbox on Rework We consider design rework to be so important, we will summarize the idea here. Important: The best way to learn is to make mistakes. Rework is a consequence of learning. All of software development can be described as various forms of knowledge capture. A project begins with many kinds of ignorance and takes steps to reduce that ignorance. Some of those steps should involve revising or consolidating previous learnings. A project without rework is suspiciously under-engineered.

2.5. Methodology, Technique and Process

Building Skills in Object-Oriented Design, Release 2.1.1-Python

For some, the word rework has a negative connotation. If you nd the word distasteful, please replace every occurance with any of the synonyms: adaptation, evolution, enhancement, mutation. We prefer the slightly negative connotation of the word rework because it helps managers realize the importance of incremental learning and how it changes the requirements, the design and the resulting software. Since learning will involve mistakes, good management plans for the costs and risks of those mistakes. Generally, our approach is to manage our ignorance; we try to create a design such that correcting a mistake only xes a few classes. We often observe denial of the amount of ignorance involved in creating IT solutions. It is sometimes very diicult to make it clear that if the problem was well-understood, or the solution was well-dened there would be immediately applicable o-the-shelf or open-source solutions. The absence of a ready-to-hand solution generally means the problem is hard. It also means that there are several degrees of ignorance: ignorance of the problem, solution and technology; not to mention ignorance of the amount of ignorance involved in each of these areas. We see a number of consequences of denying the degrees of ignorance. Programmers. For programmers, experienced in non-OO (e.g. procedural ) environments, one consequnece is that they nd learning OO design is diicult and frustrating. Our advice is that since this is new, you have to make mistakes or you wont learn eectively. Allow yourself to explore and make mistakes; feel free to rework your solutions to make them better. Above all, do not attempt to design a solution that is complete and perfect the very rst time. We cant emphasize enough the need to do design many times before understanding what is important and what is not important in coping with ignorance. Managers. For managers, experienced in non-object implementation, the design rework appears to be contrary to a fanciful expectation of reduced development eort from OO techniques. The usual form for the complaint is the following: I thought that OO design was supposed to be easier than non-OO design. Were not sure where the expectation originates, but good design takes time, and learning to do good design seems to require making mistakes. Every project needs a budget for making the necessary mistakes, reworking bad ideas to make them good and searching for simplications. Economics. Often, management attempts the false economy of attempting to minimize rework by resorting to a waterfall methodology. The idea is that having the design complete before attempting to do any development somehow magically prevents design rework. We dont see that this waterfall approach minimizes rework; rather, we see it shifting the rework forward in the process. There are two issues ignored by this approach: how we grow to understand the problem domain and providing an appropriate level of design detail. We nd that any initial high-level design can miss details of the problem domain, and this leads to rework. Forbidding rework amounts to mandating a full understanding of the problem. In most cases, our users do not fully understand their problem any more than our developers understand our users. Generally, it is very hard to understand the problem, or the solution. We nd that hands-on use of preliminary versions of software can help more than endless conversations about what could be built. In the programming arena, we nd that Java and Python (and their associated libraries) are so powerful that detailed design is done at the level of individual language statements. This leads us to write the program either in English prose or UML diagrams (sometimes both) before writing the program in the nal programming language. We often develop a strong commitment to the initial design, and subsequent revisions are merely transliterations with no time permitted for substantial revisions. While we have written the program two or three times over, the additional quality isnt worth doubling or tripling the workload. We feel that the original workload should be managed as planned cycles of work and rework.

Chapter 2. Foundations

Building Skills in Object-Oriented Design, Release 2.1.1-Python

War Story on Bad Design In one advanced programming course, we observed the following sad scenario. The instructor provided an excellent background in how to create abstract data type (ADT) denitions for the purely mathematical objects of scalar, vector and matrix. The idea was to leverage the ADTs to implement more complex operations like matrix multiplication, inversion and Gaussian elimination. The audience, primarily engineers, seemed to understand how this applied to things they did every day. The rst solution, presented after a week of work, began with the following statement: Rather than think it through from the basic denitions of matrix and vector, I looked around in my drawer and found an old FORTRAN program and rewrote that, basically transliterating the FORTRAN. We think that a lack of experience in the process of software design makes it seem that copying an inappropriate solution is more eective than designing a good solution from scratch.

2.5.3 On Decision-Making
Many of the chapters will include some lengthy design decisions that appear to be little more than handwringning over nuances. While this is true to an extent, we need to emphasize our technique for doing appropriate hand-wringing over OO design. We call it Looking For The Big Simple, and nd that managers dont often permit the careful enumeration of all the alternatives and the itemization of the pros and cons of each choice. We have worked with managers who capriciously play their schedule or budget trump cards, stopping useful discussion of alternatives. This may stem from a fundamental discomfort with the technology, and a consequent discomfort of appearing lost in front of team members and direct reports. Our suggestion in this book can be summarized as follows: Important: Good Design Good OO design comes from a good process for technical decision-making. First, admit what we dont know, and then take steps to reduce our degrees of ignorance. Which means not saying work smarter not harder unless we also provide the time and budget to actually get smarter. The learning process, as with all things, must be planned and managed. Our lesson learned from Blaise Pascal is that a little more time spent on design can result in considerable simplication, which will reduce development and maintenance costs. Its also important to note that no one in the real world is omniscient. Some of the exercises include intentional dead-ends. As a practical matter, we can rarely foresee all of the consequences of a design decision.

2.5.4 On Reuse
While there is a great deal of commonality among the three games, the exercises do not start with an emphasis on constructing a general framework. We nd that too much generalization and too much emphasis on reuse is not appropriate for beginning object designers. See Soapbox on Reuse for an opinion on reuse. Additionally, we nd that projects that begin with too-lofty reuse goals often fail to deliver valuable solutions in a timely fashion. We prefer not to start out with a goal that amounts to boiling the ocean to make a pot of tea.

2.5. Methodology, Technique and Process

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Soapbox on Reuse While a promise of OO design is reuse, this needs to be tempered with some pragmatic considerations. There are two important areas of reuse: reusing a class specication to create objects with common structure and behavior, and using inheritance to reuse structure and behavior among multiple classes of objects. Beyond these two areas, reuse can create more cost than value. The rst step in reuse comes from isolating responsibilities to create classes of objects. Generally, a number of objects that have common structure and behavior is a kind of reuse. When these objects cooperate to achieve the desired results, this is sometimes called emergent behavior : no single class contains the overall functionality, it grew from the interactions among the various objects. This is sometimes called inversion of control. When the application grows and evolves, we can preserve some class declarations, reusing them in the next revision of the application. This reduces the cost and risks associated with software change. Class denitions are the most fundamental and valuable kind of reuse. Another vehicle for OO resuse is inheritance. The simple subclass-superclass relationship yields a form of reuse; a class hierarchy with six subclasses will share the superclass code seven times over. This, by itself, has tremendous benets. We caution against any larger scope of reuse. Sharing classes between projects may or may not work out well. The complexity of achieving inter-project reuse can be paralyzing to rst-time designers. Often, dierent projects reect dierent points of view, and the amount of sharing is limited by these points of view. As an example, consider a product in a business context. An external customers view of the product (shaped by sales and marketing) may be very dierent from the internal views of the same product. Internal views of the product (for example, nance, legal, manufacturing, shipping, support) may be very dierent from each other. Reconciling these views may be far more challenging than a single software development project. For that reason, we dont encourage this broader view of reuse.

2.5.5 On Design Patterns

These exercises will refer to several of the Gang of Four design patterns in [Gamma95]. The Design Patterns book is not a prerequisite; we use it as reference material to provide additional insight into the design patterns used here. We feel that use of common design patterns signicantly expands the programmers repertoire of techniques. We note where they are appropriate, and provide some guidance in their implementation. In addition, we reference several other design patterns which are not as well documented. These are, in some cases, patterns of bad design more than patterns of good design.

2.6 Deliverables
Each chapter denes the classes to be built and the unit testing that is expected. A third deliverable is merely implied. The purpose of each chapter is to write the source les for one or more classes, the source les for one or more unit tests, and assure that a minimal set of API documentation is available. Source Files. The source les are the most important deliverable. In eect, this is the working application program. Generally, you will be running this application from within your Integrated Development Environment (IDE). You may want to create a stand-alone program. In the case of Java, we might also deliver the collection of class les. Additionally, we might bundle the class les into an executable JAR le. The source is the principle deliverable; anyone should be able to produce class and JAR les using readily available tools. In the case of Python, its the packages of .py les. There really isnt much more to deliver. The interested student might want to look at the Python distutils and setuptools to create a distribution 22 Chapter 2. Foundations

Building Skills in Object-Oriented Design, Release 2.1.1-Python

kit, or possibly a Python .egg le. Unit Test Files. The deliverables section of each chapter summarizes the unit testing that is expected, in addition to the classes to be built. We feel that unit testing is a critical skill, and emphasize it throughout the inividual exercises. We dont endorse a particular technology for implementing the unit tests. There are several approaches to unit testing that are in common use. For formal testing of some class, X, we create a separate class, TestX, which creates instances of X and exercises those instances to be sure they work. In Java, this is often done with JUnit. In Python, the unittest module is the mechanism for doing formal unit tests. Additionally, many Python developers also use the doctest module to assure that the sample code in the docstrings is actually correct. We cover these technologies in the appendices. Documentation. The job isnt over until the paperwork is done. In the case of Java and Python, the internal documentation is generally built from specially formatted blocks of comments within the source itself. Java programmers can use the javadoc tool to create documentation from the program source. Python programmers can use Epydoc (or sphinx) to create similar documentation.

2.6. Deliverables

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 2. Foundations

Part II

Roulette

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This part describes the game of Roulette. Roulette is the game with the big wheel. They spin the wheel, toss in a marble and wait for the wheel to stop spinning. Roulette is essentially a stateless game with numerous bets and a very simple process for game play. The chapters of this part present the details on the game, an overview of the solution, and a series of sixteen exercises to build a complete simulation of the game, plus a variety of betting strategies. Each exercise chapter builds at least one class, plus unit tests; in some cases, this includes rework of previous deliverables.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

CHAPTER

THREE

ROULETTE DETAILS
In the rst section, Roulette Game, we will present a summary of the game of Roulette as played in most American casinos. We will follow this with a review the various bets available on the Roulette table in Available Bets in Roulette. The denition of the various bets is an interesting programming exercise, and the rst four exercise chapters will focus on this. In Some Betting Strategies, we will describe some common betting strategies that we will simulate. The betting strategies are interesting and moderately complex algorithms for changing the amount that is used for each bet in an attempt to recoup losses.

3.1 Roulette Game

The game of Roulette centers around a wheel with thirty-eight numbered bins. The numbers include 0, 00 (double zero), 1 through 36. The table has a surface marked with spaces on which players can place bets. The spaces include the 38 numbers, plus a variety of additional bets, which will be detailed below. After the bets are placed by the players, the wheel is spun by the house, a small ball is dropped into the spinning wheel; when the wheel stops spinning, the ball will come to rest in one of the thirty-eight numbered bins, dening the winning number. The winning number and all of the related winning bets are paid o; the losing bets are collected. Roulette bets are all paid o using odds, which will be detailed with each of the bets, below. The numbers from 1 to 36 are colored red and black in an arbitrary pattern. They t into various ranges, as well as being even or odd, which denes many of the winning bets related to a given number. The numbers 0 and 00 are colored green, they t into none of the ranges, and are considered to be neither even nor odd. There are relatively few bets related to the zeroes. The geometry of the betting locations on the table denes the relationships between number bets. Note: American Rules There are slight variations in Roulette between American and European casinos. Well focus strictly on the American version.

3.2 Available Bets in Roulette

There are a variety of bets available on the Roulette table. Each bet has a payout, which is stated as n :1 where n is the multiplier that denes the amount won based on the amount bet.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A $5 bet at 2:1 will win $10. After you are paid, there will be $15 sitting on the table, your original $5 bet, plus your $10 additional winnings. Note: Odds Not all games state their odds using this convention. Some games state the odds as 2 for 1. This means that the total left on the table after the bets are paid will be two times the original bet. So a $5 bet will win $5, there will be $10 sitting on the table.

Figure 3.1: Roulette Table Layout The table is divided into two classes of bets. The inside bets are the 38 numbers and small groups of numbers; these bets all have relatively high odds. The outside bets are large groups of numbers, and have relatively low odds. If you are new to casino gambling, see Odds and Payouts for more information on odds and why they are oered.

Chapter 3. Roulette Details

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A straight bet is a bet on a single number. There are 38 possible bets, and they pay odds of 35 to 1. Each bin on the wheel pays one of the straight bets. A split bet is a bet on an adjacent pair of numbers. It pays 17:1. The table layout has the numbers arranged sequentially in three columns and twelve rows. Adjacent numbers are in the same row or column. The number 5 is adjacent to 4, 6, 2, 8; the number 1 is adjacent to 2 and 4. There are 114 of these split bet combinations. Each bin on the wheel pays from two to four of the available split bets. Any of two bins can make a split bet a winner. A street bet includes the three numbers in a single row, which pays 11:1. There are twelve of these bets on the table. A single bin selects one street bet; any of three bins make a street bet a winner. A square of four numbers is called a corner bet and pays 8:1. There are 72 of these bets available. At one end of the layout, it is possible to place a bet on the Five numbers 0, 00, 1, 2 and 3. This pays 6:1. It is the only combination bet that includes 0 or 00. A line bet is a six number block, which pays 5:1. It is essentially two adjacent street bets. There are 11 such combinations. The following bets are the outside bets. Each of these involves a group of twelve to eighteen related numbers. None of these outside bets includes 0 or 00. The only way to bet on 0 or 00 is to place a straight bet on the number itself, or use the ve-number combination bet. Any of the three 12-number ranges (1-12, 13-24, 25-36) pays 2:1. There are just three of these bets. The layout oers the three 12-number columns at 2:1 odds. All of the numbers in a given column have the same remainder when divided by three. Column 1 contains 1, 4, 7, etc., all of which have a remainder of 1 when divided by 3. There are two 18-number ranges: 1-18 is called low, 19-36 is called high. These are called even money bets because they pay at 1:1 odds. The individual numbers are colored red or black in an arbitrary pattern. Note that 0 and 00 are colored green. The bets on red or black are even money bets, which pay at 1:1 odds. The numbers (other than 0 and 00) are also either even or odd. These bets are also even money bets.

3.2. Available Bets in Roulette

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Odds and Payouts Not all of the Roulette outcomes are equal probability. Lets compare a split bet on 1-2 and a even money bet on red. The split bet wins if either 1 or 2 comes up on the wheel. This is 2 of the 38 outcomes, or a 1/19 probability, 5.26%. The red bet wins if any of the 18 red numbers come up on the wheel. The is 18 of the 38 outcomes, or a 9/19 probability, 47.4%. Clearly, the red bet is going to win almost ten times more often than the 1-2 bet. As an inducement to place bets on rare occurences, the house oers a higher payout on those bets. Since the 1-2 split bet wins is so rarely, they will pay you 17 times what you bet. On the other hand, since the red bet wins so frequently, they will only pay back what you bet. Youll notice that the odds of winning the 1-2 split bet is 1 chance in 19, but they pay you 17 times your bet. Since your bet is still sitting on the table, it looks like 18 times your bet. It still isnt 19 times your bet. This discrepency between the actual probability and the payout odds is sometimes called the house edge. It varies widely among the various bets in the game of Roulette. For example, the 5-way bet has 5/38 ways of winning, but pays only 6:1. There is only a 13.2% chance of winning, but they pay you as if you had a 16.7% chance, keeping the 3.5% dierence. You have a 5.26% chance to win a split bet, but the house pays as if it were a 5.88% chance, a .62% discrepency in the odds. The smallest discrepency between actual chances of winning (47.4%) and the payout odds (50%) is available on the even money bets: red, black, even, odd, high or low. All the betting systems that we will look at focus on these bets alone, since the house edge is the smallest.

3.3 Some Betting Strategies

Perhaps because Roulette is a relatively simple game, elaborate betting systems have evolved around it. Searches on the Internet turn up a many copies of the same basic descriptions for a number of betting systems. Our purpose is not to uncover the actual history of these systems, but to exploit them for simple OO design exercises. Feel free to research additional betting systems or invent your own. Martingale. The Martingale system starts with a base wagering amount, w, and a count of the number of losses, c, initially 0. Each loss doubles the bet. Any given spin will place an amount of w 2c on a 1:1 proposition (for example, red). When a bet wins, the loss count is reset to zero; resetting the bet to the base amount, w. This assures that a single win will recoup all losses. Note that the casinos eectively prevent successful use of this system by imposing a table limit. At a $10 Roulette table, the limit may be as low as $1,000. A Martingale bettor who lost six times in a row would be facing a $640 bet, and after the seventh loss, their next bet would exceed the table limit. At that point, the player is unable to recoup all of their losses. Seven losses in a row is only a 1 in 128 probability; making this a relatively likely situation. Waiting. Another system is to wait until some number of losses have elapsed. For example, wait until the wheel has spun seven reds in a row, and then bet on black. This can be combined with the Martingale system to double the bet on each loss as well as waiting for seven reds before betting on black. This wait for a favorable state strategy is based on a confusion between the outcome of each individual spin and the overall odds of given collections of spins. If the wheel has spun seven reds in a row, its due to spin black. 1-3-2-6 System. Another betting system is called the 1-3-2-6 system. The idea is to avoid the doubling of the bet at each loss and running into the table limit. Rather than attempt to recoup all losses in a single win, this system looks to recoup all losses by waiting for four wins in a row.

Chapter 3. Roulette Details

Building Skills in Object-Oriented Design, Release 2.1.1-Python

The sequence of numbers (1, 3, 2 and 6) are the multipliers to use when placing bets after winning. At each loss, the sequence resets to the multiplier of 1. At each win, the multiplier is advanced through the sequence. After one win, the bet is now 3w. After a second win, the bet is reduced to 2w, and the winnings of 4w are taken down or removed from play. In the event of a third win, the bet is advanced to 6w. Should there be a fourth win, the player has doubled their money, and the sequence resets. Cancellation. Another method for tracking the lost bets is called the Cancellation system or the Labouchere system. The player starts with a betting budget allocated as a series of numbers. The usual example is 1, 2, 3, 4, 5, 6, 7, 8, 9. Each bet is sum of the rst and last numbers in the last. In this case 1+9 is 10. At a win, cancel the two numbers used to make the bet. In the event of all the numbers being cancelled, reset the sequence of numbers and start again. For each loss, however, add the amount of the bet to the end of the sequence as a loss to be recouped. Heres an example of the cancellation system using 1, 2, 3, 4, 5, 6, 7, 8, 9. 1. Bet 1+9. A win. Cancel 1 and 9, leaving 2, 3, 4, 5, 6, 7, 8. 2. Bet 2+8. A loss. Add 10, leaving 2, 3, 4, 5, 6, 7, 8, 10. 3. Bet 2+10. A loss. Add 12, leaving 2, 3, 4, 5, 6, 7, 8, 10, 12. 4. Bet 2+12. A win. Cancel 2 and 12, leaving 3, 4, 5, 6, 7, 8, 10. 5. Next bet will be 3+10. A player could use the Fibonacci Sequence to structure a series of bets in a kind of cancellation system. The Fibonacci Sequence is 1, 1, 2, 3, 5, 8, 13, ... At each loss, the sum of the previous two bets the next numbers in the sequence becomes the new bet amount. In the event of a win, we simply revert to the base betting amount. This allows the player to easily track our accumulated losses, with bets that could recoup those losses through a series of wins.

3.3. Some Betting Strategies

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 3. Roulette Details

CHAPTER

FOUR

ROULETTE SOLUTION OVERVIEW

The rst section, Preliminary Survey of Classes, is a survey of the classes gleaned from the general problem statement. Refer to Problem Statement as well as the problem details in Roulette Details. This survey is drawn from a quick overview of the key nouns in these sections. Well amplify this survery with some details of the class denitions in Preliminary Roulette Class Structure. Given this preliminary of the candidate classes, A Walkthrough of Roulette is a walkthrough of the possible design that will rene the denitions, and give us some assurance that we have a reasonable architecture. We will make some changes to the preliminary class list, revising and expanding on our survey. We will also include a number of questions and answers about this preliminary design information. This should help clarify the design presentation and set the stage for the various development exercises in the chapters that follow.

4.1 Preliminary Survey of Classes

To provide a starting point for the development eort, we have to identify the objects and dene their responsibilities. The central principle behind the allocation of responsibility is encapsulation; we do this by attempting to isolate the information or isolate the processing that must be done. Encapsulation assures that the methods of a class are the exclusive users of the elds of that class. It also makes each class very loosely coupled with other classes; this permits change without a ripple through the application. For example, each Outcome contains both the name and the payout odds. That way each Outcome can be used to compute a winning amount, and no other element of the simulation needs to share the odds information or the payout calculation. In a few cases, we have looked forward to anticipate some future considerations. One such consideration is the house rake, also known as the vigorish, vig, or commission. In some games, the house makes a 5% deduction from some payouts. This complexity is best isolated in the Outcome class. Roulette doesnt have any need for a rake, since the presence of the 0 and 00 on the wheel gives the house a little over 5% edge on each bet. Well design our class so that this can be added later when we implement Craps. In reading the background information and the problem statement, we noticed a number of nouns that seemed to be important parts of the game we are simulating. Wheel Bet Bin Table Red

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Black Green Number Odds Player House One common development milestone is to be able to develop a class model in the Unied Modeling Language (UML) to describe the relationships among the various nouns in the problem statement. Building (and interpreting) this model takes some experience with OO programming. In this rst part, well avoid doing extensive modeling. Instead well simply identify some basic design principles. Well focus in on the most important of these nouns and describe the kinds of classes that you will build.

4.2 Preliminary Roulette Class Structure

Well summarize some of the classes and responsibilities that we can identify from the problem statement. This is not the complete list of classes we need to build. As we work through the exercises, well discover additional classes and rework some of these preliminary classes more than once. Well describe each class with respect to the responsibility allocated to the class and the collaborators. Some collabotors are used by an object to get work done. We have a number of uses-used by collaborative relationships among our various classes. Outcome Responsibilities. A name for the bet and the payout odds. This isolates the calculation of the payout amount. Example: Red, 1:1. Collaborators. Collected by Wheel into the bins that reect the bets that win; collected by Table into the available bets for the Player; used by Game to compute the amount won from the amount that was bet. Wheel Responsibilities. Selects the Outcome s that win. This isolates the use of a random number generator to select Outcomes; and it encapsulates the set of winning Outcomes that are associated with each individual number on the wheel. Example: the 1 bin has the following winning Outcomes: 1, Red, Odd, Low, Column 1, Dozen 1-12, Split 1-2, Split 1-4, Street 1-2-3, Corner 1-2-4-5, Five Bet, Line 1-2-3-4-5-6, 00-0-1-2-3, Dozen 1, Low and Column 1. Collaborators. Collects the Outcomes into bins; used by the overall Game to get a next set of winning Outcomes. Table Responsibilities. A collection of bets placed on Outcomes by a Player. This isolates the set of possible bets and the management of the amounts currently at risk on each bet. This also serves as the interface between the Player and the other elements of the game. Collaborators.

Chapter 4. Roulette Solution Overview

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Collects the Outcomes; used by Player to place a bet amount on a specic Outcome; used by Game to compute the amount won from the amount that was bet. Player Responsibilities. Places bets on Outcomes, updates the stake with amounts won and lost. Collaborators. Uses Table to place bets on Outcomes; used by Game to record wins and losses. Game Responsibilities. Runs the game: gets bets from Player, spins Wheel, collects losing bets, pays winning bets. This encapsulates the basic sequence of play into a single class. Collaborators. Uses Wheel, Table, Outcome, Player. The overall statistical analysis will play a nite number of games and collect the nal value of the Players stake. The class Player has the most important responsibility in the application, since we expect to update the algorithms this class uses to place dierent kinds of bets. Clearly, we need to cleanly encapsulate the Player, so that changes to this class have no ripple eect in other classes of the application.

4.3 A Walkthrough of Roulette

A good preliminary task is to review these responsibilities to conrm that a complete cycle of play is possible. This will help provide some design details for each class. It will also provide some insight into classes that may be missing from this overview. A good way to structure this task is to do a Class-Reponsibility-Collaborators (CRC) walkthrough. As preparation, get some 5 x 8 notecards. On each card, write down the name of a class, the responsibilities and the collaborators. Leave plenty of room around the responsibilities and collaborators to write notes. Weve only identied ve classes, so far, but others always show up during the walkthrough. During the walkthrough, we identify areas of responsibility, allocate them to classes of objects and dene any collaborating objects. An area of responsibility is a thing to do, a piece of information, a result. Sometimes a big piece of responsibility can be broken down into smaller pieces, and those smaller pieces assigned to other classes. There are a lot of reasons for decomposing, the purpose of this book is to explore many of them in depth. Therefore, we wont justify any of our suggestions until later in the book. For now, follow along closely to get a sense of where the exercises will be leading. The basic processing outline is the responsibility of the Game class. To start, locate the Game card. 1. Our preliminary note was that this class Runs the game. The responsibilities section has a summary of four steps involved in running the game. 2. The rst step is gets bets from :class:Player. Find the Player card. 3. Does a Player collaborate with a Game to place bets? If not, update the cards as necessary to include this. 4. One of the responsibilities of a Player is to place bets. The step in the responsibility statement is merely Places bets on :class:Outcomes. Looking at the classes, we note that the Table contains the amounts placed on the Bets. Fix the collaboration information on the Player to name the Table class. Find the Table card. 5. Does a Table collaborate with a Player to accept the bets? If not, update the cards as necessary to include this.

4.3. A Walkthrough of Roulette

Building Skills in Object-Oriented Design, Release 2.1.1-Python

6. What card has responsibility for the amount of the bet? It looks like Table. We note one small problem: the Table contains the collection of amounts bet on Outcomes. What class contains the individual amount bet on an Outcome? This class appears to be missing. Well call this new class Bet and start a new card. We know one responsibility is to hold the amount bet on a particular Outcome. We know three collaborators: the amount is paired with an Outcome, all of the Bet s are collected by a Table, and the Bet s are created by a Player. Well update all of the existing cards to name their collaboration with Bet. 7. What card has responsibility for keeping all of the Bets? Does Table list that as a responsibility? We should update these cards to clarify this collaboration. You should continue this tour, working your way through spinning the Wheel to get a list of winning Outcomes. From there, the Game can get all of the Bets from the Table and see which are based on winning Outcomes and which are based on losing Outcomes. The Game can notify the Player of each losing Bet, and notify the Player of each winning Bet, using the Outcome to compute the winning amount. This walkthrough will give you an overview of some of the interactions among the objects in the working application. You may uncover additional design ideas from this walkthrough. The most important outcome of the walkthrough is a clear sense of the responsibilities and the collaborations required to create the necessary application behavior.

4.4 Roulette Solution Questions and Answers

Why does the Game class run the sequence of steps? Isnt that the responsibility of some main program? Coee Shop Answer. We havent nished designing the entire application, so we need to reect our own ignorance of how the nal application will be assembled from the various parts. Rather than allocate too many responsibilities to Game, and possibly nding conicts or complication, wed rather allocate too few responsibilities until we know more. From another point of view, designing the main program is premature because we havent nished designing the entire application. We anticipate a Game object being invoked from some statistical data gathering object to run one game. The data gathering object will then get the nal stake from the player and record this. Game s responsibilities are focused on playing the game itself. Well need to add a responsibility to Game to collaborate with the data gathering class to run a number of games as a session. Deeper Answer. In procedural programming (especially in languages like COBOL), the main program is allocated almost all of the responsibilities. These procedural main programs usually contain a number of elements, all of which are very tightly coupled. We have seen highly skilled programmers who are able to limit the amount of work done in the main program, even in procedural languages. In OO languages, it becomes possible for even moderately skilled programmers to reduce the main program to a short list of object constructors, with the real work delegated to the objects. We nd that main program classes are relatively hard to reuse, and prefer to keep them as short as possible. Why is Outcome a separate class? Each object that is an instance of Outcome only has two attributes; why not use an array of Strings for the names, and a parallel array of integers for the odds? Representation. We prefer not to decompose an object into separate data elements. If we do decompose this object, we will have to ask which class would own these two arrays? If Wheel keeps these, then Table becomes very tightly coupled to these two arrays that should be Wheel s responsibility. If Table keeps these, then Wheel is priviledged to know details of how Table is

Chapter 4. Roulette Solution Overview

Building Skills in Object-Oriented Design, Release 2.1.1-Python

implemented. If we need to change these arrays to another storage structure, two classes would change instead of one. Having the name and odds in a single Outcome object allows us to change the representation of an Outcome. For example, we might replace the String as the identication of the outcome, with a collection of the individual numbers that comprise this outcome. This would identify a straight bet by the single winning number; an even money bet would be identied by an array of the 18 winning numbers. Responsibility. he principle of isolating responsibility would be broken by this two parallel arrays design because now the Game class would need to know how to compute odds. In more complex games, there would be the added complication of guring the rake. Consider a game where the Players strategy depends on the potential payout. Now the Game and the Player both have copies of the algorithm for computing the payout. A change to one must be paired with a change to the other. The alternative we have chosen is to encapsulate the payout algorithm along with the relevant data items in a single bundle. If Outcome encapsulates the function to compute the amount won, isnt it just a gloried subroutine? In a limited way, yes. A class can be thought of as a gloried subroutine library that captures and isolates data elements along with their associated functions. For some new designers, this is a helpful summary of the basic principle of encapsulation. Inheritance and subclasses, however, make a class more powerful than a simple subroutine library with private data. Inheritance is a way to create a family of closely-related subroutine libraries in a simple way that is validated by the compiler. What is the distinction between an Outcome and a Bet? We need to describe the propositions on the table on which you can place bets. The propositions are distinct from an actual amount of money wagered on a proposition. There are a lot of terms to choose from, including bet, wager, proposition, place, location, or outcome. We opted for using Outcome because it seemed to express the open-ended nature of a potential outcome, dierent from an amount bet on a potential outcome. In a way, were considering the Outcome as an abstract possibility, and the Bet as a concrete action taken by a player. Also, as we expand this simulation to cover other games, we will nd that the randomized outcome is not something we can directly bet on. In Roulette, however, all outcomes are something we can be bet on, as well as a great many combinations of outcomes. We will revisit this design decision as we move on to other games. Why are the classes so small? First-time designers of OO applications are sometimes uncomfortable with the notion of emergent behavior. In procedural programming languages, the applications features are always embodied in a few key procedures. Sometimes a single procedure, named main. A good OO design partitions responsibility. In many cases, this subdivision of the applications features means that the overall behavior is not captured in one central place. Rather, it emerges from the interactions of a number of objects. We have found that smaller elements, with very nely divided responsibilities, are more exible and permit change. If a change will only alter a portion of a large class, it can make that portion incompatible with other portions of the same class. A symptom of this is a bewildering nest of if-statements to sort out the various alternatives. When the design is decomposed down more nely, a change can be more easily isolated to a single class. A much simpler sequence of ifstatements can be focused on selecting the proper class, which can then simply carry out the desired functions.

4.4. Roulette Solution Questions and Answers

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 4. Roulette Solution Overview

CHAPTER

FIVE

OUTCOME CLASS
In addition to dening the fundamental Outcome on which all gambling is based, this chapter provides a sidebar discussion on the notion of object identity and object equality. This is important because we will be dealing with a number of individual Outcome objects, and we need to be sure we can test for equality of two dierent objects. This is dierent from the test for identity.

5.1 Outcome Overview

The rst class we will tackle is a small class that encapsulates each outcome. This class will contain the name of the outcome as a String, and the odds that are paid as an integer. We will use these objects when placing a bet and also when dening the Roulette wheel. There will be several hundred instances of this class on a given Roulette table. The bins on the wheel, similarly, collect various Outcomes together. The minimum set of Outcome instances we will need are the 38 numbers, Red and Black. The other instances add details to our simulation. In Roulette, the amount won is a simple multiplication of the amount bet and the odds. In other games, however, there may be a more complex calculation because the house keeps 5% of the winnings, called the rake. While it is not part of Roulette, it is good to have our Outcome class designed to cope with these more complex payout rules. Also, we know that other casino games, like Craps, are stateful. An Outcome may change the game state. We can foresee reworking this class to add in the necessary features to change the state of the game. While we are also aware that some odds are not stated as x :1, we wont include these other kinds of odds in this initial design. Since all Roulette odds are x :1, well simply assume that the denominator is always 1. We can forsee reworking this class to handle more complex odds, but we dont need to handle the other cases yet.

5.2 Design Decision Object Identity

Our design will depend on matching Outcome objects. Well be testing objects for equality. The player will be placing bets that contain Outcomes; the table will be holding bets. The wheel will select the winning Outcomes. We need a simple test to see if two objects of the Outcome class are the same. Was the Outcome for a bet equal to the Outcome contained in a spin of the wheel? It turns out that this comparison between objects has some subtlety to it. Outcomes have Names. Since each instance of Outcome has a distinct Outcome.name attribute, it seems simple enough to compare names. This is one sense of equal that seems to be appropriate. 41

Building Skills in Object-Oriented Design, Release 2.1.1-Python

To do object comparison correclty in Python, we must provide the __eq__() and __ne__() method which compares the Outcome.name attribute. However, simply comparing outcome names isnt in the long run going to work out well. More than equal. Well be creating collections of Outcome objects, and we may need to create sets or maps where hash codes are used in addition to the simple equality tests. As we look forward, the Python set and dict depend on a __hash__() method and an __eq__() method of each object in the collection. The Hash Code Problem. Every object has a hash code. The hash code is simply a unique integer, perhaps the address of the object, perhaps some summary of the bits that make up the object. The default rule is this: Distinct objects will have distinct hash codes The default rule makes it possible for us to accidentally create objects that appear to be the same thing, but internally have dierent hash codes. In Python, if we do nothing special, the __eq__() test will simply compare the object id values. These object id values are unique to each distinct object, irrespective of the attribute values. This default behavior of objects is shown by the following example:

Object Identity
>>> oc1= Outcome( "Any Craps", 8 ) >>> oc2= Outcome( "Any Craps", 8 ) >>> oc1 == oc2 False >>> id(oc1) 6291728 >>> id(oc2) 6291760

Each individual Outcome object has a distinct distinct hashcode. This makes them not equal according to the default methods inherited from object. However, we would like to have two of these objects test as equal. This example shows that we can have two objects that appear equal, but dont compare as equal. Layers of Meaning. The real issue is that we have three distinct layers of meaning for comparing objects to see if they are equal. Compare as Equal. We can call this attribute equality. The __eq__() method returns True. When we use the == operator, this is evaluated by using the __eq__() method. This must be overridden by a class to implement attribute equality. Have the same hash code. We can call this hash equality. This has to include attribute equality. The __hash__(self)() method for several objects that represent the same Outcome have the same hash code. Are references to the same object. We can call this identity. We can test that two objects are the same by using the is comparison between two objects. When we use the is comparison, were asserting that the two variables are references to the same object. This is the identity comparison.

Chapter 5. Outcome Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

If we pursue this as an implementation, we have to be sure we dont casually create instances of Outcome. Rather than create them all over our application, we need to have a single source for the oicial instances of Outcome. Problem. Which should we implement, hash equality or object identity? Forces. There are two broad implementation choices for making sure that our objects both provide appropriate hash values and pass equality tests. 1. Object Identity. We can make sure that all of the parts of our application share the same Outcome objects. Two references to the same object will pass any identity test and will also have the same hash code. This requires that we have a pool of outcomes and each Outcome is eectively a Singleton instance. We have to be sure that our application has only one instance of Outcome( "1", 35 ), and that object is shared by the Wheel, the Player and any Bets that are made. We have to be sure to never create additional Outcome objects; there would be confusion between two discrete instances that happened to have the same name and odds. We have to avoid casually creating new instances of the Outcome class. Assuring that we work with single instances of each Outcome ts with the Dont Repeat Yourself (DRY) principle: we only create each Outcome once. 2. Hash Equality. We can make sure that all instances of an Outcome act like they are the same object. That is, they have the same hash value and they test true for equality. They may not be two references to the same underlying object, but they pass the hash and equality tests, making them equal for all practical purposes. This allows us to create additional Outcome objects, condent they will behave appropriately. This, however, also requires that we include the odds when creating an outcome. This gets wordy and unpleasant because we have the various outcomes and odds stated in more than one place in our application. This would break the Dont Repeat Yourself (DRY) principle. The DRY principle is so important that this alternative is really unacceptable. Solution. We have to assure that were sharing single instances of Outcome objects. We have several choices for assuring that we have exactly one instance of each Outcome. Global Outcome Objects. We can declare global variables for the various outcomes and use those global objects as needed. Generally, globals variables are often undesirable because changes to those variables can have unexpected consequences in a large application. Global constants are no problem at all. The pool of Outcome instances are proper constant values used to create bins and bets. Outcome Factory Object. We can create a class which is a Factory for individual Outcome objects. When some part of the application asked for an Outcome which didnt yet exist, the Factory would create it, save it, and return a reference to it. When some part of the application asked for an Outcome which already exists, the Factory would return a reference to the existing object. This centralizes the pool of global objects into a single object, the Factory. Singleton Outcome Class. A Singleton class creates and maintains a single instance of itself. This requires that the class have a static getInstance() method that is a reference to the one-and-only instance of the class. This saves us from creating global variables. Instead, each class denition contains its own private reference to the one-and-only object of that class.

5.2. Design Decision Object Identity

Building Skills in Object-Oriented Design, Release 2.1.1-Python

However, this has the profound disadvantage that each distinct outcome would need to be a distinct subclass of Outcome. This is an unappealing level of complexity. All of these are acceptable techniques for implementing object identity. We dont have to choose one at the current time. Consequences. The most important consequence of this design decision is that were going to evenually create some global denitions of Outcomes, or a Factory object that centralizes Outcome creation. We may also need to add the necessary static variable and static getInstance() method.

5.3 Outcome Design

class Outcome() Outcome contains a single outcome on which a bet can be placed. In Roulette, each spin of the wheel has a number of Outcomes with bets that will be paid o. For example, the 1 bin has the following winning Outcomes: 1, Red, Odd, Low, Column 1, Dozen 1-12, Split 1-2, Split 1-4, Street 1-2-3, Corner 1-2-4-5, Five Bet, Line 1-2-3-4-5-6, 00-0-1-2-3, Dozen 1, Low and Column 1. All of these bets will payo if the wheel spins a 1.

5.3.1 Fields
name Holds the name of the Outcome. Examples include "1", "Red". odds Holds the payout odds for this Outcome. Most odds are stated as 1:1 or 17:1, we only keep the numerator (17) and assume the denominator is 1. We can use name to provide hash codes and do equality tests.

5.3.2 Constructors
__init__(self, name, odds) Parameters name (str) The name of this outcome odds (int) The payout odds of this outcome. Sets the instance name and odds from the parameter name and odds.

5.3.3 Methods
For now, well assume that were going to have global instances of each Outcome. winAmount(self, amount ) Multiply this Outcomes odds by the given amount. The product is returned. Parameter amount (number) amount being bet __eq__(self, other ) Compare the name attributes of self and other.

Chapter 5. Outcome Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Parameter other (Outcome) Another Outcome to compare against. Returns True if this name matches the other name. Return type bool __ne__(self, other ) Compare the name attributes of self and other. Parameter other (Outcome) Another Outcome to compare against. Returns True if this name does not match the other name. Return type bool __str__(self ) Easy-to-read representation of this outcome. Python formatting is most easily done with the % operator. For some tips on how to do this, see Message Formatting. Returns String of the form name (odds:1). Return type str This easy-to-read String output method is essential. This should return a String representation of the name and the odds. A form that looks like 1-2 Split (17:1) works nicely.

5.4 Outcome Deliverables

There are two deliverables for this exercise. Both will have Javadoc comments or Python docstrings. The Outcome class. A class which performs a unit test of the Outcome class. The unit test should create a three instances of Outcome, two of which have the same name. It should use a number of individual tests to establish that two Outcome with the same name will test true for equality, have the same hash code, and establish that the winAmount() method works correctly.

5.5 Message Formatting

For the very-new-to-Python, there are few variations on creating a formatted string. Generally, we simply use something like this.
def __str__( self ): return "%s (%d:1)" % ( self.name, self.odds )

5.4. Outcome Deliverables

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 5. Outcome Class

CHAPTER

SIX

BIN CLASS
This chapter will present the design for the Bin class. In addition to that, well examine the collection classes available in Java and Python. Well also present present some questions and answers on this particular class and the process of creating a design.

6.1 Bin Overview

The Roulette wheel has 38 bins, identied with a number and a color. Each of these bins denes a number of closely related winning Outcome s. At this time, we wont enumerate each of the 38 Bins of the wheel and each of the winning Outcomes (from two to fourteen in each Bin); well save that for a later exercise. At this time, well dene the Bin class, and use this class to contain a number of Outcome objects. Two of the Bins have relatively few Outcomes. Specically, the 0 Bin only contains the basic 0 Outcome and the 00-0-1-2-3 Outcome . The 00 Bin , similarly, only contains the basic 00 Outcome and the 00-0-1-2-3 Outcome . The other 36 Bins contain the straight bet, split bets, street bet, corner bets, line bets and various outside bets (column, dozen, even or odd, red or black, high or low) that will win if this Bin is selected. Each number bin has from 12 to 14 individual winning Outcomes. Some Outcomes, like red or black, occur in as many as 18 individual Bins. Other Outcomes, like the straight bet numbers, each occur in only a single Bin. We will have to be sure that our Outcome objects are shared appropriately by the Bins. Since a Bin is just a collection of individual Outcome objects, we have to select a collection class to contain the objects. In Java we have ve basic choices, with some variations based on performance needs. In Python, we have two basic choices.

6.2 Design Decision Choosing A Collection

There are ve basic Python types that are a containers for other objects. Immutable Sequence tuple. A tuple is immutable sequence of objects. This is a good candidate for the kind of collection we need, since the elements of a Bin dont change. Howver, a tuple allows duplicates and retains things in a specic order; we cant tolerate duplicates, and order doesnt matter. Mutable Sequence list. A list is a sequence of objects that can be changed. While handy for the initial construction of the bin, this isnt really useful because the contents of a bin dont change once they have been enumerated.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Mapping dict. A dict associates a key with a value. This key-value pair is called an item in the map. We dont need this feature at all. A map does more than we need for representing a Bin. Mutable Set set. A set is a mutable collection of objects. Duplicates arent allowed, and theres no inherent ordering. This looks close to what we need. Immutable Set frozenset. A frozenset is an immutable collection of objects. Duplicates arent allowed, and theres no inherent ordering. This looks close to what we need. Having looked at the fundamental collection varieties, we will elect to use a frozenset.

6.3 Bin Questions and Answers

Why wasnt Bin in the design overview? The denition of the Roulette game did mention the 38 bins of the wheel. However, when identifying the nouns, it didnt seem important. Then, as we started designing the Wheel class, the description of the wheel as 38 bins came more fully into focus. Rework of the preliminary design is part of detailed design. This is the rst of several instances of rework. Why introduce an entire class for the bins of the wheel? Why cant the wheel be an array of 38 individual arrays? There are two reasons for introducing Bin as a separate class: to improve the delity of our object model of the problem, and to reduce the complexity of the Wheel class. The denition of the game describes the wheel as having 38 bins, each bin causes a number of individual Outcomes to win. Without thinking too deeply, we opted to dene the Bin class to hold a collection of Outcomes. At the present time, we cant foresee a lot of processing that is the responsibility of a Bin. But allocating a class permits us some exibility in assigning responsibilities there in the future. Additionally, looking forward, it is clear that the Wheel class will use a random number generator and will pick a winning Bin. In order to keep this crisp denition of responsibilities for the Wheel class, it makes sense to delegate all of the remaining details to another class. Isnt an entire class for bins a lot of overhead? The short answer is no, class denitions are almost no overhead at all. Class denitions are part of the compilers world; at run-time they amount to a few simple persistent objects that dene the class. Its the class instances that cause run-time overhead. In a system where were are counting individual instruction executions at the hardware level, this additional class may slow things down somewhat. In most cases, however, the extra few instructions required to delegate a method to an internal object is oset by the benets gained from additional exibility. How can you introduce Set, List, Vector when these dont appear in the problem? We have to make a distinction between the classes that are uncovered during analysis of the problem in general, and classes are that just part of the implementation of this particular solution. This emphasizes the distinction between the problem as described by users and a solution as designed by software developers. The collections framework are part of a solution, and hinted at by the denition of the problem. Generally, these solution-oriented classes are part of frameworks or libraries that came with our tools, or that we can license for use in our application. The problem-oriented classes, however, are usually unique to our problem.

Chapter 6. Bin Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

6.4 Bin Design

class Bin() Bin contains a collection of Outcomes which reect the winning bets that are paid for a particular bin on a Roulette wheel. In Roulette, each spin of the wheel has a number of Outcome s. Example: A spin of 1, selects the 1 bin with the following winning Outcomes: 1 , Red , Odd , Low , Column 1 , Dozen 1-12 , Split 1-2 , Split 1-4 , Street 1-2-3 , Corner 1-2-4-5 , Five Bet , Line 1-2-3-4-5-6 , 00-0-1-2-3 , Dozen 1 , Low and Column 1 . These are collected into a single Bin .

6.4.1 Fields
outcomes A frozenset that holds the collection of individual Outcomes.

6.4.2 Constructors
Python programmers should provide an initializer that uses the * modier so that all of the individual arguments appear as a single list within the initializer. __init__(self, * outcomes) Parameter outcomes any number of outcomes used to populate the Bin initially. This constructor can be used as follows.

Python Bin Construction

five= Outcome( "00-0-1-2-3", 6 ) zero= Bin( Outcome("0",35), five ) zerozero= Bin( Outcome("00",35), five )

1. zero is based on references to two objects: the 0 Outcome and the 00-0-1-2-3 Outcome. 2. zerozero is based on references to two objects: the 00 Outcome and the 00-0-1-2-3 Outcome.

6.4.3 Methods
add(self, outcome) Adds an Outcome to this Bin. This can be used by a builder to construct all of the bets in this Bin. Since this class is really just a faade over the underlying collection object, this method could simply delegate the real work to the underlying collection. Note that a frozenset is immutable; unlike a list it does not have an append() method. Instead, a new frozenset can be constructed by using the | operator which creates a new frozenset from a frozenset and a frozenset or set
>>> zero= Outcome("0",35) >>> zerozero= Outcome("00",35) >>> b = frozenset( [zero] ) >>> b frozenset([<roulette.Outcome object at 0x6002d0>])

6.4. Bin Design

Building Skills in Object-Oriented Design, Release 2.1.1-Python

>>> map(str,b) ['0 (35:1)'] >>> b |= set( [zerozero] ) >>> map(str,b) ['0 (35:1)', '00 (35:1)'] >>> b frozenset([<roulette.Outcome object at 0x6002d0>, <roulette.Outcome object at 0x600330>])

Parameter outcome (Outcome) An outcome to add to this Bin __str__(self ) An easy-to-read representation of the list of Outcomes in this Bin. A handy technique for displaying collections is the following. This maps the str() function to each element of Bin.outcomes, then joins the resulting string with ", " separators
', '.join( map(str,self.outcomes) )

Returns String of the form [outcome, outcome, ...]. Return type str

6.5 Bin Deliverables

There are two deliverables for this exercise. Both will have Javadoc comments or Python docstrings. The Bin class. A class which performs a unit test of the Bin class. The unit test should create several instances of Outcome, two instances of Bin and establish that Bin s can be constructed from the Outcomes. Programmers who are new to OO techniques are sometimes confused when reusing individual Outcome instances. This unit test is a good place to examine the ways in which object references are shared. A single Outcome object can be referenced by several Bins. We will make increasing use of this in later sections.

Chapter 6. Bin Class

CHAPTER

SEVEN

WHEEL CLASS
This chapter builds on the previous two chapters, creating a more complete composite object from the Outcome and Bin classes we have already dened. Well introduce some basic subclass constructor techniques, also.

7.1 Wheel Overview

The wheel has two responsibilities: it is a container for the Bins and it picks one Bin at random. Well also have to look at how to initialize the various Bins that comprise a standard Roulette wheel.

7.1.1 The Container Responsibility

Since the Wheel is 38 Bins, it is a collection. We can review our survey of available collections in Design Decision Choosing A Collection for some guidance here. In this case, the choice of the winning Bin will be selected by a random numeric index. We need some kind of sequential collection. This makes an immutable tuple very appealing. This is a subclass of collections.Sequence and has the features were looking for. One consequence of using a sequential collection is that we have to choose an index for the various Bins. Each Bin contains a variety of individual Outcomes. However, each single number Outcome will be unique to each Bin. The index values of 1 to 36 become the index to their respective Bins. The Bin at index 1 is the only bin which contains Outcome( "1", 35 ). We have a small problem, however, with 0 and 00: we need two separate indexes. While 0 is a valid index, what do we do with 00? Enumerate some possible solutions before reading on.

Since the index of the Bin doesnt have any signicance at all, we can assign the Bin that has the 00 Outcome to position 37 in the Vector. This gives us a unique place for all 38 Bins.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

7.1.2 The Random Bin Selection Responsibility

In order for the Wheel to select a Bin at random, well need a random number from 0 to 37 that we can use an an index. Python programmers can use the random module. This module oers a Random.choice() function which picks a random value from a sequence. This is ideal for returning a randomly selected Bin from our list of Bins. Testability. Note that testing a class using random numbers isnt really the best policy. To do testing properly, well need to create a non-random random number generator that we can use in place of the published library random number generator. To create a non-random random-number generator, we can so something like the following. 1. Design our Wheel to accept a Random Number Generator that in inserted by the main program. For testing a non-random generator is used. For real simulation, a proper random generator is used. 2. Create a subclass of the system-supplied random number generator. One that Override a few methods required for testing.

7.1.3 Constructing a Wheel

Each instance of Bin has a list of Outcomes. The 0 and 00 Bin s only have two Outcome s. The other numbers have anywhere from twelve to fourteen Outcomes. Constructing the entire collection of Bins would be a tedious undertaking. Well apply a common OO principle of deferred binding and work on that algorithm later, after we have the basic design for the Wheel nished.

7.2 Wheel Design

class Wheel() Wheel contains the 38 individual bins on a Roulette wheel, plus a random number generator. It can select a Bin at random, simulating a spin of the Roulette wheel.

7.2.1 Fields
bins Contains the individual Bin instances. This is a tuple of 38 elements. This can be built with tuple( Bin() for i in range(38) ) rng The random number generator to use to select a Bin from the bins collection. This is not always simply random.Random(). For testing, we need to inject a non-random random number generator in place of the system random number generator.

7.2.2 Constructors
__init__(self, rng)

Chapter 7. Wheel Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Parameter rng (random.Random) A random number generator. For testing, this may be a non-random number generator. Creates a new wheel with 38 empty Bins. It will also create a new random number generator instance. At the present time, this does not do the full initialization of the Bins. Well rework this in a future exercise.

7.2.3 Methods
addOutcome(number, outcome) Adds the given Outcome to the Bin with the given number. Parameters bin (int) bin number, in the range zero to 37 inclusive. outcome (Outcome) The Outcome to add to this Bin next() Generates a random number between 0 and 37, and returns the randomly selected Bin. The Random.choice() function of the random module will select one of the available Bin s from the bins list. Returns A Bin selected at random from the wheel. Return type Bin get(bin) Returns the given Bin from the internal collection. Parameter bin (int) bin number, in the range zero to 37 inclusive. Returns The requested Bin. Return type Bin

7.3 Non-Random Design

We need a controlled kind of random number generation for testing purposes. Well dene a class NonRandom that extends random.Random, but provides a more testable sequence of values. One approach is to simply return a given integer value.

7.3.1 Fields
value The non-random value to return.

7.3.2 Constructors
__init__(self ) Creates a non-random random number generator.

7.3. Non-Random Design

Building Skills in Object-Oriented Design, Release 2.1.1-Python

7.3.3 Methods
setSeed(self, value) Saves this value as the next value to return. Parameter value (long) the value to return next. choice(self, sequence) Use the given seed value as an index and return the requested item from the sequence. Parameter sequence the sequence of values from which a random value is selected.

7.4 Wheel Deliverables

There are three deliverables for this exercise. The new class and the unit test will have Javadoc comments or Python docstrings. The Wheel class. A class which performs a unit test of building the Wheel class. The unit test should create several instances of Outcome, two instances of Bin, and an instance of Wheel. The unit test should establish that Bins can be added to the Wheel. A Non-Random Random Number Generator class, to be used for testing. A class which tests the Wheel and NonRandom class by selecting values from a Wheel object.

Chapter 7. Wheel Class

CHAPTER

EIGHT

BIN BUILDER CLASS

This is a rather long chapter, which describes a number of design considerations surrounding Outcome, Bin and Wheel. The second section presents a number of algorithms to group Outcomes based on the geometry of the table. These algorithms are presented at a high level, leaving much of the detailed design to the student.

8.1 Bin Builder Overview

It is clear that enumerating each Outcome in the 38 Bins is a tedious undertaking. Most Bins contain about fourteen individual Outcomes. It is often helpful to create a class that is used to build an instance of another class. This is a design pattern sometimes called a Builder. Well design an object that builds the various Bins and assigns them to the Wheel. This will ll the need left open in the Wheel Class. Additionally, we note that the complex algorithms to construct the Bins are only tangential to the operation of the Wheel object. While not essential to the design of the Wheel class, we nd it is often helpful to segregate these rather complex builder methods into a separate class. The BinBuilder class will have a method that enumerates the contents of each of the 36 number Bins, building the individual Outcome instances. We can then assign these Outcome objects to the Bins of a Wheel instance. We will use a number of steps to create the various types of Outcomes, and depend on the Wheel to assign each Outcome object to the correct Bin. The Roulette Outcomes. Looking at the Available Bets in Roulette gives us a number of geometric rules for determining the various Outcomes that are combinations of individual numbers. These rules apply to the numbers from one to thirty-six. A dierent and much simpler set of rules applies to 0 and 00. First, well survey the table geometry, then well develop specic algorithms for each kind of bet. Split Bets. Each number is adjacent to two, three or four other numbers. The four corners (1, 3, 34, and 36) only participate in two split bets. The center column of numbers (5, 8, 11, ..., 32) each participate in four split bets. The remaining edge numbers participate in three split bets. While this is moderately complex, the bulk of the layout (from 4 to 32) can be handled with a simple rule to distinguish the center column from the edge columns. The ends (1, 2, 3, 34, 35 and 36) are a little more complex. Street Bets. Each number is a member of one of the twelve street bets. Corner Bets. Each number is a member of one, two or four corner bets. As with split bets, the bulk of the layout can be handled with a simple rule to distinguish the column, and hence the corners. A number in the center column (5, 8, 11, ..., 32) is a member of four corners. At the ends, 1, 3, 34, and 36, are members of just one corner. All of the remaining numbers are along an edge and are members of two corners.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Line Bets. Six numbers comprise a line; each number is a member of one or two lines. The ends (1, 2, 3 and 34, 35, 36) are each part of a single line. The remaining 10 rows are each part of two lines. Dozen Bets. Each number is a member of one of the three dozens. The three ranges are from 1 to 12, 13 to 24 and 25 to 36, making it very easy to associate numbers and ranges. Column Bets. Each number is a member of one of the three columns. Each of the columns has a number numeric relationship. The values are 3c + 1, 3c + 2, and 3c + 3, where 0 c < 12. The Even-Money Bets. These include Red, Black, Even, Odd, High, Low. Each number has three of the six possible even money Outcome s. An easy way to handle these is to create the 6 individual Outcome instances. For each number, n, individual if-statements can be used to determine which of the Outcome objects are associated with the given Bin. The Bins for zero and double zero can easily be enumerated. Each bin has a straight number bet Outcome, plus the Five Bet Outcome (00-0-1-2-3, which pays 6:1). One other thing well probably want are handy names for the various kinds of odds. While can dene an Outcome as Outcome( "Number 1", 35 ) , this is a little opaque. A slightly nicer form is Outcome( "Number 1", RouletteGame.StraightBet ). These are specic to a game, not general features of all Outcomes. We havent designed the game yet, but we can provide a stub class with these these outcome odds denitions.

8.2 Bin Builder Algorithms

This section provides the algorithms for nine kinds of bets.

8.2.1 Generating Straight Bets

Straight bet Outcomes are the easiest to generate. For All Numbers. For each number, n, 1 n < 37: Create Outcome. Create an Outcome from the number, n, with odds of 35:l. Assign to Bin. Assign this new Outcome to Bin n. Zero. Create an Outcome from the 0 with odds of 35:l. Assign this to Bin 0. Double Zero. Create an Outcome from the 00 with odds of 35:l. Assign this to Bin 37.

8.2.2 Generating Split Bets

Split bet Outcomes are more complex because of the various cases: corners, edges and down-the-middle. We note that there are two kinds of split bets: left-right pairs. These pairs all have the form ( n , n +1). up-down paris. These paris have the form ( n , n +3). We can look at the number 5 as being part of 4 dierent pairs: (4,4+1), (5,5+1), (2,2+3), (5,5+3). The corner number 1 is part of 2 split bets: (1,1+1), (1,1+3). We can generate the left-right split bets by iterating through the left two columns; the numbers 1, 4, 7, ..., 34 and 2, 5, 8, ..., 35. For All Rows. For each row, r, where 0 r < 12:

Chapter 8. Bin Builder Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

First Column Number. Set n 3r + 1. This will create values 1, 4, 7, ..., 34. Column 1-2 Split. Create a n, n +1 split Outcome with odds of 17:1. Assign to Bins. Associate this object with two Bins: n and n +1. Second Column Number. Set n 3r + 2. This will create values 2, 5, 8, ..., 35. Column 2-3 Split. Create a n , n +1 split Outcome. Assign to Bins. Associate this object to two Bins: n and n +1. A similar algorithm must be used for the numbers 1 through 33, to generate the up-down split bets. For each number, n, we generate a n, n +3 split bet. This Outcome belongs to two Bins: n and n +3.

8.2.3 Generating Street Bets

Street bet Outcomes are very simple. We can generate the street bets by iterating through the twelve rows of the layout. For All Rows. For each row, r, 0 r < 12: First Column Number. Set n 3r + 1. This will create values 1, 4, 7, ..., 34. Street. Create a n, n +1, n +2 street Outcome with odds of 11:1. Assign to Bins. Associate this object to three Bins: n, n +1, n +2.

8.2.4 Generating Corner Bets

Corner bet Outcomes are as complex as split bets because of the various cases: corners, edges and downthe-middle. Eacn corner has four numbers, n , n +1, n +3, n +4. This is two numbers in the same row, and two numbers in the next higher row. We can generate the corner bets by iterating through the numbers 1, 4, 7, ..., 31 and 2, 5, 8, ..., 32. For each number, n, we generate four corner bets: n , n +1, n+3 , n +4 corner bet. This Outcome object belongs to four Bins. We generate corner bets by iterating through the various corners based on rows and columns. There is room for two corners within the three columns of the layout: one corner starts at column 1 and the other corner starts at column 2. There is room for 11 corners within the 12 rows of the layout. For All Lines Between Rows. For each row, r, 0 r < 11: First Column Number. Set r 3r + 1. This will create values 1, 4, 7, ..., 31. Column 1-2 Corner. Create a n , n +1, n +3, n +4 corner Outcome with odds of 8:1. Assign to Bins. Associate this object to four Bins: n, n +1, n +3, n +4. Second Column Number. Set n 3r + 2. This will create values 2, 5, 8, ..., 32. Column 2-3 Corner. Create a n , n +1, n +3, n +4 corner Outcome withs odds of 8:1. Assign to Bins. Associate this object to four Bins: n, n +1, n +3, n +4.

8.2. Bin Builder Algorithms

Building Skills in Object-Oriented Design, Release 2.1.1-Python

8.2.5 Generating Line Bets

Line bet Outcomes are similar to street bets. However, these are based around the 11 lines between the 12 rows. For lines s numbered 0 to 10, the numbers on the line bet can be computed as follows: 3s + 1, 3s + 2, 3s + 3, 3s + 4, 3s + 5, 3s + 6. This Outcome object belongs to six individual Bins. For All Lines Between Rows. For each row, r, 0 r < 10: First Column Number. Set n 3r + 1. This will create values 1, 4, 7, ..., 31. Line. Create a n, n +1, n +2, n +3, n +4, n +5 line Outcome withs odds of 5:1. Assign to Bins. Associate this object to six Bins: n, n +1, n +2, n +3, n +4, n +5.

8.2.6 Generating Dozen Bets

Dozen bet Outcomes require enumerating all twelve numbers in each of three groups. For All Dozens. For each dozen, d, 0 d < 3: Create Dozen. Create an Outcome for dozen d +1 with odds of 2:1. For All Numbers. For each number, m, 0 m < 12: Assign to Bin. Associate this object to Bin 12d + m + 1.

8.2.7 Generating Column Bets

Column bet Outcomes require enumerating all twelve numbers in each of three groups. While the outline of the algorithm is the same as the dozen bets, the enumeration of the individual numbers in the inner loop is slightly dierent. For All Columns. For each column, c, 0 c < 3: Create Column. Create an Outcome for column c +1 with odds of 2:1. For All Rows. For each row, r, 0 r < 12: Assign to Bin. Associate this object to Bin 3r + c + 1.

8.2.8 Generating Even-Money Bets

The even money bet Outcomes are relatively easy to generate. Create the Red outcome, with odds of 1:1. Create the Black outcome, with odds of 1:1. Create the Even outcome, with odds of 1:1. Create the Odd outcome, with odds of 1:1. Create the High outcome, with odds of 1:1. Create the Low outcome, with odds of 1:1. For All Numbers. For each number, n, 1 n < 37:

Chapter 8. Bin Builder Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Low? If 1 n < 19, associate the low Outcome with Bin n. High? Otherwise, 19 n < 37, associate the high Outcome with Bin n. Even? If nmod2 = 0, associate the even Outcome with Bin n. Odd? Otherwise, nmod2 = 0, associate the odd Outcome with Bin n. Red? If n is one of 1, 3, 5, 7, 9, 12, 14, 16, 18, 19, 21, 23, 25, 27, 30, 32, 34, or 36, associate the red Outcome with Bin n. Black? Otherwise, associate the black Outcome with Bin n.

8.3 BinBuilder Design

class BinBuilder() BinBuilder creates the Outcomes for all of the 38 individual Bin on a Roulette wheel.

8.3.1 Constructors
__init__(self ) Initializes the BinBuilder.

8.3.2 Methods
buildBins(self, wheel ) Creates the Outcome instances and uses the addOutcome() method to place each Outcome in the appropropriate Bin of wheel . Parameter wheel (Wheel) The Wheel with Bins that must be populated with Outcomes. There should be separate methods to generate the straight bets, split bets, street bets, corner bets, line bets, dozen bets and column bets, even money bets and the special case of zero and double zero. Each of the methods will be relatively simple and easy to unit test. Details are provided in Bin Builder Algorithms.

8.4 Bin Builder Deliverables

There are three deliverables for this exercise. The new classes will have Javadoc comments or Python docstrings. The BinBuilder class. A class which performs a unit test of the BinBuilder class. The unit test invoke each of the various methods that create Outcome instances. Rework the unit test of the Wheel class. The unit test should create and initialize a Wheel. It can use the Wheel.getBin() method to check selected Bins for the correct Outcomes.

8.3. BinBuilder Design

Building Skills in Object-Oriented Design, Release 2.1.1-Python

8.5 Internationalization and Localization

An an advanced topic, we need to avoid hard coding the names of the bets. Both Java and Python provide extensive tools for localization (l10n) of programs. Since both applications already use Unicode strings, they support non-Latin characters, handling much of the internationalization (i18n) part of the process. Python localization depends on the locale and gettext modules. The locale module provides a number of functions and constants to help format dates and times for the users selected locale. The gettext module provides a mechanism for getting translated text from a message catalog with messages in a variety of languages. We wont dwell on the l10n issue. However, the right preparation for l10n is to isolate all Strings used for messages, and format all dates and numbers with formatters found in the locale module. In this case, our Outcome names may be displayed, and should be isolated into a separate object. Even something as simple as blackBetName = _("Black") will set the stage for fetching strings from a gettext message catalog. In order to dene the _(string) function, the following fragment should be used. While not ubiquitous in Python programs, doing this now establishes a standard practice that will permit easy localization of an application program.
import gettext gettext.NullTranslations().install() print _("String To Be Translated")

1. This constructs an instance of NullTranslations. This is then installed as a function named _(), which nds a translation for the default C-locale string. This class does nothing; but it can be replaced with the GNUTranslations class, which uses a message catalog to replace the default strings with localized strings. 2. This shows how to use the _() function to translate a default C-locale string to a localized string.

Chapter 8. Bin Builder Class

CHAPTER

NINE

ROULETTE BET CLASS

In addition to the design of the Bet class, this chapter also presents some additional questions and answers on the nature of an object, identity and state change. This continues some of the ideas from Design Decision Object Identity.

9.1 Roulette Bet Overview

A Bet is an amount that the player has wagered on a specic Outcome. This class has the responsibilty for maintaining the association between an amount, an Outcome, and a specic Player. This is the classic record declaration: a passive association among data elements. The only methods we can see are three pairs of getters and setters to get and set each of the three attributes. The general scenario is to have a Player construct a number of Bet instances. The wheel is spun to select a winning Bin. Then each of the Bet objects will be checked to see if they are winners or losers. A winning Bet has an Outcome that matches one in the winning Bin; winners will return money to the Player. All other bets arent in the winning Bin; they are losers, which removes the money from the Player. Locating Outcomes. We have a design decision to make. Do we create a fresh Outcome object with each Bet or do we locate an existing Outcome object?

9.2 Design Decision Create or Locate an Outcome

Building a Bet involves two parts: an Outcome and an amount. Problem. Do we get an create an Outcome object with a Bet? Or do we locate an Outcome object in some collection or as a variable? Forces. One option is to create an Outcome object as part of constructing a Bet object. Heres what it might look like in Python.
my_bet= Bet( Outcome("red", 2), 25 )

One unfortunate feature of this is that we have to repeat the odds when creating an Outcome. This violates the DRY principle. The other option is to locate an existing Outcome object as part of creating a Bet instance. Heres what it might look like in Python.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

my_bet= Bet( someSource["red"], 25 )

Solution. We want to get a complete Outcome from just the name of the outcome. The will prevent repeating the odds information. Consequence. There are several consequences to this decision. We need to pick a collection. When we look at the collections (see Design Decision Choosing A Collection) we can see that a Map from name to complete Outcome instance is ideal. This helps us associate a Bet with an Outcome given just the name of the Outcome. Some object needs to build the collection of distinct Outcomes. Some object needs to maintain the collection of Outcomes for use by the Player in building Bets. If the builder and maintainer are the same object, then things would be somewhat simpler because all the responsibilities would fall into a single place.

9.3 Design Decision Where to Keep the Outcome Map

Building a Bet involves two parts: an Outcome and an amount. Since we cant create an outcome, we need to get an outcome from some other object. There are several objects that could contain this Outcome map. Problem. What class contains the mapping from name to Outcome so we can reuse references to those objects? Forces. In Outcome Class we looked at three ways to achieve a single instance of an Outcome. This assured that we could have multiple references to a single instance of an Outcome. Well modify that list slightly and reuse it here. Global Outcome Mapping. We can declare some global mapping object to keep references to out Outcomes. This object is more-or-less immutable, once the initial set of Outcomes have been built. Its not a global variable, so it is a usable technique. Outcome Factory Object. We can dene a class which produces references to Outcomes as needed. By using this object, we can be sure were getting a reference to the one-and-only instance of the Outcome. The Outcome Class Itself. We can dene the Outcome class to retain a static mapping to each Outcome that gets created. By using the static Outcome.getInstance() method, we can be sure were getting a reference to the one-and-only instance of the Outcome. This, however, limits our ability to create subclasses of Outcome. After creating the BinBuilder, we can see that we in eect chose the Factory design for creating Outcome instances. However, the class:BinBuilder doesnt currently have a handy mapping so that we can look up an Outcome based on the name of the outcome. But is this the right place to do the lookup? What about the Wheel? Alternative Solutions. We have a number of potential ways to gather all Outcome objects that were created by the BinBuilder. Clearly, the BinBuilder can create the mapping from name to each distinct Outcome. To do this, wed have to do several things. First, we expand the BinBuilder to keep a simple Map of the various Outcomes that are being assigned via the Wheel.add() method.

Chapter 9. Roulette Bet Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Second, we would have to add specic Outcome getters to the BinBuilder. We could, for example, include a getOutcome() method that returns an Outcome based on its name. Heres what it might look like in Python.
class BinBuilder( object ): ... def apply( self, outcome, bin, wheel ): self.all_outcomes.add( outcome ) wheel.add( bin, outcome ) def getOutcome( self, name ): ...

Access the Wheel. A better choice is to get Outcome obects from the Wheel. To do this, wed have to do several things. First, we expand the Wheel to keep a simple Map of the various Outcomes created by the BinBuilder. This Map would be maintained by the Wheel.add(). Second, we would have to add specic Outcome getters to the Wheel. We could, for example, include a getOutcome() method that returns an Outcome based on its name. In Python, for example, we might a a method function like the following to Wheel.
class Wheel( object ): ... def add( self, bin, outcome ): self.all_outcomes.add( outcome ) ... def getOutcome( self, name ): return set( [ oc for oc in self.all_outcomes if oc.name.lower().contains( name.lower() ) ] )

Solution. The allocation of responsibility seems to be a toss-up. We can see that the amount of programming is almost identical. This means that the real question is one of clarity: which allocation more clearly states our intention? The Wheel is a rst-class part of the game of Roulette. It showed up in our initial noun analysis. The BinBuilder was an implementation convenience to separate the one-time construction of the Bins from the overall work of the Wheel. Since Wheel is a rst-class part of the problem, we should augment the Wheel to keep track of our individual Outcome objects by name.

9.4 Roulette Bet Questions and Answers

Why not update each Outcome with the amount of the bet? We are isolating the static denition of the Outcome from the presence or absence of an amount wagered. Note that the Outcome is shared by the wheels Bin s, and the available betting spaces on the Table, possibly even the Player class. Also, if we have multiple Player, then we need to distinguish bets placed by the individual players. Changing a elds value has an implication that the thing has changed state. In Roulette, there isnt any state change in the denition of an Outcome. However, when we look at Craps, we will see that changes in the games state will enable and disable whole sets of Outcome s. Does an individual bet really have unique identity? Isnt it just anonymous money?

9.4. Roulette Bet Questions and Answers

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Yes, the money is anonymous. In a casino, the chips all look alike. However, the placement of the bet, really does have unique identity. A Bet is owned by a particular player, it lasts for a specic duration, it has a nal outcome of won or lost. When we want to create summary statistics, we could do this by saving the individual Bet objects. We could update each Bet with a won or lost indicator, then we can total the wins and losses. This points up another reason why we know a Bet is an object in its own right: it changes state. A bet that has been placed can change to a bet that was won or a bet that was lost.

9.5 Roulette Bet Design

class Bet() Bet associates an amount and an Outcome. In a future round of design, we can also associate a Bet with a Player.

9.5.1 Fields
amountBet The amount of the bet. outcome The Outcome on which the bet is placed.

9.5.2 Constructors
__init__(self, amount, outcome) Parameters amount (int) The amount of the bet. outcome (Outcome) The Outcome were betting on. Create a new Bet of a specic amount on a specic outcome. For these rst exercises, well omit the Player. Well come back to this class when necessary, and add that capability back in to this class.

9.5.3 Methods
winAmount(self ) Returns amount won Return type int Uses the Outcomes winAmount to compute the amount won, given the amount of this bet. Note that the amount bet must also be added in. A 1:1 outcome (e.g. a bet on Red) pays the amount bet plus the amount won. loseAmount(self ) Returns amount lost Return type int

Chapter 9. Roulette Bet Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Returns the amount bet as the amount lost. This is the cost of placing the bet. __str__(self ) Returns string representation of this bet with the form "amount on outcome" Return type str Returns a string representation of this bet. Note that this method will delegate the much of the work to the __str__() method of the Outcome.

9.6 Roulette Bet Deliverables

There are four deliverables for this exercise. The new classes will have Javadoc comments or Python docstrings. The expanded Wheel class which creates a mapping of string name to Outcome. Expanded unit tests of Wheel that conrm that the mapping is being built correctly. The Bet class. A class which performs a unit test of the Bet class. The unit test should create a couple instances of Outcome, and establish that the winAmount() and loseAmount() methods work correctly.

9.6. Roulette Bet Deliverables

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 9. Roulette Bet Class

CHAPTER

TEN

ROULETTE TABLE CLASS

This section provides the design for the Table to hold the bets. It also introduces the concepts behind exception handling, and the proper role of exceptions.

10.1 Roulette Table Overview

The Table has the responsibility to keep the Bets created by the Player. Additionally, the house imposes table limits on the minimum amount that must be bet and the maximum that can be bet. Clearly, the Table has all the information required to evaluation these conditions. Note: Betting Constraints Casinos prevent the Martingale betting system from working by imposing a table limit on each game. To cover the cost of operating the table game, the casino also imposes a minimum bet. Typically, the maximum is a multiplier of the minimum bet, often in the range of 10 to 50; a table with a $5 minimum might have a $200 limit, a $10 minimum may have only a $300 limit. It isnt clear where the responsibility lies for determining winning and losing bets. The money placed on Bets on the Table is at risk of being lost. If the bet is a winner, the house pays the Player an amount based on the Outcomes odds and the Bets amount. If the bet is a loser, the amount of the Bet is forfeit by the Player. Looking forward to stateful games like Craps, well place the responsibility for determining winners and losers with the game, and not with the Table object. Well wait, then, until we write the game to nalize paying winning bets and collecting losing bets. Winning vs. Losing. Our second open question is the timing of the payment for the bet from the players stake. In a casino, the payment to the casino eectively happens when the bet is placed on the table. In our Roulette simulation, this is a subtlety that doesnt have any practical consequences. We could deduct the money as part of Bet creation, or we could deduct the money as part of resolving the spin of the wheel. In other games, however, there may several events and several opportunities for placing additional bets. For example, splitting a hand in blackjack, or placing additional odds bets in Craps. Because we cant allow a player to bet more than their stake, we should deduct the payment as the Bet is created. A consequence of this is a change to our denition of the Bet class. We dont need to compute the amount that is lost. Were not going to deduct the money when the bet resolved, were going to deduct the money from the Players stake as part of creating the Bet. This will become part of the design of Player and Bet. Looking forward a little, a stateful game like Craps will introduce a subtle distinction that may be appropriate for a future subclass of Table. When the game is in the point o state, some of the bets on the table are not allowed, and others become inactive. When the game is in the point on state, all bets are allowed and

Building Skills in Object-Oriented Design, Release 2.1.1-Python

active. In Craps parlance, some bets are not working or working depending on the game state. This does not apply to the version of Table that will support Roulette. Container Implementation. A Table is a collection of Bets. We need to choose a concrete class for the collection of the bets. We can review the survey of collections in Design Decision Choosing A Collection for some guidance here. In this case, the bets are placed in no particular order, and are simply visited in an arbitrary order for resolution. Bets dont have specic names. Since the number of bets varies, we cant use a Python tuple; a list will do. Table Limits. Table limits can be checked by providing a public method isValid() that compares the total of a new prospective amount plus all existing Bet s to the table limit. This can be used by the Player to evaluate each potential bet prior to creating it. In the unlikely event of the Player object creating an illegal Bet, we can also throw (or raise) an exception to indicate that we have a design error that was not detected via unit testing. This should be a subclass of Exception that has enough information to debug the problem with the Player that attempted to place the illegal bet. Additionally, the game can check the overall state of a Players Bets to be sure that the table minimum is met. Well need to provide a public method isValid() that is used by the game. In the event of the minimum not being met, there are serious design issues, and an exception should be thrown. Generally, this situation arises because of a bug where the Player should have declined to bet rather than placing incorrect bets that dont meet the table minimum. Bet Resolution. An important consideration is the collaboration between Table and some potential game class for resolving bets. The Table has the collection of Bets, each of which has a specic Outcome. The Wheel selects a Bin, which has a collection of Outcomes. All bets with a winning outcome will be resolved as a winner. So well have multiple Outcomes in the winning Bin to be matched against multiple Bets active on the Table. Because some games are stateful, and the winning and losing bets depend on game state, we will defer the details of this collaboration design until we get to the Game class. For now, well simply collect the Bets. Adding and Removing Bets. A Table contains Bets. Instances of Bet are added by a Player . Later, Bets will be removed from the Table by the Game. When a bet is resolved, it must be deleted. Some games, like Roulette resolve all bets with each spin. Other games, like Craps, involve multiple rounds of placing and resolving some bets, and leaving other bets in play. For Bet deletion to work, we have to provide a method to remove a bet. When we look at Game and bet resolution well return to bet deletion. Its import not to over-design this class at this time; we will often add features as we develop designs for additional use cases.

10.2 InvalidBet Exception Design

exception InvalidBet InvalidBet is raised when the Player attempts to place a bet which exceeds the tables limit. This class simply inherits all features of its superclass.

Chapter 10. Roulette Table Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

10.3 Roulette Table Design

class Table() Table contains all the Bet s created by the Player. A table also has a betting limit, and the sum of all of a players bets must be less than or equal to this limit. We assume a single Player in the simulation.

10.3.1 Fields
limit This is the table limit. The sum of a Player s bets must be less than or equal to this limit. bets This is a list of the Bets currently active. These will result in either wins or losses to the Player.

10.3.2 Constructors
Table() Creates an empty list of bets. Methods isValid(self, bet ) Parameter bet (Bet) A Bet instance to be validated. Validates this bet. If the sum of all bets is less than or equal to the table limit, then the bet is valid, return true. Otherwise, return false. placeBet(self, bet ) Parameter bet (Bet) A Bet instance to be validated. Raises InvalidBet Adds this bet to the list of working bets. If the sum of all bets is greater than the table limit, then an exception should be thrown (Java) or raised (Python). This is a rare circumstance, and indicates a bug in the Player more than anything else. __iter__() Returns an iterator over the available list of Bet instances. This simply returns the iterator over the list of Bet objects. Note that we need to be able remove Bets from the table. Consequently, we have to update the list, which requires that we create a copy of the list. This is done with bets[:]. Returns iterator over all bets __str__() Return String representation of all current bets.

10.4 Roulette Table Deliverables

There are three deliverables for this exercise. Each of these will have complete Javadoc comments or Python docstring comments.

10.3. Roulette Table Design

Building Skills in Object-Oriented Design, Release 2.1.1-Python

An InvalidBet exception class. This is a simple subclass of Exception. Since theres no unique programming here, theres no compelling need to construct a unit test. The Table class. A class which performs a unit test of the Table class. The unit test should create at least two instances of Bet, and establish that these Bet s are managed by the table correctly.

Chapter 10. Roulette Table Class

CHAPTER

ELEVEN

ROULETTE GAME CLASS

Between Player and Game, we have a chicken-and-egg design problem. In this chapter, well describe the design for Game in detail. However, in order to create the deliverables, we have to create a version of Player that we can use just to get started. In the long run, well need to create a sophisticated hierarchy of players. Rather than digress too far, well create a simple player, Pssenger57 (they always bet on black), which will be the basis for further design in later chapters.

11.1 Roulette Game Overview

The RouletteGames responsibility is to cycle through the various steps of the game procedure, getting bets from the player, spinning the wheel and resolving the bets. This is an active class that makes use of the classes we have built so far. The hallmark of an active class is longer or more complex methods. This is distinct from most of the classes we have considered so far, which have relatively trivial methods that are little more than getters and setters of instance variables. The sequence of operations in one round of the game is the following.

A Single Round of Roulette

1. Place Bets. Notify the Player to create Bet s. The real work of placing bets is delegated to the Player class. Note that the money is committed at this point; they players stake should be reduced as part of creating a Bet. 2. Spin Wheel. Get the next spin of the Wheel, giving the winning Bin, w. This is a collection of individual Outcomes which will be winners. 3. Resolve All Bets. For each Bet, b placed by the Player: (a) Winner? If Bet bs Outcome is in the winning Bin, w, then notify the Player that Bet b was a winner and update the Players stake. (b) Loser? If Bet bs Outcome is not in the winning Bin, w, then notify the Player that Bet b was a loser. This allows the Player to update the betting amount for the next round. Matching Algorithm. This Game class will have the responsibility for matching the collection of Outcomes in the Bin of the Wheel with the collection of Outcomes of the Bets on the Table. We have two collections that must be matched: Bin and Table. Well need to structure a loop or nested loops to compare individual elements from these two collections.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Driven by Bin. We could use a loop to visit each Outcome in the winning Bin. For each Outcome, we would then visit each of the Bets contained by the Table. A Bets Outcome that matches the Bins Outcome is a winner and is paid o. The other bets are losers. This involves two nested loops: one to visit the winning Outcomes of a Bin and one to visit the Bet s of a Table. Driven by Table. The alternative is to visit each Bet contained by the Table. Since the winning Bin is has a Set of Outcomes, we can exploit set membership methods to test for presence or absence of an Bets Outcome in the winning Bins Set. If the Bin contains the Outcome, the Bet is a winner; otherwise the Bet is a loser. This only requires a single loop to visit the Bet s of a Table. Player Interface. The Game collaborates with Player. We have a chicken and egg problem in decomposing the relationship between these classes. We note that the Player is really a complete hierarchy of subclasses, each of which provides a dierent betting strategy. For the purposes of making the Game work, we can develop our unit tests with a stub for Player that simply places a single kind of Bet. Well call this player Passenger57 because it always bets on Black. In several future exercises, well revisit this design to make more sophisticated players. For some additional design considerations, see Additional Roulette Design Considerations. This provides some more advanced game options that our current design can be made to support. Well leave this as an exercise for the more advanced student.

11.2 Passenger57 Design

class Passenger57() Passenger57 constructs a Bet based on the Outcome named "Black". This is a very persistent player. Well need a source for the Black outcome. We have several choices; we looked at these in Roulette Bet Class. We will query the Wheel for the needed Outcome object. In the long run, well have to dene a Player superclass, and make Passenger57 class a proper subclass of Player. However, our focus now is on getting the Game designed and built.

11.2.1 Fields
black This is the outcome on which this player focuses their betting. This Player will get this from the Wheel using a well-known bet name. table The Table that is used to place individual Bets.

11.2.2 Constructors
__init__(self, table) Parameter table (Table) A Table instance on which bets are placed. Constructs the Player with a specic table for placing bets. This also creates the black Outcome. This is saved in a variable named Passenger57.black for use in creating bets.

Chapter 11. Roulette Game Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

11.2.3 Methods
placeBets(self ) Updates the Table with the various bets. This version creates a Bet instance from the black Outcome. It uses Table placeBet() to place that bet. win(self, bet ) Parameter bet (Bet) The bet which won. Notication from the Game that the Bet was a winner. The amount of money won is available via a the winAmount() method of theBet. lose(self, bet ) Parameter bet (Bet) The bet which won. Notication from the Game that the Bet was a loser.

11.3 Roulette Game Design

class Game() Game manages the sequence of actions that denes the game of Roulette. This includes notifying the Player to place bets, spinning the Wheel and resolving the Bets actually present on the Table.

11.3.1 Fields
wheel The Wheel that returns a randomly selected Bin of Outcomes. table The Table which contains the Bets placed by the Player. player The Player which creates Bets at the Table.

11.3.2 Constructors
We based the Roulette Game constructor on a design that allows any of the elds to be replaced. This is the Strategy design pattern. Each of these collaborating objects is a replaceable strategy, and can be changed by the client that uses this game. Additionally, we specically do not include the Player instance in the constructor. The Game exists independently of any particular Player, and we defer binding the Player and Game until we are gathering statistical samples. __init__(self, wheel, table) Parameters wheel (Wheel) The wheel which produces random events table (Table) The table which holds bets to be resolved. Constructs a new Game, using a given Wheel and Table.

11.3. Roulette Game Design

Building Skills in Object-Oriented Design, Release 2.1.1-Python

11.3.3 Methods
cycle(self, player ) Parameter player (Player) the individual player that places bets, receives winnings and pays losses. This will execute a single cycle of play with a given Player. It will call thePlayers placeBets() method to get bets. It will call theWheels next() method to get the next winning Bin. It will then call theTables iterator to get an Iterator over the Bets. Stepping through this Iterator returns the individual Bet objects. If the winning Bin contains the Outcome, call the thePlayer win() method otherwise call the thePlayer lose() method.

11.4 Roulette Game Questions and Answers

Why are Table and Wheel part of the constructor while Player is given as part of the cycle() method? We are making a subtle distinction between the casino table game (a Roulette table, wheel, plus casino sta to support it) and having a player step up to the table and play the game. The game exists without any particular player. By setting up our classes to parallel the physical entities, we give ourselves the exibility to have multiple players without a signicant rewrite. We allow ourselves to support multiple concurrent players or multiple simulations each using a dierent player object. Also, as we look forward to the structure of the future simulation, we note that the game objects are largely xed, but there will be a parade of variations on the player. We would like a main program that simplies inserting a new player subclass with minimal disruption. Why do we have to include the odds with the Outcome ? This pairing makes it diicult to create an Outcome from scratch. The odds are an essential ingredient in the Outcome . It turns out that we want a short-hand name for each Outcome. We have three ways to provide a short name. A variable name. Since each variable is owned by a specic class instance, we need to allocate this to some class. The Wheel or the BinBuilder make the most sense for owning this variable. A key in a mapping. In this case, we need to allocate the mapping to some class. Again, the Wheel or BinBuilder make the most sense for owning the mapping. A method which returns the Outcome . The method can use a xed variable or can get a value from a mapping.

11.5 Roulette Game Deliverables

There are three deliverables for this exercise. The stub does not need documentation, but the other classes do need complete Javadoc or Python docstrings. The Passenger57 class. We will rework this design later. This class always places a bet on Black. Since this is simply used to test Game , it doesnt deserve a very sophisticated unit test of its own. It will be replaced in a future exercise. The RouletteGame class.

Chapter 11. Roulette Game Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A class which performs a demonstration of the Game class. This demo program creates the Wheel , the stub Passenger57 and the Table . It creates the Game object and cycles a few times. Note that the Wheel returns random results, making a formal test rather diicult. Well address this testability issue in the next chapter.

11.6 Additional Roulette Design Considerations

In European casinos, the wheel has a single zero. In some casinos, the zero outcome has a special en prison rule: all losing bets are split and only half the money is lost, the other half is a push and is returned to the player. The following design notes discuss the implementation of this additional rule. This is a payout variation that depends on a single Outcome. We will need an additional subclass of Outcome that has a more sophisticated losing amount method: it would push half of the amount back to the Players stake. Well call this subclass the the PrisonOutcome class. In this case, we have a kind of hybrid resolution: it is a partial loss of the bet. In order to handle this, well need to have a loss() method in Bet as well as a win() method. Generally, the loss() method does nothing (since the money was removed from the Players stake when the bet was created.) However, for the PrisonOutcome class, the loss() method returns half the money to the Player. We can also introduce a subclass of BinBuilder that creates only the single zero, and uses this new PrisonOutcome subclass of Outcome for that single zero. We can call this the EuroBinBuilder . The EuroBinBuilder does not create the ve-way Outcome of 00-0-1-2-3, either; it creates a four-way for 0-1-2-3. After introducing these two subclasses, we would then adjust Game to invoke the loss() method of each losing Bet, in case it resulted in a push. For an American-style casino, the loss() method does nothing. For a European-style casino, the los() method for an ordinary Outcome also does nothing, but the los() for a PrisonOutcome would implement the additional rule pushing half the bet back to the Player . The special behavior for zero then emerges from the collaboration between the various classes. We havent designed the Player yet, but we would have to bear this push rule in mind when designing the player. The uniform interface between Outcome and PrisonOutcome is a design pattern called polymorphism. We will return to this principle many times.

11.6. Additional Roulette Design Considerations

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 11. Roulette Game Class

CHAPTER

TWELVE

REVIEW OF TESTABILITY
This chapter presents some design rework and implementation rework for testability purposes. While testability is very important, new programmers can be slowed to a crawl by the mechanics of building test drivers and test cases. We prefer to emphasize the basic design considerations rst, and address testability as a feature to be added to a working class. Additionally, well address some issues in construction of class instances and an idealized structure for the main procedure.

12.1 Testability Overview

Without pausing, we charged past the elephant standing in the saloon. Its time to pause a moment a take a quick glance back at the pachyderm we ignored. Back in Wheel Class we skipped over the problem of testing an application that includes a random number generator (RNG). There are two questions raised: 1. How can we develop formalized unit tests when we cant predict the random outcomes? This is a serious testability issue in randomized simulations. This question also arises when considering interactive applications, particularly for performance tests of web applications where requests are received at random intervals. 2. Are the numbers really random? This is a more subtle issue, and is only relevant for more serious applications. Cryptographic, applications may care more deeply about the randomness of random numbers. This is a large subject, and well beyond the scope of this book. Well just assume that our random number generator is good enough. To address the testing issue, we need to develop some scaolding that permits controlled testing. There are three approaches to replacing the random behavior with something more controlled. One approach is to create a subclass of Wheel to create a more testable version without the random number generator. This MockWheel call be used to test classes that require a specic sequence of outcomes. A second approach is to record the sequence of random numbers actually generated from a particular seed value and use this to dene the exected test results. Instead of changing Wheel or analyzing the results of a random number generator, we can dene a subclass of java.util.Random (or Pythons random.Random) that isnt actually random. Without much analysis, we suggest this in Wheel Class. This is called the Mock Objects approach. We created a Mock Random Number Generator.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

On Random Numbers Random numbers arent actually random. Since they are generated by an algorithm, they are sometimes called pseudo-random. The distinction is important because pseudo-random numbers are generated in a xed sequence from a given seed value. Computing the next value in the sequence involves a simple calculation that is expected to overow the available number of bits of precision leaving apparently random bits as the next value. This leads to results which, while predictable, are arbitrary enough that they pass rigorous statistical tests and are indistinguishable from data created by random processes. We can make an application less predictable by choosing a very hard to predict seed value. One popular choice for the seed is the system clock. In some operating systems a special device is available for producing random seed values. In Linux this is typically /dev/random. As a consequence, we can make an application more predictable by setting a xed seed value and noting the sequence of numbers generated. We can write a short demonstration program to see the eect of setting a xed seed. This will also give us a set of predictable answers for unit testing.

12.2 Test-Driven Design

Good testability is achieved when there are no changes to the target software. For this reason, we dont want to have two versions of Wheel. Instead of having two versions of Wheel, its slightly better to have a random number generator that creates a known sequence with which we can test. To get this known sequence, we have a choice between creating a non-random subclass of java.util.Random or controlling the seed for the random number generator used by Wheel. Both will produce a known sequence of non-random values. An important consequence of these design decisions is that we have to make the random number generator in Wheel visible. This creates a tension between the encapsulation principle and the testability principle. By Encapsulation we mean the design strategy where we dene a class to encapsulate the details of its implementation. Its unclear if the random number generator is an implementation detail or an explicit part of the wheel. Generally, for must of the normal use cases, the Wheels random number generator is an invisible implementation detail. However, for testing purposes, the random number generator needs to be a congurable feature of the Wheel instance. Our favorite approach to making something more visible is to provide default values in the constructor for the object.
class Wheel( object ): def __init___( self, rng=None ): self.rng= rng if rng is not None else random.Random() ... rest of Wheel construction ...

Note: Python Subclass Design Because of the late binding in Python, and the lack of compile-time type checking, we dont have to use strict subclass inheritance to replace the Python random number generator with our own non-random number generator. We only have to provide a class with the minimal interface that our application requires. While legal, we prefer not to bypass the inheritance tree in Python. We nd that applications are much easier to maintain if reasonable care is taken to assure that the classes and interfaces are matched up as precisely as possible. In this case, well want to make our non-random generator a strict subclass of the Random class. 78 Chapter 12. Review of Testability

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Note that the Mock Objects approach follows the principle of deferred binding or dependency injection. A class doesnt choose which implementation it will use; the decision is made somewhere else in the application. Ideally, by the user. Weve allowed the following options. Wheel can be built with an instance of Random initialized with a given seed. Wheel can be build with instance of a new subclass, NonRandom. Wheel can create its own random number generator if no other choice is given.

12.3 Capturing Pseudo-Radom Data

Well need to build and execute a program that reveals the xed sequence of spins that are created by the non-random number generator. Create an instance of NonRandom or Random. Set the seed of the this random number generator to a xed value, for example, 3. Create an instance of Wheel using this non-random number generator. Call the Wheel.next() method six times, and print the winning Bin instances. This sequence will always be the result for a seed value of 3. This discovery procedure will give you the results needed to create unit tests for Wheel and anything that uses it, for example, Game.

12.4 Testability Questions and Answers

Why are we making the random number generator more visible? Isnt object design about encapsulation? Encapsulation isnt the same thing as information hiding . For some people, the information hiding concept can be a useful way to begin to learn about encapsulation. However, information hiding is misleading because it is often taken to exremes. In this case, we want to encapsulate the bins of the wheel and the procedure for selecting the winning bin into a single object. However, the exact random-number generator (RNG) is a separate component, allowing us to bind any suitable RNG. Consider the situation where we are generating random numbers for a cryptographic application. In this case, the built-in random number generator may not be random enough. In this case, we may have a third-party Super-Random-Generator that should replace the built-in generator. We would prefer to minimize the changes required to introduce this new class. Our initial design has isolated the changes to the Wheel, but required us to change the constructor. Since we are changing the source code for a class, we must to unit test that change. Further, we are also obligated unit test all of the classes that depend on this class. Changing the source for a class deep within the application forces us to endure the consequence of retesting every class that depends on this deeply buried class. This is too much work to simply replace one object with another. We do, however, have an alternative. We can change the top-level main() method, altering the concrete object instances that compose the working application. By making the change at the top of the application, we dont need to change a deeply buried class and unit test all the classes that depend on the changed class. Instead, we are simply choosing among objects with the same superclass or interface.

12.3. Capturing Pseudo-Radom Data

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This is why we feel that constructors should be made very visible using the various design patterns for Factories and Builders. Further, we look at the main method as a kind of master Builder that assembles the objects that comprise the current execution of our application. See our Roulette Solution Questions and Answers FAQ for more on this subject. Looking ahead, we will have additional notes on this topic as we add the SevenReds Player Class subclass of Player. If setting the seed works so well, why make a non-random subclass? While setting the seed is an excellent method for setting up a unit test, not everyone is comfortable with random number generators. The presence of an arbitrary but predictable sequence of values looks too much like luck for some Quality Assurance (QA) managers. While the algorithms for both Python and Java random number generators are published and well-understood, this isnt always suiciently clear and convincing for non-mathematicians. Its often seems simpler to build a non-random subclass than to control the existing class. From another point of view, creating a subclass of a built-in class is a necessary skill. The random number generators are easy classes to extend. Extending the abstract collection classes is also a very useful skill, but somewhat more complex than extending the random number generator.

12.5 Testable Random Events Design

Well present the design modication for Wheel rst. This will be followed by design information for NonRandom in both Java and Python.

12.5.1 Wheel Rework

Wheel contains the 38 individual bins on a Roulette wheel, plus a random number generator. It can select a Bin at random, simulating a spin of the Roulette wheel. Note that we will be rewriting some methods to change their implementation. The external denitions of the methods dont change, but the implementations can (and do) change. This is the real strength of OO design. Once we have a well-dened interface (dened by the public methods and attributes), we are free to change the implementation as necessary. This kind of rework making a fundamental change to the implementation without touching the interface is essential to successful OO programming. Constructors __init__(rng=None) Create a wheel with the given random number generator. This will dene the various bins and outcomes using an instance of BinBuilder. Parameter rng (random.Random) Random number generator Note: Common n00b Mistake Note that you cannot provide an instance of random.Random as the default argument. A random.Random is a mutable object, and using a mutable object as a default value of a function will lead to weird problems because that single object instance will be reused for every Wheel built during testing and actual simulation runs.

Chapter 12. Review of Testability

Building Skills in Object-Oriented Design, Release 2.1.1-Python

12.6 Testability Deliverables

There are three deliverables for this exercise. All of these deliverables need appropriate Javadoc comments or Python docstrings. A modied design for Wheel. This add the second constructor that allows us to omit the random number generator when constructing Wheel. Revised unit tests for Wheel based on the xed sequence of responses from the non-random number generator. Revised unit tests for Game based on this revised version of Wheel based on the xed sequence of responses from the non-random number generator.

12.6. Testability Deliverables

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 12. Review of Testability

CHAPTER

THIRTEEN

PLAYER CLASS
The variations on Player, all of which reect dierent betting strategies, is the heart of this application. In Roulette Game Class, we roughed out a stub class for Player. In this chapter, we will complete that design. We will also expand on it to implement the Matingale betting strategy.

13.1 Roulette Player Overview

We have now built enough infrastructure that we can begin to add a variety of players and see how their betting strategies work. Each player is betting algorithm that we will evaluate by looking at the players stake to see how much they win, and how long they play before they run out of time or go broke. The Player has the responsibility to create bets and manage the amount of their stake. To create bets, the player must create legal bets from known Outcomes and stay within table limits. To manage their stake, the player must deduct money when creating a bet, accept winnings or pushes, report on the current value of the stake, and leave the table when they are out of money. We have an interface that was roughed out as part of the design of Game and Table. In designing Game, we put a placeBets() method in Player to place all bets. We expected the Player to create Bet s and use the placeBet() method of Table class to save all of the individual Bets. In an earlier exercise, we built a stub version of Player in order to test Game. See Passenger57 Design. When we nish creating the nal superclass, Player, we will also revise our Passenger57 to be a subclass of Player, and rerun our unit tests to be sure that our more complete design still handles the basic test cases correctly. Objectives. Our objective is to have a new abstract class, Player, with two new concrete subclasses: a revision to Passenger57 and a new player that follows the Martingale betting system. Well defer some of the design required to collect detailed measurements for statistical analysis. In this rst release, well simply place bets. There are four design issues tied up in Player : tracking stake, keeping within table limits, leaving the table, and creating bets. Well tackle them in separate subsections. Tracking the Stake. One of the more important features we need to add to Player are the methods to track the players stake. The initial value of the stake is the players budget. There are two signicant changes to the stake. Each bet placed will deduct the bet amount from the Players stake. We are stopped from placing bets when our stake is less than the table minimum. Each win will credit the stake. The Outcome will compute this amount for us. Additionally, a push will put the original bet amount back. This is a kind of win with no odds applied.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Well have to design an interface that will create Bet s, reducing the stake. and will be used by Game to notify the Player of the amount won. Additionally, we will need a method to reset the stake to the starting amount. This will be used as part of data collection for the overall simulation. Table Limits. Once we have our superclass, we can then dene the Martingale player as a subclass. This player doubles their bet on every loss, and resets their bet to a base amount on every win. In the event of a long sequence of losses, this player will have their bets rejected as over the table limit. This raises the question of how the table limit is represented and how conformance with the table limit is assured. We put a preliminary design in place in Roulette Table Class. There are several places where we could isolate this responsibility. 1. The Player stops placing bets when they are over the Table limit. In this case, we will be delegating responsibility to the Player hierarchy. In a casino, a sign is posted on the table, and both players and casino sta enforce this rule. This can be modeled by providing a method in Table that simply returns the table limit for use by the Player to keep bets within the limit. 2. The Table provides a valid bet method. This reects a more general situation where a stateful game has bets that change. In Craps, for example, most bets cannot be placed until a point is established. 3. The Table throws an illegal bet exception when an illegal bet is placed. While permissable, this kind of use for exceptions pushes the envelope on clarity and simplicity. One viewpoint is that exceptions should be reserved for situations that are truly unexpected. In this case, we expect to run into the table limit situation fairly often using Martigale betting. We recommend the second choice: adding a isValid() method to the Table class. This has the consequence of allocating responsibility to Table , and permits us to have more advanced games where some bets are not allowed during some game states. It also obligates Player to validate each bet with the Table. It also means that at some point, the player may be unable to place a legal bet. We could also implement this by adding to the responsibilities of the existing placeBet() method. For example, we could return true if the bet was accepted, and false if the bet violated the limits or other game state rules. We prefer to isolate responsibility and create a second method rather than pile too much into a single method. Leaving the Table. In enumerating the consequences of checking for legal bets, we also uncovered the issue of the Player leaving the game. We can identify a number of possible reasons for leaving: out of money, out of time, won enough, and unwilling to place a legal bet. Since this decision is private to the Player , we need a way of alerting the Game that the Player is nished placing bets. There are three mechanisms for alerting the Game that the Player is nished placing bets. 1. Expand the responsibilities of the placeBets() to also indicate if the player wishes to continue or is withdrawing from the game. While most table games require bets on each round, it is possible to step up to a table and watch play before placing a bet. This is one classic strategy for winning at blackjack: one player sits at the table, placing small bets and counting cards, while a confederate places large bets only when the deck is favorable. We really have three player conditions: watching, betting and nished playing. It becomes complex trying to bundle all this extra responsibility into the placeBets() method. 2. Add another method to Player that the Game can use to determine if the Player will continue or stop playing. This can be used for a player who is placing no bets while waiting; for example, a player who is waiting for the Roulette wheel to spin red seven times in a row before betting on black. 3. The Player can throw an exception when they are done playing. This is an exceptional situation: it occurs exactly once in each simulation. However, it is a well-dened condition, and doesnt deserve to be called exceptional . It is merely a terminating condition for the game.

Chapter 13. Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

We recommend adding a method to Player to indicate when Player is done playing. This gives the most exibility, and it permits Game to cycle until the player withdraws from the game. A consequence of this decision is to rework the Game class to allow the player to exit. This is relatively small change to interrogate the Player before asking the player to place bets. Note: Design Evolution In this case, these were situations which we didnt discover during the initial design. It helped to have some experience with the classes in order to determine the proper allocation of responsibilities. While design walkthroughs are helpful, an alternative is a prototype, a piece of software that is incomplete and can be disposed of. The earlier exercise created a version of Game that was incomplete, and a version of PlayerStub that will have to be disposed of. Creating Bets from Outcomes. Generally, a Player will have a few Outcomes on which they are betting. Many systems are similar to the Martingale system, and place bets on only one of the Outcomes. These Outcome objects are usually created during player initialization. From these Outcome s, the Player can create the individual Bet instances based on their betting strategy.

13.2 Player Design

class Player() Well design the base class of Player and a specic subclass, Martingale. This will give us a working player that we can test with. Player places bets in Roulette. This an abstract class, with no actual body for the placeBets() method. However, this class does implement the basic win() method used by all subclasses.

13.2.1 Fields
stake The players current stake. Initialized to the players starting budget. roundsToGo The number of rounds left to play. Initialized by the overall simulation control to the maximum number of rounds to play. In Roulette, this is spins. In Craps, this is the number of throws of the dice, which may be a large number of quick games or a small number of long-running games. In Craps, this is the number of cards played, which may be large number of hands or small number of multi-card hands. table The Table used to place individual Bets. The Table contains the current Wheel from which the player can get Outcomes used to build Bets.

13.2.2 Constructors
__init__(self, table) Constructs the Player with a specic Table for placing Bets. Parameter table (Table) the table to use

13.2.3 Methods
playing(self )

13.2. Player Design

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Returns true while the player is still active. placeBets(self ) Updates the Table with the various Bet s. When designing the Table, we decided that we needed to deduct the amount of a bet from the stake when the bet is created. See the Table Roulette Table Overview for more information. win(self, bet ) Parameter bet (Bet) The bet which won Notication from the Game that the Bet was a winner. The amount of money won is available via bet method winAmount(). lose(self, bet ) Parameter bet (Bet) The bet which won Notication from the Game that the Bet was a loser. Note that the amount was already deducted from the stake when the bet was created.

13.3 Martingale Player Design

class Martingale() Martingale is a Player who places bets in Roulette. This player doubles their bet on every loss and resets their bet to a base amount on each win.

13.3.1 Fields
lossCount The number of losses. This is the number of times to double the bet. betMultiple The the bet multiplier, based on the number of losses. This starts at 1, and is reset to 1 on each win. It is doubled in each loss. This is always equal to 2lossCount .

13.3.2 Methods
placeBets(self ) Updates the Table with a bet on black. The amount bet is 2lossCount , which is the value of betMultiple. win(self, bet ) Parameter bet (Bet) The bet which won Uses the superclass win() method to update the stake with an amount won. This method then resets lossCount to zero, and resets betMultiple to 1. lose(self, bet ) Parameter bet (Bet) The bet which won Uses the superclass lose() to do whatever bookkeeping the superclass already does. Increments lossCount by 1 and doubles betMultiple.

Chapter 13. Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

13.4 Player Deliverables

There are six deliverables for this exercise. The new classes must have Javadoc comments or Python docstrings. The Player abstract superclass. Since this class doesnt have a body for the placeBets(), it cant be unit tested directly. A revised Passenger57 class. This version will be a proper subclass of Player, but still place bets on black until the stake is exhausted. The existing unit test for Passenger57 should continue to work correctly after these changes. The Martingale subclass of Player. A unit test class for Martingale. This test should synthesize a xed list of Outcome s, Bin s, and calls a Martingale instance with various sequences of reds and blacks to assure that the bet doubles appropriately on each loss, and is reset on each win. A revised Game class. This will check the players playing() method before calling placeBets(), and do nothing if the player withdraws. It will also call the players win() and lose() methods for winning and losing bets. A unit test class for the revised Game class. Using a non-random generator for Wheel, this should be able to conrm correct operation of the Game for a number of bets.

13.4. Player Deliverables

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 13. Player Class

CHAPTER

FOURTEEN

OVERALL SIMULATION CONTROL

This section starts to address the overall main program that pulls the various classes together to create a nished application. Additionally, well also address some issues in how Java handles collections of primitive data types like integers.

14.1 Simulation Control Overview

We can now use our application to generate some more usable results. We can perform a number of simulation runs and evaluate the long-term prospects for the Martingale betting system. We want to know a few things about the game: How long can we play with a given budget? In other words, how many spins before weve lost our stake. How much we can realistically hope to win? How large a streak can we hope for? How far ahead can we hope to get before we should quit? The Simulator will be an active class with a number of responsibilities Create the Wheel, Table and Game objects. Simulate a number of sessions (typically 100), saving the maximum stake and length of each session. For each session: initialize the Player and Game , cycle the game a number of times, collect the size of the Player s stake after each cycle. Write a nal summary of the results. At this point, well need for formalize some denitions, just to be crystal clear on the responisbility allocation.

14.2 Simulation Terms

cycle A single cycle of betting and bet resolution. This depends on a single random event: a spin of the wheel or a throw of the dice. Also known as a round of play. session One or more cycles. The session begins with a player having their full stake. A session ends when the play elects to leave or can no longer participate. A player may elect to leave because of elapsed time (typically 250 cycles), or they have won a statistically signicant amount. A player can no longer participate when their stake is too small to make the minimum bet for the table.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

game Some games have intermediate groupings of events between an individual cycles and an entire session. Blackjack has hands where a number of player decisions and a number of random events contribute to the payo. Craps has a game, which starts with the dice roll when the point is o, and ends when the point is made or the shooter gets Craps; consequently, any number of individual dice rolls can make up a game. Some bets are placed on the overall game, while others are placed on individual dice rolls. The sequence of operations for the simulator looks like this.

Controlling the Simulation

1. Empty List of Maxima. Create an empty maxima list. This is the maximum stake at the end of each session. 2. Empty List of Durations. Create an empty durations list. This is the duration of each session, measured in the number of cycles of play before the player withdrew or ran out of money. 3. For All Sessions. For each of 100 sessions: Empty List of Stake Details. Create an empty list to hold the history of stake values for this session. This is raw data that we will summarize into two metrics for the session: maximum stake and duration. We could also have two simple variables to record the maximum stake and count the number of spins for the duratioon. However, the list allows us to gather other statistics, like maximum win or maximum loss. While The Player Is Active. Play One Cycle. Play one cycle of the game. See the denition in Roulette Game Class. Save Outcomes. Save the players current stake in the list of stake values for this session. An alternative is to update the maximum to be the larger of the current stake and the maximum, and increment the duration. Get Maximum. Get the maximum stake from the list of stake values. Save the maximum stake metric in the maxima list. Get Duration. Get the length of the list of stake values. Save the duration metric in the durations list. Durations less than the maximum mean the strategy went bust. 4. Statistical Description of Maxima. Compute the average and standard deviation of the values in the maxima list. 5. Statistical Description of Durations. Compute the average and standard deviation of values in the durations list. Both this overall Simulator and the Game collaborate with the Player. The Simulators collaboration, however, initalizes the Player and then monitors the changes to the Players stake. We have to design two interfaces for this collaboration. Player Initialization. The Simulator will initialize a Player for 250 cycles of play, assuming about one cycle each minute, and about four hours of patience. We will also initialize the player with a generous budget of the table limit, 100 betting units. For a $10 table, this is $1,000 bankroll. Currently, the Player class is designed to play one session and stop when their duration is reached or their stake is reduced to zero. We have two alternatives for reinitializing the Player at the beginning of each session.

Chapter 14. Overall Simulation Control

Building Skills in Object-Oriented Design, Release 2.1.1-Python

1. Provide some setters that allow a client class like this overall simulator control to reset the stake and roundsToGo values of a Player. 2. Provide a Factory that allows a client class to create new, freshly initialized instances of Player. While the rst solution is quite simple, there are some advantages to creating a PlayerFactory. If we create an Abstract Factory, we have a single place that creates all Players. Further, when we add new player subclasses, we introduce these new subclasses by creating a new subclass of the factory. In this case, however, only the main program creates instances of Player, reducing the value of the factory. While design of a Factory is a good exercise, we can scrape by with adding setter methods to the Player class. Player Interrogation. The Simulator will interrogate the Player after each cycle and capture the current stake. An easy way to manage this detailed data is to create a List that contains the stake at the end of each cycle. The length of this list and the maximum value in this list are the two metrics the Simulator gathers for each session. Our list of maxima and durations are created sequentially during the session and summarized sequentially at the end of the session. A Java LinkedList will do everything we need. For a deeper discussion on the alternatives available in the collections framework, see Design Decision Choosing A Collection. Statistical Summary. The Simulator will interrogate the Player after each cycle and capture the current stake. We dont want the sequence of stakes for each cycle, however, we want a summary of all the cycles in the session. We can save the length of the sequence as well as the maximum of the sequence to determine these two aggregate performance parameters for each session. Our objective is to run several session simulations to get averages and a standard deviations for duration and maximum stake. This means that the Simulator needs to retain these statistical samples. We will defer the detailed design of the statistical processing, and simply keep the duration and maximum values in Lists for this rst round of design.

14.3 Simulator Design

class Simulator() Simulator exercises the Roulette simulation with a given Player placing bets. It reports raw statistics on a number of sessions of play. Handling Lists of Values in Java Java programmers cant easily accumulate a List of int values. This is because a primitive int value is not an object. The usual solution is to make an Integer value from the int, and put the Integer object into the list. When getting values back out of the list, there is a two-step dance: cast the item in the list to be an Integer, then get the int value of that object. The following examples show this as two steps.
for( Iterator<Integer> i = someList.iterator(); i.hasNext(); ) { int value = i.next().intValue(); ... processing of value ... }

14.3.1 Fields
initDuration The duration value to use when initializing a Player for a session. A default value of 250 is a good

14.3. Simulator Design

Building Skills in Object-Oriented Design, Release 2.1.1-Python

choice here. initStake The stake value to use when initializing a Player for a session. This is a count of the number of bets placed; i.e., 100 $10 bets is $1000 stake. A default value of 100 is sensible. samples The number of game cycles to simulate. A default value of 50 makes sense. durations A List of lengths of time the Player remained in the game. Each session of play producrs a duration metric, which are collected into this list. maxima A List of maximum stakes for each Player. Each session of play producrs a maximum stake metric, which are collected into this list. player The Player; essentially, the betting strategy we are simulating. game The casino game we are simulating. This is an instance of Game, which embodies the various rules, the Table and the Wheel.

14.3.2 Constructors
__init__(self, game, player ) Saves the Player and Game instances so we can gather statistics on the performance of the players betting strategy. Parameters game (Game) The Game were simulating. This includes the Table and Wheel. player (Player) The Player. This encapsulates the betting strategy.

14.3.3 Methods
session(self ) Returns list of stake values. Return type list Executes a single game session. The Player is initialized with their initial stake and initial cycles to go. An empty List of stake values is created. The session loop executes until the Player playing() returns false. This loop executes the Game cycle(); then it gets the stake from the Player and appends this amount to the List of stake values. The List of individual stake values is returned as the result of the session of play. gather(self ) Executes the number of games sessions in samples. Each game session returns a List of stake values. When the session is over (either the play reached their time limit or their stake was spent), then the length of the session List and the maximum value in the session List are the resulting duration and maximum metrics. These two metrics are appended to the durations list and the maxima list. A client class will either display the durations and maxima raw metrics or produce statistical summaries. 92 Chapter 14. Overall Simulation Control

Building Skills in Object-Oriented Design, Release 2.1.1-Python

14.4 Player Rework

The current design for the Player doesnt provide all the methods we need. Well can add two new methods: one will set the stake and the other will set the roundsToGo. setStake(self, stake) Parameter stake (integer) the Players initial stake setRounds(self, rounds) Parameter rounds (integer) the Players duration of play

14.5 Simulation Control Deliverables

There are ve deliverables for this exercise. Each of these classes needs complete Javadoc or Python docstring comments. Revision to the Player class. Dont forget to update unit tests. The Simulator class. The expected outcomes from the non-random wheel can be rather complex to predict. Because of this, one of the deliverables is a demonstration program that enumerates the actual sequence of non-random spins. From this we can derive the sequence of wins and losses, and the sequence of Player bets. This will allow us to predict the nal outcome from a single session. A unit test of the Simulator class that uses the non-random generator to produce the predictable sequence of spins and bets. A main application function that creates the necessary objects, runs the Simulators gather() method, and writes the available outputs to sys.stdout For this initial demonstration program, it should simply print the list of maxima, and the list of session lengths. This raw data can be redirected to a le, loaded into a spreadsheet and analyzed.

14.4. Player Rework

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 14. Overall Simulation Control

CHAPTER

FIFTEEN

SEVENREDS PLAYER CLASS

This section introduces an additional specialization of the Martingale strategy. Additionally, well also address some issues in how an overall application is composed of individual class instances. Adding this new subclass should be a small change to the main application class.

15.1 SevenReds Player Overview

The SevenReds player waits for seven red wins in a row before betting black. This is a subclass of Player. We can create a subclass of our main Simulator to use this new SevenReds class. We note that Passenger57s betting is stateless: this class places the same bets over and over until they are cleaned out or their playing session ends. The Martingale players betting, however, is stateful. This player changes the bet based on wins and losses. The state is a loss counter than resets to zero on each win, and increments on each loss. Our SevenReds player will have two states: waiting and betting. In the waiting state, they are simply counting the number of reds. In the betting state, they have seen seven reds and are now playing the Martingale system on black. We will defer serious analysis of this stateful betting until some of the more sophisticated subclasses of Player. For now, we will simply use an integer to count the number of reds. Game Changes. Currently, a Player is not informed of the nal outcome unless they place a bet. We designed the Game to evaluate the Bet instances and notify the Player of just their Bets that were wins or losses. We will need to add a method to Player to be given the overall list of winning Outcomes even when the Player has not placed a bet. Once we have updated the design of Game to notify Player, we can add the new SevenReds class. Note that we are can intoduce each new betting strategy via creation of new subclasses. A relatively straightforward update to our simulation main program allows us to use these new subclasses. The previously working subclasses are left in place, allowing graceful evolution by adding features with minimal rework of existing classes. In addition to waiting for the wheel to spin seven reds, we will also follow the Martingale betting system to track our wins and losses, assuring that a single win will recoup all of our losses. This makes SevenReds a further specialization of Martingale. We will be using the basic features of Martingale, but doing additional processing to determine if we should place a bet or not. Introducing a new subclass should be done by upgrading the main program. See Soapbox on Composition for comments on the ideal structure for a main program. Additionally, see the Roulette Solution Questions and Answers FAQ entry.

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Soapbox on Composition Generally, a solution is composed of a number of objects. However, the consequences of this are often misunderstood. Since the solution is a composition of objects, it falls on the main method to do just the composition and nothing more. Our ideal main program creates and composes the working set of objects. In this case, it should decode the command-line parameters, and use this information to create and initialize the simulation objects, then start the processing. For these simple exercises, however, were omitting the parsing of command-line parameters, and simply creating the necessary objects directly. Our main program should, therefore, look something like the following:
theRNG= random.Random() theWheel= Wheel( theRNG ) theTable= Table() theGame= Game( theWheel, theTable ) thePlayer= SevenReds( table ) sim= new Simulator( theGame, thePlayer ) sim.gather()

1. The theRNG variable can also be initialized to an instance of NonRandom. The idea is that we build up the functionality of Wheel by composition. We compose the wheel of the bins plus the number generator. 2. the thePlayer variable can also be Martingale or Passenger57. Again, we have assembled the nal simulation by composition of a Wheel (and its generator), the Table and the specic Player algorithm. 3. This is the actual work of the application, done by the Simulator, using the Game and Player objects built by main. In many instances, the construction of objects is not done directly by the main method. Instead, the main method will construct a base group of Abstract Factory and Builder objects. Given these objects and the command-line parameters, the application objects can be built. The application objects then collaborate to perform the required functions.

15.2 SevenReds Design

class SevenReds() SevenReds is a Martingale player who places bets in Roulette. This player waits until the wheel has spun red seven times in a row before betting black.

15.2.1 Fields
redCount The number of reds yet to go. This starts at 7 , is reset to 7 on each non-red outcome, and decrements by 1 on each red outcome. Note that this class inherits betMultiple. This is initially 1, doubles with each loss and is reset to one on each win.

15.2.2 Methods
placeBets(self )

Chapter 15. SevenReds Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

If redCount is zero, this places a bet on black, using the bet multiplier. winners(self, outcomes) Parameter outcomes (Set of Outcome) The Outcome set from a Bin. This is notication from the Game of all the winning outcomes. If this vector includes red, redCount is decremented. Otherwise, redCount is reset to 7.

15.3 Player Rework

Well need to revise the Player class to add the following method. The superclass version doesnt do anything with this information. Some subclasses, however, will process this. winners(self, outcomes) Parameter outcomes (Set of Outcome) The set of Outcome instances that are part of the current win. The game will notify a player of each spin using this method. This will be invoked even if the player places no bets.

15.4 Game Rework

Well need to revise the Game class to extend the cycle method. This method must provide the winning bins Outcome set.

15.5 SevenReds Player Deliverables

There are six deliverables from this exercise. The new classes will require complete Javadoc comments or Python docstrings. A revision to the Player to add the Player.winners() method. The superclass version doesnt do anything with this information. Some subclasses, however, will process this. A revision to the Player unit tests. A revision to the Game class. This will call the winners() with the winning Bin instance before paying o the bets. The SevenReds subclass of Player. A unit test of the SevenReds class. This test should synthesize a xed list of Outcomes, Bins and the call a SevenReds instance with various sequences of reds and blacks. One test cases can assure that no bet is placed until 7 reds have been seen. Another test case can assure that the bets double (following the Martingale betting strategy) on each loss. A main application function that creates the necessary objects, runs the Simulators gather() method, and writes the available outputs to sys.stdout For this initial demonstration program, it should simply print the list of maxima, and the list of session lengths. This raw data can be redirected to a le, loaded into a spreadsheet and analyzed.

15.3. Player Rework

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Chapter 15. SevenReds Player Class

CHAPTER

SIXTEEN

STATISTICAL MEASURES
This section presents two ordinary statistical algorithms: mean and standard deviation. In one sense, this section could be treated as optional. Instead of having the simulator present useful summary statistics, raw data can be put into a spreadsheet for creating the statistics. However, the process of designing these statistical processing classes is very interesting. This chapter examines some ways to add features to the built-in collection classes.

16.1 Statistics Overview

Currently, the Simulator class collects two Lists. One has the length of a session, the other has the maximum stake during a session. We need to create some descriptive statistics to summarize these stakes and session lengths. On Statistics In principle, we could apply basic probability theory to predict the statistics generated by this simulation. Indeed, some of the statistics we are gathering are almost matters of denition, rather than actual unknown data. We are, however, much more interested in the development of the software than we are in applying probability theory to a relatively simple casino game. As a consequence, the statistical analysis here is a little o-base. In spite of our problems, we will build a relatively simple-minded set of descriptive statistics. We will average two metrics: the peak stake for a session and the duration of the session. Both of these values have some statistical problems. The peak stake, for instance, is already a summary number. We have to be cautious in our use of summarized numbers, since we have lost any sense of the frequency with which the peaks occured. For example, a player may last 250 cycles of play, with a peak stake of 130. We dont know if that peak occurred once or 100 times out of that 250. Additionally, the length of the session has an upper bound on the number of cycles. This forces the distribution of values to be skewed away from the predictable distribution. We will design a Statistics class with responsibility to retain and summarize a list of numbers and produce the average (also known as the mean) and standard deviation. The Simulator can then use this this Statistics class to get an average of the maximum stakes from the list of session-level measures. The Simulator can also apply this Statistics class to the list of session durations to see an average length of game play before going broke. We can encapsulate this statistical processing in two ways: we could extend an existing class or we could delegate the statistical functions to a separate class. If we extend the library java.util.List, we can add the needed statistical summary features; given this new class, we can replace the original List of sample values with a StatisticalList that both saves the values and computes descriptive statistics. In addition 99

Building Skills in Object-Oriented Design, Release 2.1.1-Python

to delegation, we have two choices for implementation of this extended version of List, giving us three alternative designs. 1. Extend LinkedList. In this case, we are simply adding two new methods to the existing LinkedList implementation. However, our new methods would not apply to ArrayList or Vector. 2. Extend AbstractList. However, in doing this we would need to provide the implementation details of the list structure itself. In eect, we would reimplement LinkedList or ArrayList. 3. Create a separate class that computes statistics on a List given as an argument to the methods of the statistical class. In this case, our statistical methods will work with any subclass that implements the interface List. This can be applied to any of LinkedList, ArrayList or Vector. Alternative three a separate statistical class has the most reuse potential. Well create a simple class with two methods: mean() and stdev() to compute the mean and standard deviation of a List of Integer objects. This can be used in a variety of contexts, and allows us the freedom to switch List implementation classes without any other changes. The detailed algorithms for mean and standard deviation are provided in Statistical Algorithms. Since our processes are relative simple, stateless algorithms, we dont need any instance variables. Consequently, well never need to create an instance of this class. As with Java.lang.Math , all of these methods can be declared as static, making them features of the class itself.

16.2 Some Foundations

For those programmers new to statistics, this section covers the Sigma operator, .
n i=0

f (i)

The operator has three parts to it. Below it is a bound variable, i, and the starting value for the range, written as i = 0. Above it is the ending value for the range, usually something like n. To the right is some function to execute for each value of the bound variable. In this case, a generic function, f (i) is shown. This is read as sum f ( i ) for i in the range 0 to n. One common denition of uses a closed range, including the end values of 0 and n . However, since this is not a helpful denition for software, we will dene to use a half-open interval. It has exactly n elements, including 0 and n -1; mathematically, 0 i < n. Consequently, we prefer the following notation, but it is not often used. Since statistical and mathematical texts often used 1-based indexing, some care is required when translating formulae to programming languages that use 0-based indexing. f (i)
0i<n

Our two statistical algorithms have a form more like the following function. In this we are applying some function, f , to each value, xi of an list, x. When computing the mean, as a special case, there is no function applied to the values in the list. When computing standard deviation, the function involves subtracting and multiplying. f (xi )
0i<n

We can transform this denition directly into a for loop that sets the bound variable to all of the values in the range, and does some processing on each value of the List of Integers. 100 Chapter 16. Statistical Measures

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This is the Python implementation of Sigma. This computes two values, the sum, s and the number of elements, n.
s= sum( theList ) n = len( theList )

When computing the standard deviation, we do something like the following

s= sum( f(x) for x in theList ) n = len( theList )

Where the f(x) calculuation computes the measure of deviation from the average.

16.3 Statistical Algorithms

16.3.1 Mean
Computing the mean of a list of values is relatively simple. The mean is the sum of the values divided by the number of values in the list. Since the statistical formula is so closely related to the actual loop, well provide the formula, followed by an overview of the code. xi x =
0i<n

The denition of the mathematical operator leads us to the following method for computing the mean:

Computing The Mean of a List, X, with n elements.

1. Initialize. Set s 0. 2. Iterate. For each value, xi in the list: Sum and Count. Set s s + xi . 3. Result. Return s n.

16.3.2 Standard Deviation

The standard deviation can be done a few ways. Well use the formula shown below. This computes a deviation measurement as the square of the dierence between each sample and the mean. The sum of these measurements is then divided by the number of values times the number of degrees of freedom to get a standardized deviation measurement. Again, the formula summarizes the loop, so well show the formula followed by an overview of the code. x =
0i<n

(xi x )2 n1

The denition of the mathematical operator leads us to the following method for computing the standard deviation:

16.3. Statistical Algorithms

101

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Computing Standard Deviation of a List, X, with n elements.

1. Get the Mean. Compute the mean, m gets mean(x). 2. Initialize. Set s 0. 3. Iterate. For each value, xi in the list: Dierence. Compute the dierence from the mean, d xi m. Sums. Compute the sum, s s + d2 . This is typically done as d*d in Java, since there is no squared operator. In Python, this can be d**2. 4. Variance. Set v s (n 1). This n 1 value reects the statistical notion of degrees of freedom, which is beyond the scope of this book. 5. Result. Return the square root of the variance, v . Tip: Math Functions For programmers new to Python, the math module contains the math.sqrt(), which computes square roots.

16.4 IntegerStatistics Design

class IntegerStatistics() IntegerStatistics computes several simple descriptive statistics of Integer values in a List. In Python, we dont actually need a class denition, we can simply declare two functions outside any class. However, to be consistent with Java, its easier to dene these functions in a class.

16.4.1 Constructors
Since this class has entirely static methods, it doesnt really need a constructor.

16.4.2 Methods
mean(values) This requires the @staticmethod decorator. Parameter values (list) The list of values we are summarizing. Computes the mean of the List of Integer values. stdev(values) This requires the @staticmethod decorator. Parameter values (list) The list of values we are summarizing. Computes the standard deviation of the List of Integer values.

102

Chapter 16. Statistical Measures

Building Skills in Object-Oriented Design, Release 2.1.1-Python

16.5 Statistics Deliverables

There are three deliverables for this exercise. These classes will include the complete Javadoc comments or Python dostring. The IntegerStatistics class. A unit test of the IntegerStatistics class. Prepare some simple list (or tuple) of test data. The results can be checked with a spreadsheet An update to the overall Simulator that gets uses an IntegerStatistics object to compute the mean and standard deviation of the peak stake. It also computest the mean and standard deviation of the length of each session of play. Here is some standard deviation unit test data. Sample Value 9 8 5 9 9 4 5 8 10 7 8 8 Here are some intermediate results and the correct answers given to 6 signicant digits. Your answers should be the same to the precision shown. sum 90 count 12 mean 7.5 sum d*d 608.668 stdev 1.88293

16.5. Statistics Deliverables

103

Building Skills in Object-Oriented Design, Release 2.1.1-Python

104

Chapter 16. Statistical Measures

CHAPTER

SEVENTEEN

RANDOM PLAYER CLASS

This section will introduce a simple subclass of Player who bets at random.

17.1 Random Player Overview

One possible betting strategy is to bet completely randomly. This serves as an interesting benchmark for other betting strategies. Well write a subclass of Player which steps through all of the bets available on the Wheel, selecting one or more of the available outcomes at random. This Player, like others, will have a xed initial stake and a limited amount of time to play. The Wheel class can provide an Iterator over the collection of Bin instances. We could revise Wheel to provide a binIterator() method that we can use to return all of the Bins. From each Bin, we will need an iterator we can use to return all of the Outcomes. To collect a list of all possible Outcomes, we would use the following algorithm:

Locating all Outcomes

1. Empty List of Outcomes. Create an empty set of all Outcomes, all_OC. 2. Get Bin Iterator. Get the Iterator from the Wheel that lists all Bins. 3. For each Bin. Get Outcome Iterator. Get the Iterator that lists all Outcomes. For each Outcome. Save Outcome. Save a reference to each Outcome in the set of all known outcomes, all_OC. To place a random bet, we would use the following algorithm:

Placing a Random Bet

1. Get the size of the pool of all possible Outcome s, s. 2. Get a random number, u, from zero to the total size-1. That is, 0 <= u <= s -1. 3. Return element u from the pool of Outcome s.

105

Building Skills in Object-Oriented Design, Release 2.1.1-Python

17.2 Random Player Design

class PlayerRandom() PlayerRandom is a Player who places bets in Roulette. This player makes random bets around the layout.

17.2.1 Fields
rng A Random Number Generator which will return the next random number.

17.2.2 Constructors
__init__(table, rng) This uses the super() construct to invoke the superclass constructor using the Table. Then it saves the random number generator. This could be an instance of either Random or NonRandom. Parameters table (Table) The Table which will accept the bests. rng (random.Random) A :class:random.Random random number generator.

17.2.3 Methods
placeBets(self ) Updates the Table with a randomly placed bet.

17.3 Random Player Deliverables

There are ve deliverables from this exercise. The new classes need Javadoc comments or Python docstrings. Updates to the class Bin to return an iterator over available Outcomes. Updates to unittests for the class Bin, also. Updates to the Wheel to return an iterator over available Bins. Updates to the unittests for the class Wheel, also. The PlayerRandom class. A unit test of the PlayerRandom class. This should use the NonRandom number generator to iterate through all possible Outcomes. An update to the overall Simulator that uses the PlayerRandom.

106

Chapter 17. Random Player Class

CHAPTER

EIGHTEEN

PLAYER 1-3-2-6 CLASS

This section will describe a player who has a complex internal state. We will also digress on the way the states can be modeled using a group of polymorphic classes. This section also has an advanced exercise that shows some alternative implementations for the state objects, using the Singleton design pattern.

18.1 Player 1-3-2-6 Overview

On the Internet, we found descriptions of a betting system called the 1-3-2-6 system. This system looks to recoup losses by waiting for four wins in a row. The sequence of numbers (1, 3, 2 and 6) are the multipliers to use when placing bets after winning. At each loss, the sequence resets to the multiplier of 1. At each win, the multiplier is advanced. After one win, the bet is now 3x. After a second win, the bet is reduced to 2x, and the winnings of 4x are taken down or removed from play. In the event of a third win, the bet is advanced to 6x. Should there be a fourth win, the player has doubled their money, and the sequence resets. This betting system makes our player more stateful than in previous betting systems. When designing SevenReds, we noted that this player was stateful; that case, the state was a simple count. In this case, the description of the betting system seems to identify four states: no wins, one win, two wins, and three wins. In each of these states, we have specic bets to place, and state transition rules that tell us what to do next. The following table summarizes the states, the bets and the transition rules. Table 18.1: 1-3-2-6 Betting States State No Wins One Win Two Wins Three Wins Bet 1 3 2 6 Next State On Win One Win Two Wins Three Wins No Wins On Loss No Wins No Wins No Wins No Wins

When we are in a given state, the table gives us the amount to bet in the Bet column. If this bet wins, we transition to the state named in the On Win column, otherwise, we transition to the state named in the On Loss column. We always start in the No Wins state. Design Pattern. We can exploit the State design pattern to implemente this more sophisticated player. This pattern suggests that we design a hierarchy of classes to represent these four states. Each state will have a slightly dierent bet amount, and dierent state transition rules. Each individual state class will be relatively simple, but we will have isolated the processing unique to each state into separate classes. One of the consequences of the State design pattern is that it obligates us to dene the interface between the Player and the object that holds the Players current state.

107

Building Skills in Object-Oriented Design, Release 2.1.1-Python

It seems best to have the state object follow our table and provide three methods: currentBet(), nextWon(), and nextLost(). The Player can use these methods of the state object to place bets and pick a new state. A states currentBet() method will construct a Bet from an Outcome that the Player keeps, and the multiplier that is unique to the state. As the state changes, the multiplier moves between 1, 3, 2 and 6. A states nextWon() method constructs a new state object based on the state transition table when the last bet was a winner. A states nextLost() method constructs a new state based on the state transition table when the last bet was a loser. In this case, all of the various states create a new instance of the NoWins object, resetting the multiplier to 1 and starting the sequence over again.

18.2 On Polymorphism
One very important note is that we never have to check the class of any object. This is so important, we will repeat it here. Important: We dont use isistance(). We use polymorphism and design all subclasses to have the same interface. Python relies on duck typing (if it walks like a duck and quacks like a duck, it is a duck). Additionally, Python relies on the principle that its better to seek forgiveness than ask permission. What this translates to is an approach where an objects methods are simply invoked. If the object implements the methods, then it walked like a duck and for all practical purposes actually was a duck. If the method is not implemented, the application has a serious design problem and needs to crash. We nd that isinstance() is sometimes used by beginning programmers who have failed to properly delegate processing to the subclass. Often, when a responsibility has been left out of the class hierarchy, it is allocated to the client object. The typical Pretty-Poor Polymorphism looks like the following:
class SomeClient( object def someMethod( self if isinstance(x, Special Case ): ): AClass): that should have been part of AClass

In all cases, uses of isinstance() must be examined critically. This will usually lead to refactoring the special case out of the collaborating class. There will be three changes as part of the refactoring. 1. We will move the special-case the functionality into the class being referenced by instanceof. In the above example, the special case is moved to AClass. 2. We will usually have to add default processing to the superclass of AClass so that all other sibling classes of AClass lack this special-case feature. 3. We simply call the refactored method from the client class. This refactoring leads to a class hierarchy that has the property of being polymorphic: all of the subclasses have the same interface: all objects of any class in the hierarchy are interchangable. Each object is, therefore, responsible for correct behavior. More important, a client object does not need to know which subclass the object is a member of: it simply invokes methods which are dened with a uniform interface across all subclasses. 108 Chapter 18. Player 1-3-2-6 Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

18.3 Player 1-3-2-6 Questions and Answers

Why code the state as objects? The reason for encoding the states as objects is to encapsulate the information and the behavior associated with that state. In this case, we have both the bet amount and the rules for transition to next state. While simple, these are still unique to each state. Since this is a book on design, we feel compelled to present the best design. In games like blackjack, the players state may have much more complex information or behavior. In those games, this design pattern will be very helpful. In this one case only, the design pattern appears to be over-engineered. We will use this same design pattern to model the state changes in the Craps game itself. In the case of the Craps game, there is additional information as well as behavior changes. When the state changes, bets are won or lost, bets are working or not, and outcomes are allowed or prohibited. Isnt it simpler to code the state as a number? We can just increment when we win, and reset to zero when we lose. The answer to all isnt it simpler questions is yes, but... In this case, the full answer is Yes, but what happens when you add a state or the states become more complex? This question arises frequently in OO programming. Variations on this question include Why is this an entire object? and Isnt an object over-engineering this primitive type? See Design Decision Object Identity FAQ entry on the Outcome class for additional background on object identity. Our goal in OO design is to isolate responsibility. First, and most important, we can unambiguously isolate the responsibilities for each individual state. Second, we nd that it is very common that only one state changes, or new states get added. Given these two conditions, the best object model is separate state objects. Doesnt this create a vast number of state objects? Yes. There are two usual follow-up questions: Arent all those objects a lot of memory overhead? or ...a lot of processing overhead? Since Java and Python remove unused objects, the old state denitions are cleaned up by the garbage collection thread. A few tests will demonstrate that the Java memory management is eicient and reliable. Object creation is an overhead that we can control. One common approach is to use the Singleton design pattern. In this case, this should be appropriate because we only want a single instance of each of these state classes. Note that using the Singleton design pattern doesnt change any interfaces except the initialization of the Player1326 object with the starting state. Is Polymorphism necessary? In some design patterns, like State and Command, it is essential that all subclasses have the same interface and be uniform, indistinguishable, almost anonymous instances. Because of this polymorphic property, the objects can be invoked in a completely uniform way. In our exercise, we will design a number of dierent states for the player. Each state has the same interface. The actual values for the instance variables and the actual operation implemented by

18.3. Player 1-3-2-6 Questions and Answers

109

Building Skills in Object-Oriented Design, Release 2.1.1-Python

a subclass method will be unique. Since the interfaces are uniform, however, we can trust all state objects to behave properly. There are numerous designs where polymorphism doesnt matter at all. In many cases, the anonymous uniformity of subclasses isnt relevant. When we move on to example other casino games, we will see many examples of non-polymorphic class hierarchies. This will be due to the profound dierences between the various games and their level of interaction with the players.

18.4 Player1326 State Design

class Player1326State() Player1326State is the superclass for all of the states in the 1-3-2-6 betting system.

18.4.1 Fields
player The Player1326 player who is currently in this state. This player will be used to provide the Outcome that will be used to create the Bet.

18.4.2 Constructors
__init__(self, player ) The constructor for this class saves the Player1326 which will be used to provide the Outcome on which we will bet.

18.4.3 Methods
currentBet(self ) Constructs a new Bet from the players preferred Outcome. Each subclass has a dierent multiplier used when creating this Bet. In Python, this method should return NotImplemented. This is a big debugging aid, it helps us locate subclasses which did not provide a method body. This raise statement is the functional equivalent of the Java abstract declaration. nextWon(self ) Constructs the new Player1326State instance to be used when the bet was a winner. In Python, this method should return NotImplemented. This is a big debugging aid, it helps us locate subclasses which did not provide a method body. This raise statement is the functional equivalent of the Java abstract declaration. nextLost(self ) Constructs the new Player1326State instance to be used when the bet was a loser. This method is the same for each subclass: it creates a new instance of Player1326NoWins. This dened in the superclass to assure that it is available for each subclass.

110

Chapter 18. Player 1-3-2-6 Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

18.5 Player1326 No Wins Design

class Player1326NoWins() Player1326NoWins denes the bet and state transition rules in the 1-3-2-6 betting system. When there are no wins, the base bet value of 1 is used.

18.5.1 Methods
currentBet(self ) Constructs a new Bet from the players outcome information. The bet multiplier is 1. nextWon(self ) Constructs the new Player1326OneWin instance to be used when the bet was a winner.

18.6 Player1326 One Win Design

class Player1326OneWin() Player1326OneWin denes the bet and state transition rules in the 1-3-2-6 betting system. When there is one wins, the base bet value of 3 is used.

18.6.1 Methods
currentBet(self ) Constructs a new Bet from the players outcome information. The bet multiplier is 3. nextWon(self ) Constructs the new Player1326TwoWins instance to be used when the bet was a winner.

18.7 Player1326 Two Wins Design

class Player1326TwoWins() Player1326TwoWins denes the bet and state transition rules in the 1-3-2-6 betting system. When there are two wins, the base bet value of 2 is used.

18.7.1 Methods
currentBet(self ) Constructs a new Bet from the players outcome information. The bet multiplier is 2. nextWon(self ) Constructs the new Player1326ThreeWins instance to be used when the bet was a winner.

18.5. Player1326 No Wins Design

111

Building Skills in Object-Oriented Design, Release 2.1.1-Python

18.8 Player1326 Three Wins

class Player1326ThreeWins() Player1326ThreeWins denes the bet and state transition rules in the 1-3-2-6 betting system. When there are three wins, the base bet value of 6 is used.

18.8.1 Methods
currentBet(self ) Constructs a new Bet from the players outcome information. The bet multiplier is 6. nextWon(self ) Constructs the new Player1326NoWins instance to be used when the bet was a winner. An alternative is to update the player to indicate that the player is nished playing.

18.9 Player1326 Design

class Player1326() Player1326 follows the 1-3-2-6 betting system. The player has a preferred Outcome, an even money bet like red, black, even, odd, high or low. The player also has a current betting state that determines the current bet to place, and what next state applies when the bet has won or lost.

18.9.1 Fields
outcome This is the players preferred Outcome. During construction, the Player must fetch this from the Wheel. state This is the current state of the 1-3-2-6 betting system. It will be an instance of a subclass of Player1326State. This will be one of the four states: No Wins, One Win, Two Wins or Three Wins.

18.9.2 Constructors
__init__(self ) Initializes the state and the outcome. The state is set to the initial state of an instance of Player1326NoWins. The outcome is set to some even money proposition, for example "Black".

18.9.3 Methods
placeBets(self ) Updates the Table with a bet created by the current state. This method delegates the bet creation to state objects currentBet() method. win(self, bet )

112

Chapter 18. Player 1-3-2-6 Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Parameter bet (Bet) The Bet which won Uses the superclass method to update the stake with an amount won. Uses the current state to determine what the next state will be by calling states objects nextWon() method and saving the new state in state lose(self, bet ) Parameter bet (Bet) The Bet which lost Uses the current state to determine what the next state will be. This method delegates the next state decision to state objects nextLost() method, saving the result in state.

18.10 Player 1-3-2-6 Deliverables

There are eight deliverables for this exercise. Additionally, there is an optional, more advanced design exercise in a separate section. The ve classes that make up the Player1326State class hierarchy. The Player1326 class. A Unit test for the entire Player1326State class hierarchy. Its possible to unit test each state class, but theyre so simple that its often easier to simply test the entire hierarchy. A unit test of the Player1326 class. This test should synthesize a xed list of Outcomes, Bins, and calls a Player1326 instance with various sequences of reds and blacks. There are 16 dierent sequences of four winning and losing bets. These range from four losses in a row to four wins in a row. An update to the overall Simulator that uses the Player1326.

18.11 Advanced Exercise Refactoring

Initially, each subclass of Player1326State has a unique currentBet() method. This class can be simplied slightly to have the bet multiplier coded as an instance variable, betAmount. The currentBet() method can be refactored into the superclass to use the betAmount value. This would simplify each subclass to be only a constructor that sets the betAmount multiplier to a value of 1, 3, 2 or 6. Similarly, we could have the next state after winning dened as an instance variable, nextStateWin. We can initialized this during construction. Then the nextWon() method could also be refactored into the superclass, and would return the value of the nextStateWin instance variable. The deliverable for this exercise is an alternative Player1326State class using just one class and distinct constructors that create each state object with an appropriate bet multipler and and next state win value.

18.12 Advanced Exercise Less Object Creation

The object creation for each state change can make this player rather slow. There are a few design pattens that can reduce the number of objects that need to be created.

18.10. Player 1-3-2-6 Deliverables

113

Building Skills in Object-Oriented Design, Release 2.1.1-Python

1. Global objects for the distinct state objects. While truly global objects are usually a mistake, we can justify this by claiming that were only creating objects that are shared among a few objects. 2. The Singleton design pattern. With some cleverness, this can be made transparent in Python. However, its easiest to make this explicit, by dening a formal instance() method that fetches (or creates) the one-and-only instance of the class. 3. A Factory object that produces state objects. This Factory can retain a small pool of object instances, eliminating needless object construction. The deliverables for this exercise are revisions to the Player1326State class hierarchy to implement the Factory design pattern. This will not change any of the existing unit tests, demonstrations or application programs.

18.12.1 Player1326 State Factory Design

values This is a map from a class name to an object instance. __init__(self ) Create a new mapping from the class name to object instance. There are only four objects, so this is relatively simple. get(self, name) Parameter name (String) name of one of the subclasses of Player1326State. Return the Player1326State instance for the given class name.

114

Chapter 18. Player 1-3-2-6 Class

CHAPTER

NINETEEN

CANCELLATION PLAYER CLASS

This section will describe a player who has a complex internal state that can be modeled using existing library classes.

19.1 Cancellation Player Overview

One method for tracking the lost bets is called the cancellation system or the Labouchere system. The player starts with a betting budget allocated as a series of numbers. The usual example sequence is [ 1, 2, 3, 4, 5, 6 ]. Each bet will be the sum of the rst and last numbers in the last. In this example, the end values of 1+6 leads the player to bet 7. When the player wins, the player cancels the two numbers used to make the bet. In the event that all the numbers are cancelled, the player has doubled their money, and can retire from the table happy. For each loss, however, the player adds the amount of the bet to the end of the sequence; this is a loss to be recouped. This adds the loss to the amount bet to assure that the next winning bet both recoups the most recent loss and provides a gain. Multiple winning bets will recoup multiple losses, supplemented with small gains. Example. Heres an example of the cancellation system using [ 1, 2, 3, 4, 5, 6 ] 1. Bet 1+6. A win. Cancel 1 and 6 leaving [ 2, 3, 4, 5 ] 2. Bet 2+5. A loss. Add 7 leaving [ 2, 3, 4, 5, 7 ] 3. Bet 2+7. A loss. Add 9 leaving [ 2, 3, 4, 5, 7, 9 ] 4. Bet 2+9. A win. Cancel 2 and 9 leaving [ 3, 4, 5, 7 ] 5. Next bet will be 3+7. State. The players state is a list of individual bet amounts. This list grows and shrinks; when it is empty, the player leaves the table. We can keep a List of individual bet amounts. The total bet will be the rst and last elements of this list. Wins will remove elements from the collection; losses will add elements to the collection. Since we will be accessing elements in an arbitrary order, we will want to use an ArrayList. We can dene the players state with a simple list of values.

19.2 PlayerCancellation Design

class PlayerCancellation() 115

Building Skills in Object-Oriented Design, Release 2.1.1-Python

PlayerCancellation uses the cancellation betting system. This player allocates their available budget into a sequence of bets that have an accelerating potential gain as well as recouping any losses.

19.2.1 Fields
sequence This List keeps the bet amounts; wins are removed from this list and losses are appended to this list. THe current bet is the rst value plus the last value. outcome This is the players preferred Outcome.

19.2.2 Constructors
__init__(self ) This uses the PlayerCancellation.resetSequence() method to initalize the sequence of numbers used to establish the bet amount. This also picks a suitable even money Outcome, for example, black.

19.2.3 Methods
resetSequence(self ) Puts the initial sequence of six Integer instances into the sequence variable. These Integers are built from the values 1 through 6. placeBets(self ) Creates a bet from the sum of the rst and last values of sequence and the preferred outcome. win(self, bet ) Parameter bet (Bet) The bet which won Uses the superclass method to update the stake with an amount won. It then removes the st and last element from sequence. lose(self, bet ) Parameter bet (Bet) The bet which lost Uses the superclass method to update the stake with an amount lost. It then appends the sum of the rst and list elements of sequence to the end of sequence as a new Integer value.

19.3 Cancellation Player Deliverables

There are three deliverables for this exercise. The PlayerCancellation class. A unit test of the PlayerCancellation class. This test should synthesize a xed list of Outcomes, Bin s, and calls a PlayerCancellation instance with various sequences of reds and blacks. There are 16 dierent sequences of four winning and losing bets. These range from four losses in a row to four wins in a row. This should be suicient to exercise the class and see the changes in the bet amount. An update to the overall Simulator that uses the PlayerCancellation. 116 Chapter 19. Cancellation Player Class

CHAPTER

TWENTY

FIBONACCI PLAYER CLASS

This section will describe a player who has an internal state that can be modeled using methods and simple values instead of state objects. This is a variation on the Martingale System. See Martingale Player Design for more information.

20.1 Fibonacci Player Overview

A player could use the Fibonacci Sequence to structure a series of bets in a kind of cancellation system. The Fibonacci Sequence is 1 1 2 3 5 8 13 ... At each loss, the sum of the previous two bets is used, which is the next number in the sequence. In the event of a win, we revert to the basic bet. Example. Heres an example of the Fibonacci system. 1. Bet 1. A win. 2. Bet 1. A loss. The next value in the sequence is 1. 3. Bet 1. A loss. The next value in the sequence is 2. 4. Bet 2. A loss. The next value in the sequence will be 3 5. Bet 3. In the event of a loss, the next bet is 5. Otherwise, the bet is 1. State. In order to compute the Fibonacci sequence, we need to retain the two previous bets as the players state. In the event of a win, we revert to the basic bet value of 1. In the event of a loss, we can update the two numbers to show the next step in the sequence. The players state is just these two numeric values.

117

Building Skills in Object-Oriented Design, Release 2.1.1-Python

20.2 PlayerFibonacci Design

class PlayerFibonacci() PlayerFibonacci uses the Fibonacci betting system. This player allocates their available budget into a sequence of bets that have an accelerating potential gain.

20.2.1 Fields
recent This is the most recent bet amount. Initially, this is 1. previous This is the bet amount previous to the most recent bet amount. Initially, this is zero.

20.2.2 Constructors
__init__(self ) Initialize the Fibonacci player.

20.2.3 Methods
win(self, bet ) Parameter bet (Bet) The bet which won Uses the superclass method to update the stake with an amount won. It resets recent and previous to their initial values of 1 and 0. lose(self, bet ) Parameter bet (Bet) The bet which lost Uses the superclass method to update the stake with an amount lost. This will go forwards in the sequence. It updates recent and previous as follows. next recent + previous previous recent recent next param bet The Bet which lost. type bet Bet

20.3 Fibonacci Player Deliverables

There are three deliverables for this exercise. The PlayerFibonacci class. A unit test of the PlayerFibonacci class. This test should synthesize a xed list of Outcomes, Bins, and calls a PlayerFibonacci instance with various sequences of reds and blacks. There are 16 dierent sequences of four winning and losing bets. These range from four losses in a row to four wins in a row. This should be suicient to exercise the class and see the changes in the bet amount. 118 Chapter 20. Fibonacci Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

An update to the overall Simulator that uses the PlayerFibonacci.

20.3. Fibonacci Player Deliverables

119

Building Skills in Object-Oriented Design, Release 2.1.1-Python

120

Chapter 20. Fibonacci Player Class

CHAPTER

TWENTYONE

CONCLUSION
The game of Roulette has given us an opportunity to build an application with a considerable number of classes and objects. It is comfortably large, but not complex; we have built tremendous delity to a real-world problem. Finally, this program produces moderately interesting simulation results. We note that a great many of our design decisions were not easy to make without exploring a great deal of the overall applications design. There are two ways to do this exploration: design everything in advance or design just enough but remain tolerant of our own ignorance. Experienced designers can often design an entire, complex application before building any software. The process of doing this design, however, is internally iterative. Some parts are designed in detail, with tolerance for future changes; then other parts are designed in detail and the two design elements reconciled. For new designers, we cant give enough emphasis to the importance of creating a trial design, exploring the consequences of that design, and then doing rework of that design. Too often, we have seen trial designs nalized into deliverables with no opportunity for meaningful rework. In Review of Testability, we presented one common kind of rework to support more complete testing. In Player Class, we presented another kind of rework to advance a design from a stub to a complete implementation. We also feel compelled to point out the distinction between relatively active and passive classes in this design. We had several passive classes, like Outcome, Bet and Table, which had few responsibilities beyond collecting a number of related attributes and providing simple functions. We also had several complex, active classes, like Game, BinBuilder and all of the variations on Player. These classes, typically, had fewer attributes and more complex methods. In the middle of the spectrum is the Wheel. We nd this distinction to be an enduring feature of OO design: there are things and actors; the things tend to be passive, acted upon by the actors. The overall system behavior emerges from the collaboration among all of the objects in the system; primarily but not exclusively the behavior of the active classes.

121

Building Skills in Object-Oriented Design, Release 2.1.1-Python

122

Chapter 21. Conclusion

Part III

Craps

123

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This part describes parts of the more complex game of Craps. Craps is played with dice. A player throws the dice; sometimes its an immediate win (or loss). Other times, the number on the dice become the point, and you contine throwing dice until you make your point or crap out. Craps is a game with two states and a number of state-change rules. It has a variety betting alternatives, some of which are quite complex. The chapters of this part presents the details on the game, an overview of the solution, and a series of eleven exercises to build a complete simulation of the game, with a variety of betting strategies. The exercises in this part are more advanced; unlike Roulette, we will often combine several classes into a single batch of deliverables. There are several examples of rework in this part, some of them quite extensive. This kind of rework reects three more advanced scenarios: refactoring to generalize and add features, renaming to rationalize the architecture, and refactoring to extract features. Each of these is the result of learning; they are design issues that cant easily be located or explained in advance.

125

Building Skills in Object-Oriented Design, Release 2.1.1-Python

126

CHAPTER

TWENTYTWO

CRAPS DETAILS
In the rst section we will present elements of the game of Craps. Craps is a stateful game, and well describe the state changes in some detail. We will review some of the bets available on the Craps table in some depth. Craps oers some very obscure bets, and part of this obscurity stems from the opaque names for the various bets. Creating this wide variety of bets is an interesting programming exercise, and the rst four exercise chapters will focus on this. Finally, we will describe some common betting strategies that we will simulate. The betting strategies described in Roulette will be applied to Craps bets, forcing us to examine the Roulette solution for reuse opportunities.

22.1 Craps Game

Craps centers around a pair of dice with six sides. The 36 combinations form eleven numbers from 2 to 12; additionally, if the two die are equal, this is called a hardways , because the number was made the hard way . The table has a surface marked with spaces on which players can place bets . The names of the bets are opaque, and form an obscure jargon. There are several broad classication of bets, including those which follow the sequence of dice throws that make up a complete game , propositions based on only the next throw of the dice, and hardways bets. These bets will be dened more completely, below. When the bets are placed, the dice are thrown by one of the players. This number resolves the one-roll bets that are winners or losers; it also determines the state change in the game, which may resolve any game-based winners and losers. The rst roll of the dice can be an immediate win or loss for the game, or it can establish the point for the following rounds of this game. Subsequent rolls either make the point and win the game, roll a seven and lose the game, or the game continues. When the game is won, the winning player continues to shoot; a losing player passes the dice around the table to the next player. In addition to the bets marked on the table, there are some additional bets which are not marked on the table. Some of these are called odds bets, or free odds bets, and pay o at odds that depend on the dice, not on the bet itself. An odds bet must be associated with a bet called a line bet, and are said to be behind the original bet; they can be called behind the line odds bets. The four line bets will be dened more fully, below. Note: Casino Variation There are slight variations in the bets available in dierent casinos. Well focus on a common subset of bets. The dice form a frequency distribution with values from 2 to 12. An interesting preliminary exercise is to produce a table of this distribution with a small demonstration program. This frequency distribution is central to understanding the often cryptic rules for Craps. See Creating A Dice Frequency Distribution for a small application to develop this frequency distribution.

127

Building Skills in Object-Oriented Design, Release 2.1.1-Python

The Craps game has two states: point o and point on . When a player begins a game, the point is o; the initial throw of the dice is called the come out roll . The number thrown on the dice determines the next state, and also the bets that win or lose. The rules are summarized in the following table. Table 22.1: Craps Game States State Point O; the Come Out Roll. Only Pass and Dont Pass bets allowed. Roll 2, 3, 12 Bet Resolution Craps: Pass bets lose, Dont Pass bets win. Winner: Pass bets win, Dont Pass bets lose. Next State Point O

7, 11 4, 5, 6, 8, 9, 10 Point On. Any additional bets may be placed. 2, 3, 12 11 7 Point, p Non-p number

Point O Point On the number rolled, p. Point still on Point still on Point O Point O Point still on

Nothing Nothing Loser: all bets lose. The table is cleared. Winner: point is made, Pass bets win, Dont Pass bets lose. Nothing; Come bets are activated

There is a symmetry to these rules. The most frequent number, 7, is either an immediate winner or an immediate loser. The next most frequent numbers are the six point numbers, 4, 5, 6, 8, 9, 10, which form three sets: 4 and 10 are high odds, 5 and 9 are middle odds, 6 and 8 are low odds. The relatively rare numbers (2, 3, 11 and 12) are handled specially on the rst roll only, otherwise they are ignored. Because of the small distinction made between the craps numbers (2, 3 and 12), and 11, there is an obscure C-E bet to cover both craps and eleven.

22.2 Creating A Dice Frequency Distribution

Since Python is interpreted, you can enter the following directly in the interpreter and get the frequency of each number. In Python, you will have to write and run a short program to produce the frequency distribution.
from collections import defaultdict freq= defaultdict(int) for d1 in range(6): for d2 in range(6): n= d1+d2+2 freq[n] += 1 print freq

22.3 Available Bets

There are four broad classication of bets: those which follow the sequence of dice throws that make up a complete game, those based on just the next throw of the dice (called proposition bets), the hardways bets and the unmarked bets. The game bets are called line bets, can be supplemented with additional odds bets,

128

Chapter 22. Craps Details

Building Skills in Object-Oriented Design, Release 2.1.1-Python

placed behind the line bet. There are actually two sets of game bets based on the pass line and the come line. The following illustration shows the left half of a typical craps table layout. On the left are the game bets. On the right (in the center of the table) are the single-roll propositions.

Figure 22.1: Craps Table Layout Game Bets. The Game bets have two states. There is a set of bets allowed for the rst roll, called the come out roll. If a point is established a number of additional bets become available. These are the pass line bets. In addition to the basic pass line bets, there are also come line bets, which are similar to game bets. Come Out Roll. When the point is o, the player will be making the come out roll, and only certain bets are available. Once the point is on, all bets are available. The bets which are allowed on the come out roll are the Pass Line and Dont Pass Line bets. These bets follow the action of the players game. There are several possible outcomes for these two bets. Player throws craps (2, 3, or 12). A Pass Line bet loses. A Dont Pass Line bet wins even money on 2 or 3; however, on 12, it returns the bet. This last case is called a push ; this rule is noted on the table by the bar 12 legend in the Dont Pass area. Player throws 7 or 11. A Pass Line bet wins even money. A Dont Pass Line bet loses. Player throws a point number. The point is now on. A large white token, marked on is placed in one of the six numbered boxes to announce the point. The player can now place an additional odds bet behind the line. Placing a bet behind the Pass Line is called buying odds on the point. Placing a bet behind the Dont Pass Line is called laying odds against the point. Point Rolls. Once the point is on, there are three possible outcomes. The player makes the point before rolling 7. A Pass Line bet wins even money. A Dont Pass Line bet loses. The point determines the odds used to pay the behind the Pass Line odds bet. The player throws 7 before making the point. A Pass Line bet loses. A Dont Pass Line bet pays even money. The point determines the odds used to pay the behind the Dont Pass Line odds bet. The player throws a number other than 7 and the point: the game continues. Come Line Bets. Once the point is on, the bets labeled Come and Dont Come are allowed. These bets eectively form a game parallel to the main game, using a come point instead of the main game point. Instead of dropping the large on token in a numbered square on the layout, the house will move the 22.3. Available Bets 129

Building Skills in Object-Oriented Design, Release 2.1.1-Python

individual Come and Dont Come bets from their starting line into the numbered boxes (4, 5, 6, 8, 9, 10) to show which bets are placed on each point number. There are several possible outcomes for these two bets. Player throws craps (2, 3, or 12). A Come Line bet loses. A Dont Come bet wins even money on 2 or 3; it is a push on 12. Established come point bets in the numbered squares remain. Player throws 11. A Come Line bet wins even money. A Dont Come bet loses. Established come point bets in the numbered squares remain. Player throws a point number (4, 5, 6, 8, 9, or 10). Any come bets in that numbered box are winners the come point was made and the point determines the odds used to pay the behind the line odds. (Note that if this is the point for the main game, there will be no bets in the box, instead the large on token will be sitting there.) New bets are moved from the Come Line and Dont Come Line to the numbered box, indicating that this bet is waiting on that point number to be made. Additional behind the line odds bets can now be placed. If the main games point was made, the on token will be removed (and ipped over to say o ), and the next roll will be a come out roll for the main game. These bets (with the exception of the free odds behind a Come Line bet) are still working bets. The player throws 7 before making the main point. A Come bet loses. A Dont Come bet pays even money. The bets in the numbered boxes are all losers. Propositions. There are also a number of one-roll propositions, including the individual numbers 2, 3, 7, 11, 12; also eld (2, 3, 4, 9, 10, 11 or 12), any craps (2, 3 or 12), the horn (2, 3, 11 or 12), and a hop bet. These bets have no minimum, allowing a player to cover a number of them inexpensively. Hardways. There are four hardways bets. These are bets that 4, 6, 8 or 10 will come up on two equal die (the hard way) before they come up on two unequal die (the easy way) or before a 7 is rolled. Hard 6, for example, is a bet that the pair (3,3) will be rolled before any of the other four combinations (1,5), (2,4), (5,1), (4,2) that total 6, or any of the 6 combinations that total 7. These bets can only be placed when the point is on, but they are neither single-roll propositions, nor are they tied to the game point. Unmarked Bets. There are additional unmarked bets. One is to make a place bet on a number; these bets have a dierent set of odds than the odds bets behind a line bet. Other unmarked bets are to buy a number or lay a number, both of which involve paying a commission on the bet. Typically, only the six and eight are bought, and they can only be bought when they are not the point. For programmers new to casino gambling, see Odds and Payouts for more information on odds payments for the various kinds of bets.

130

Chapter 22. Craps Details

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Odds and Payouts Not all of the Craps outcomes are equal probability. Some of the bets pay xed odds, dened as part of the bet. Other bets pay odds that depend on the point established. The basic Pass Line, Dont Pass, Come Line and Dont Come bets are even money bets. A $5 bet wins an additional $5. The odds of winning a Pass Line bet is 49.3%. The 1:1 payout does not reect the actual probability of winning. The points of 4 and 10 have odds of 3 in 36 to win and 6 in 36 to lose. The points of 5 and 9 have odds of 4 in 36 to win and 6 in 36 to lose. The points of 6 and 8 have odds of 5 in 36 to win and 6 in 36 to lose. The odds, therefore are 2:1, 3:2 and 6:5 for these behind-the-line odds bets, respectively. The combination of a Pass Line bet plus odds behind the line provides the narrowest house edge. For example, a bet on the Pass Line has an expected return of -1.414%. If we place double odds behind the line with an expected return of 0, the net expected return for the combination is -0.471%. The various proposition bets have odds stated on the table. Note that these rarely reect the actual odds of winning. For example, a bet on 12 is a 1/36 probability, but only pays 30:1. There are 1/36 ways to win a hard six or hard eight bet, 10/36 ways to loose and the remaining 25 outcomes are indeterminate. While a hard 6 or hard 8 should pay 10:1, the actual payment is 9:1. Similarly for a hard 4 or hard 10: they should pay 8:1, the actual payment is 7:1.

22.4 Some Betting Strategies

All of the Roulette betting strategies apply to Craps. However, the additional complication of Craps is the ability to have multiple working bets by placing additional bets during the game. The most important of these additional bets are the behind the line odds bets. Since these are paid at proper odds, they are the most desirable bets on the layout. The player, therefore, should place a Line bet, followed by a behind the line odds bet. The other commonly used additional bets are the Come (or Dont Come) bets. These increase the number of working bets. Since the additional odds bets for these pay proper odds, they are also highly desirable. A player can place a Come bet, followed by a behind the line odds bet on the specic number rolled. Beyond this, there are the various proposition and hardways bets that can be placed in addition to other bets. While these propositions have odds that are incorrect, some players will count the number of throws in game, and place a 7 bet if the game lasts more than six throws. This is a form of betting against ones self: there is a Pass Line bet that the game will be won, diluted by a 7 bet that the game will be lost. While the common rationale is that the second bet protects against a loss, it also reduces the potential win. We can simulate this kind of betting and examine the potential outcomes.

22.5 Wrong Betting

Craps has two forms of odds betting: right and wrong. The Pass Line and Come Line, as well as buying odds, are all right. The Dont Pass, Dont Come and Laying Odds are all wrong. Sometimes wrong betting is also called the dark side. Right betting involves odds that provide a large reward for a small wager. The odds on a Pass Line bet for the number 4 are 2:1, you will win double your bet if the point is made. You put up $10 at risk to make $20. Wrong betting involves odds that provide a small reward for a large wager. The odds on a Dont Pass Line bet for the number 4 are described as laid at 2:1 in eect it is a 1:2 bet, you will win half of what you bet if the point is missed. You put $20 at risk to make $10.

22.4. Some Betting Strategies

131

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This distinction can be ignored by novice programmers. None of our simulated players are wrong bettors, since this involves putting a large amount at risk for a small payout, which violates one of the basic rules of gambling.

132

Chapter 22. Craps Details

CHAPTER

TWENTYTHREE

CRAPS SOLUTION OVERVIEW

We will present a survey of the classes gleaned from the general problem statement in Problem Statement as well as the problem details in Craps Details. This survey is drawn from a quick overview of the key nouns in these sections. Given this survey of the candidate classes, we will then do a walkthrough to rene the denitions, and assure ourselves that we have a reasonable architecture. We will make some changes to the preliminary class list, revising and expanding on our survey. We will also include a number of questions and answers about this preliminary design information. This should help clarify the design presentation and set the stage for the various development exercises in the chapters that follow.

23.1 Preliminary Survey of Classes

We have divided up the responsibilities to provide a starting point for the development eort. The central principle behind the allocation of responsibility is encapsulation . In reading the background information and the problem statement, we noticed a number of nouns that seemed to be important parts of the Craps game. Dice Bet Table Point Proposition Number Odds Player House The following table summarizes some of the classes and responsibilities that we can identify from the problem statement. This is not the complete list of classes we need to build. As we work through the exercises, well discover additional classes and rework some of these classes more than once. We also have a legacy of classes available from the Roulette solution. We would like to build on this infrastructure as much as possible.

133

Building Skills in Object-Oriented Design, Release 2.1.1-Python

23.2 Preliminary Class Structure

Outcome Responsibilities A name for a particular betting opportunity. Most outcomes have xed odds, but the behind the line odds bets have odds that depend on a point value. Collaborators Collected by Table into the available bets for the Player; used by Game to compute the amount won from the amount that was bet. Dice Responsibilities Selects any winning propositions as well as next state of the game. Collaborators Used by the overall Game to get a next set of winning Outcome s, as well as change the state of the Game . Table Responsibilities A collection of bets placed on Outcome s by a Player . This isolates the set of possible bets and the management of the amounts currently at risk on each bet. This also serves as the interface between the Player and the other elements of the game. Collaborators Collects the Outcome s; used by Player to place a bet amount on a specic Outcome ; used by Game to compute the amount won from the amount that was bet. Player Responsibilities Places bets on Outcome s, updates the stake with amounts won and lost. This is the most important responsibility in the application, since we expect to update the algorithms this class uses to place dierent kinds of bets. Clearly, we need to cleanly encapsulate the Player , so that changes to this class have no ripple eect in other classes of the application. Collaborators Uses Table to place Bet s on preferred Outcome s; used by Game to record wins and losses. Game Responsibilities Runs the game: gets bets from Player , throws the Dice , updates the state of the game, collects losing bets, pays winning bets. This encapsulates the basic sequence of play into a single class. The overall statistical analysis is based on playing a nite number of games and seeing the nal value of the Player s stake. Collaborators Uses Dice , Table , Outcome , Player .

23.3 A Walkthrough of Craps

Building Skills in Object-Oriented Design, Release 2.1.1-Python

1. Our preliminary note was that this class Runs the game. The responsibilities section has a summary of ve steps involved in running the game. 2. The rst step is gets bets from :class:Player. Find the Player card. 3. Does a Player collaborate with a Game to place bets? Note that the game state inuences the allowed bets. Does Game collaborate with Player to provide the state information? If not, add this information to one or both cards. 4. The Games second step is to throw the Dice. Is this collaboration on the Dice card? 5. The Games third step is to update the state of the game. While the state appears to be internal to the Game, requiring no collaboration, we note that the Player needs to know the state, and therefore should collaborate with Game. Be sure this collaboration is documented. 6. The Games fourth and fth steps are to pay winning bets and collect losing bets. Does the Game collaborate with the Table to get the working bets? If not, update the collaborations. Something well need to consider is the complex relationship between the dice, the number rolled on the dice, the game state and the various bets. In Roulette, the wheel picked a random Bin which had a simple list of winning bets; all other bets were losers. In Craps, however, we nd that we have game bets that are based on changes to the game state, not simply the number on the dice. The random outcome is used to resolve one-roll proposition bets, resolve hardways bets, change the game state, and resolve game bets. We also note that the house moves Come Line (and Dont Come) bets from the Come Line to the numbered spaces. In eect, the bet is changed from one Outcome to another Outcome . This means that a Bet has a kind of state change, in addition to the Game s state change and any possible Player state change. Important: Stateful Objects To continue this rant, we nd that most interesting IT applications involve stateful objects. Everything that has a state or status, or can be updated, is stateful. Often, designers overlook these state changes, describing them as simple updates. However, state changes are almost universally accompanied by rules that determine legal changes, events that precipitate changes, and actions that accompany a state change. Belittling stateful objects causes designers to overlook these additional details. The consequence of ignoring state is software that performs invalid or unexpected state transitions. These kinds of bugs are often insidious, and diicult to debug. A walkthrough will give you an overview of the interactions among the objects in the working application. You may uncover additional design ideas from this walkthrough. The most important outcome of the walkthrough is a clear sense of the responsibilities and the collaborations required to create the necessary application behavior.

23.4 Craps Solution Questions and Answers

Why is Outcome a separate class? Each object that is an instance of Outcome is merely a number from 2 to 12. Here we have complex interdependency between the dice, the game states, the bets and outcomes. An outcome has dierent meanings in dierent game states: sometimes a 7 is an immediate winner, other times it as an immediate loser. Clearly, we need to isolate these various rules into separate objects to be sure that we have captured them accurately without any confusion, gaps or conicts. We can foresee three general kinds of Outcome s: the propositions that are resolved by a single throw of the dice, the hardways that are resolved periodically, and the game bets which are resolved when a point is made or missed. Some of the outcomes are only available in certain game states.

23.4. Craps Solution Questions and Answers

135

Building Skills in Object-Oriented Design, Release 2.1.1-Python

The alternative is deeply nested if-statements. Multiple objects introduce some additional details in the form of class declarations, but objects have the advantage of clearly isolating responsibilities, making us more condent that our design will work properly. If-statements only conate all of the various conditions into a tangle that includes a huge risk of missing an important and rare condition. See the discussion under Design Decision Object Identity for more discussion on object identity and why each Outcome is a separate object. What is the dierence between Dice and Wheel? Dont they both represent simple collections with random selection? Perhaps. At the present time, the distinction appears to be in the initialization of the two collections of Bins of the Wheel or Throws of the Dice. Well revisit the question in some depth in Soapbox on Overengineering. Generally, we are slow to merge classes together without evidence that they are really the same thing. In this case, they appear very similar, so we will note the similarities and dierences as we work through the design details. There is a ne line between putting too many things together and splitting too many things apart. Generally, the mistake we see most often is putting too many things together, and resolving the dierences through complex if-statements and other hidden processing logic.

136

Chapter 23. Craps Solution Overview

CHAPTER

TWENTYFOUR

OUTCOME CLASS
This chapter will examine the Outcome class, and its suitability for the game of Craps. Well present some additional code samples to show a way to handle the dierent kinds of constuctors that this class will need.

24.1 Outcome Overview

For Craps, we have to be careful to disentangle the random events produced by the Dice and the outcomes on which a Player creates a Bet. In Roulette, this relationship was simple: a Bin was a container of Outcomes; a players Bet referenced one of these Outcomes. In Craps, however, we have one-roll outcomes, hardways outcomes, plus outcomes that arent resolved until the end of the game. Keeping this in mind, well defer details of dice and game problem until later, and focus on just the bet Outcomes rst. Varietals. There is a rich variety of bet Outcomes. Well itemize them so that we can conrm the responsibilities for this class. The Line Bets: these are the Pass Line, Dont Pass Line, Come Line, Dont Come Line. These outcomes have xed odds of 1:1. The four Hardways bets: 4, 6, 8 and 10. These outcomes also have xed odds which depend on the number. The various one-roll propositions. All of these have xed odds. These are most like the original Outcome used for Roulette. The six Come-point bets (4, 5, 6, 8, 9 and 10). Each of these has xed odds. Also, the initial line bet is moved to a point number from the Come Line bet, based on the number shown on the dice. Well examine these is some detail, below. The Odds bets placed behind the Line bets. These have odds based on the point, not the outcome itself. Well have to look at these more closely, also. The six Placed Numbers have odds are based on the number placed. These outcomes have xed odds. Once the bet is placed, these bets are resolved when the number is rolled or when a game losing seven is rolled. The Buy and Lay bets require a commission payment, called a vigorish, when the bet is placed. The outcomes have simple, xed odds. As with the placed number bets, these bets are resolved when the number is rolled or when a game losing seven is rolled. Looking more closely at the bets with payout odds that depend on the point rolled, we note that the Come Line (and Dont Come) odds bets are moved to a specic number when that point is established. For example, the player places a Come bet, the dice roll is a 4; the come bet is moved into the box labeled 4. Any additional odds bet is placed in this box, also.

137

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This leaves us with the Pass Line and Dont Pass Line odds bet, which also depend on the point rolled. In this case, the bets are eectively moved to a specic numbered box. In a casino, a large, white, on token in placed in the box. The eect is that same as if the house had moved all the Pass Line bets. The odds bet, while physically behind the Pass Line, is eectively in the box identied by the point. Again, the bet is moved from the 1:1 Pass Line to one of the six numbered boxes; any odds bet that is added will be on a single Outcome with xed odds. In this case, the existing Outcome class still serves many of our needs. Looking forward, we will have to rework Bet to provide a method that will change to a dierent Outcome. This will move a line bets to one of the six numbered point boxes. Additional Features. There are two additional responsibilities that we will need in the Outcome class: more complex odds and a house commission. In Roulette, all odds were stated as n :1, and our winAmount() depended on that. In craps, many of the odds have non-unit denominators. Example odds include 6:5, 3:2, 7:6, 9:5. In a casino, the bets are multiples of $5, $6 or $10 to accomodate the fractions. Complex Odds. In our simulation, we are faced with two choices for managing these more complex odds: exact fractions or approximate oating-point values. Since exact rational fractions are not part of the Java class library, well use simple double values instead of int values for the players stake and the amount won or lost. An interesting additional feature is to use rational numbers for the stakes and amounts. There are a number of rational number classes available on the Internet, any of which would be suitable for this problem. We will leave it to the interested student to replace the double values with Rational values for the players stake and the amounts won and lost with each bet. In Java, we will have to replace many instances of int with double. This is an unpleasant change, but a consequence of starting out with a built-in primitive type. The only eective escape from this kind of change is to use a class in all cases instead of a primitive type. This would allow us to use dierent subclasses as our design evolved. If, for example, we use java.lang.Number, we can switch to any subclass without a complex or sweeping change. The down-side of using a class in Python is that the built-in expression syntax only applies to primitive types. With primitive types we can write a + b . With the language classes like Integer, we have to write something like new Integer( a.intValue() + b.intValue() ). Withbclasses like BigInteger, we can write something like a.add( b ) . In Python, we dont have static compile-time type checking. We can easily replace an integer result with a double result with no impact throughout the application. Commission Payments. The second extension we have to consider is for the two bets which have a commission when they are created: buy bets and lay bets. The buy bet involves an extra 5% placed with the bet: the player puts down $21, a $20 bet and a $1 commission. A lay bet, which is a wrong bet, involves a risk of a large amount of money against a small win, and the commission is based on the potential winning. For a 2:3 wrong bet, the commission is 5% of the outcome; the player puts down $31 to win $20 if the point is not made. In both buy and lay cases, the Player sees a price to create a bet of a given amount. Indeed, this generalizes nicely to all other bets. Most bets are simple and the price is the amount of the bet. For buy bets, however, the price is 5% of the amount of the bet; for lay bets, the price is 5% of the possible payout. The open question is the proper allocation of responsibility for this price. Is the price related to the Outcome or the Bet? When we look at the buy and lay bets, we see that they are based on existing point number Outcome s and share the same odds. However, there are three very dierent ways create a Bet on one of these point number Outcome s: a bet on the Pass Line (or Dont Pass Line), a bet on the Come Line (or Dont Come Line), and a buy (or lay) bet on the number. When we bet via the Pass Line or Come Line, the Line bet was moved to the point number, and the odds bet follows the Line bet. For this reason, the price is a feature of creating the Bet. Therefore, well consider the commission as the price

138

Chapter 24. Outcome Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

of creating a bet and defer it to the Bet class design. We observe that the slight error in the Line bet odds is the houses edge on the Line bet. When we put an odds bet behind the line, the more correct odds dilutes this edge. When we buy a number, on the other hand, the odds are correct and the house takes the commission directly.

24.2 Overloaded Methods

Well be extending the denition of Outcome. We dont want to disturb the simple case of Roulette when we add the more complex case of Craps. Well do this with an overloaded constructor; a single method name that two dierent sets of parameters. In Python, we use optional parameters to provide default values. We declare the default value as part of the method denition. This works well when we are initializing values to any of the immutable object classes. It doesnt, however, work well for mutable lists and mappings; these must be handled specially. The following example shows a simple required argument, followed by an optional argument.
class SomeClass: def __init__( self, reqArg, optArg=1 ): initialize using reqArg and optArg

Note that Python mutable types (lists, sets, maps) should not be provided as default values for an initializer. This is because one initialization object is created when the class is dened, and all instances will share this singleton initializer object. To avoid this undesirable sharing of an instance of the default value, we have to do the following.
class SomeClass: def __init__( self, aList=None ): if aList: initialize using aList else: default initialization using a fresh []

24.3 Outcome Rework

Outcome contains a single outcome on which a bet can be placed. In Roulette, each spin of the wheel has a number of Outcomes. For example, the 1 bin has the following winning Outcomes: 1, Red, Odd, Low, Column 1, Dozen 1-12, Split 1-2, Split 1-4, Street 1-2-3, Corner 1-2-4-5, Five Bet, Line 1-2-3-4-5-6, 00-0-1-2-3, Dozen 1, Low and Column 1 .

24.3.1 Fields
name Holds the name of the Outcome. Examples include "1", "Red", "Pass Line". numerator Holds the numerator portion of odds for this Outcome. This is the multiplier in the odds fraction. For right bets in Craps, this is larger than the denominator. For wrong bets, it is smaller than the denominator. denominator Holds the denominator portion of odds for this Outcome. For Roulette, this is always 1. 24.2. Overloaded Methods 139

Building Skills in Object-Oriented Design, Release 2.1.1-Python

24.3.2 Constructors
__init__(name, numerator, denominator=1) Parameters numerator (integer) the payout odds numerator denominator (integer) the payout odds denominator Sets the name and odds from the parameters. Example 1: 6:5 is a right bet, the player will win 6 for each 5 that is bet. Example 2: 2:3 is a wrong bet, they player will win 2 for each 3 that is bet.

24.3.3 Methods
winAmount(self, amount ) Parameter amount (integer) amount of the bet Returns the product of this Outcomes odds numerator by the given amount, divided by the odds denominator. __str__(self ) An easy-to-read String output method is also very handy. This should return a String representation of the name and the odds. A form that looks like 1-2 Split (17:1) works nicely.

24.4 Outcome Deliverables

There are three deliverables for this exercise. The revised Outcome class that handles more complex odds and returns type double from winAmount(). A class which performs a unit test of the Outcome class. The unit test should create a couple instances of Outcome, and establish that the winAmount() method works correctly. A revision to each subclass of Player to correctly implement the revised result from winAmount(). Currently, there are six subclasses of Player : Passenger57, SevenReds, PlayerRandom, Player1326, PlayerCancellation, and PlayerFibonacci.

24.5 Advanced Exercise

For the more advanced student, locate or develop a class to work with rational numbers, represented as an integer numerator and denominator. Use this for all stakes, odds and bet payos, replacing any double variables.

140

Chapter 24. Outcome Class

CHAPTER

TWENTYFIVE

THROW CLASS
In Craps, a throw of the dice may change the state of the game. This close relationship between the Throw and CrapsGame leads to another chicken-and-egg design problem. Well design Throw in detail, but provide a rough stub for CrapsGame. Additionally, this chapter will introduce the subtle issue of over-engineering when doing design work. Well revist these issues a number of times in order to provde examples of good design and the issues that can lead to poor design.

25.1 Throw Overview

The pair of dice can throw a total of 36 unique combinations. These are summarized into fteen distinct outcomes: the eleven numbers from 2 to 12, plus the four hardways variations for 4, 6, 8 and 10. In Roulette, the randomized positions on the wheel were called Bin s and each one had a very simple collection of winning Outcome s. In Craps, however, the randomized throws of the dice serve three purposes: they resolve simple one-roll proposition bets, they may resolve hardways bets, and they change the game state (which may resolve game bets). From this we can allocate three responsibilities. Well look at each of these responsibilities individually. One-Throw Propositions. A Throw of the dice includes a collection of proposition Outcomes which are immediate winners. This collection will be some combination of 2, 3, 7, 11, 12, Field, Any Craps, or Horn. For completeness, we note that each throw could also contain one of the 21 hop-bet Outcomes; however, well ignore the hop bets. Multi-Throw Propositions. A Throw of the dice may resolve hardways bets (as well as place bets and buy bets). There are three possible conditions for a given throw: some bets may be winners, some bets may be losers, and some bets may remain unresolved. This tells us that a Throw may be more than a simple collection of winning Outcomes. A Throw must also contain a list of losing Outcomes. For example, any of the two easy 8 rolls (6-2 or 5-3) would contain winning Outcomes for the place-8 bet and buy-8 bet, as well as a losing Outcome for a hardways-8 bet. The hard 8 roll (4-4), however, would contain winning Outcome s for the place-8 bet, buy-8 bet, and hardways-8 bet Game State Change. Most importantly, a Throw of the dice can lead to a state change of the Game. This may resolve game-level bets. From the Craps Game, we see that the state changes depend on both the Game state plus the kind of Throw. The rules identify the following species of Throw. Craps. These are throws of 2, 3 or 12. On a come-out roll, this is an immediate loss. On any other roll, this is ignored. There are 4 of these throws. Natural. This is a throw of 7. On a come-out roll, this is an immediate win. On any other roll, this is an immediate loss and a change of state. There are 6 of these throws.

141

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Eleven. This is a throw of 11. On a come-out roll, this is an immediate win. On any other roll, this is ignored. There are 2 of these throws. Point. This is a throw of 4, 5, 6, 8, 9, or 10. On a come-out roll, this estanblishes the point, and changes the game state. On any other roll, this is is compared against the established point: if it matches, this is a win and a change of game state. Otherwise, it doesnt match and no game state change occurs. There are a total of 24 of these throws, with actual frequencies between three and ve. The state change can be implemented by dening methods in Game that match the varieties of Throw. We can imagine that the design for Game will have four methods: craps(), natural(), eleven(), and point(). Each kind of Throw will call the matching method of Game, leading to state changes, and possibly game bet resolution. The game state changes lead us to design a hierarchy of Throw classes to enumerate the four basic kinds of throws. We can then initialize a Dice object with 36 Throw objects, each of the appropriate subclass. When all of the subclasses have an identical interface, this embodies the principle of polymorphism. For additional information, see On Polymorphism. In looking around, we have a potential naming problem: both a wheels Bin and the dices Throw are somehow instances of a common abstraction. Looking forward, we may wind up wrestling with a deck of cards trying to invent a common nomenclature for all of these randomizers. They create random events, and this leads us to a possible superclass for Bin and Throw: RandomEvent. Currently, we cant identify any features that we can refactor up into the superclass. Rather than overengineer this, well hold o on complicating the design until we nd something else that is common between our sources of random events. See the Soapbox on Over-Engineering as a continuation of this rant. Soapbox on Over-Engineering Some of the horror stories of failed OO design projects seem to have over-engineering as the root cause. Over-engineering often includes the creation of superclasses and interfaces to embody potentially useful features. We suggest examining all engineering eorts by asking does it solve a real problem? Focus on potential value tends to dilute the creation of real value. We note that software is developed to solve a particular problem. If a design consideration does not help solve the problem as presented, we have to classify the design as potential over-engineering. The most common response to this classication is that we will tend to focus too narrowly and build a point solution; the presumption is that point solutions are not as adaptable as something more broadly focused. Our experience is that when the object design focuses on tangible real-world objects, and models those objects with a high degree of delity, these objects are very adaptable; they can be applied to solve a number of real-world problems. When the object design includes a number of irrelevant considerations, it takes too much eort to solve the initial problem, diluting the value of the software. Additionally, some over-engineering is the result of conating a number of features into a single class. We prefer to add features slowly so as to clearly separate the various concerns. As an example, it is very easy to conate Bet and Outcome into a single class, losing site of the fact that bet amounts can change, but the denition of the outcome itself is invariant.

25.2 Throw Design

class Throw() Throw is the superclass for the various throws of the dice. Each subclass is a dierent grouping of the numbers, based on the rules for Craps.

142

Chapter 25. Throw Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

25.2.1 Fields
outcomes A Set of one-roll Outcomes that win with this throw. d1 One of the two die values, from 1 to 6. d2 The other of the two die values, from 1 to 6.

25.2.2 Constructors
__init__(self, d1, d2, * outcomes) Creates this throw, and associates the given Set of Outcomes that are winning propositions. Parameters d1 The value of one die d2 The value of the other die outcomes The various outcomes for this Throw

25.2.3 Methods
hard(self ) Returns true if d1 is equal to d2. This helps determine if hardways bets have been won or lost. updateGame(self, game) Parameter game (CrapsGame) the Game to be updated based on this throw. Calls one of the Game state change methods: craps(), natural(), eleven(), point(). This may change the game state and resolve bets. __str__(self ) An easy-to-read String output method is also very handy. This should return a String representation of the dice. A form that looks like 1,2 works nicely.

25.3 Natural Throw Design

class NaturalThrow() Natural Throw is a subclass of Throw for the natural number, 7.

25.3.1 Constructors
__init__(self, d1, d2 ) Parameters d1 The value of one die d2 The value of the other die

25.3. Natural Throw Design

143

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Creates this throw. The constraint is that d1 + d2 = 7. If the constraint is not satised, simply raise an exception. This uses the superclass constructor to add appropriate Outcomes for a throw of 7.

25.3.2 Methods
hard(self ) A natural 7 is odd, and can never be made the hard way. This method always returns false. updateGame(self, game) Parameter game (CrapsGame) the Game to be updated based on this throw. Calls the natural() method of a game Game. This may change the game state and resolve bets.

25.4 Craps Throw Design

class CrapsThrow() Craps Throw is a subclass of Throw for the craps numbers 2, 3 and 12.

25.4.1 Constructors
__init__(self, d1, d2 ) Parameters d1 The value of one die d2 The value of the other die Creates this throw. The constraint is that d1 + d2 {2, 3, 12}. If the constraint is not satised, simply raise an exception. This uses the superclass constructor to add appropriate Outcomes for a throw of craps.

25.4.2 Methods
hard(self ) The craps numbers are never part of hardways bets. This method always returns false. updateGame(self, game) Parameter game (CrapsGame) the Game to be updated based on this throw. Calls the craps() method of a game Game. This may change the game state and resolve bets.

25.5 Eleven Throw Design

class ElevenThrow() Eleven Throw is a subclass of Throw for the number, 11. This is special because 11 has one eect on a come-out roll and a dierent eect on point rolls. 144 Chapter 25. Throw Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

25.5.1 Constructors
__init__(self, d1, d2 ) Parameters d1 The value of one die d2 The value of the other die Creates this throw. The constraint is that d1 + d2 = 11. If the constraint is not satised, simply raise an exception. This uses the superclass constructor to add appropriate Outcomes for a throw of 11.

25.5.2 Methods
hard(self ) Eleven is odd and never part of hardways bets. This method always returns false. updateGame(self, game) Parameter game (CrapsGame) the Game to be updated based on this throw. Calls the eleven() method of a game Game. This may change the game state and resolve bets.

25.6 Point Throw Design

class PointThrow() Point Throw is a subclass of Throw for the point numbers 4, 5, 6, 8, 9 or 10.

25.6.1 Constructors
__init__(self, d1, d2 ) Parameters d1 The value of one die d2 The value of the other die Creates this throw. The constraint is that d1 + d2 {4, 5, 6, 8, 9, 10}. If the constraint is not satised, simply raise an exception. This uses the superclass constructor to add appropriate Outcomes for a throw of craps.

25.6.2 Methods
hard(self ) Eleven is odd and never part of hardways bets. This method always returns false. Returns true if d1 is equal to d2. This helps determine if hardways bets have been won or lost. updateGame(self, game) Parameter game (CrapsGame) the Game to be updated based on this throw.

25.6. Point Throw Design

145

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Calls the point() method of a game Game. This may change the game state and resolve bets.

25.7 Craps Game Design

class CrapsGame() CrapsGame is a preliminary design for the game of Craps. This initial design contains the interface used by the Throw class hierarchy to implement game state changes.

25.7.1 Fields
point The current point. This will be replaced by a proper State design pattern.

25.7.2 Constructors
__init__(self ) Creates this Game. This will be replaced by a constructor that uses Dice and CrapsTable.

25.7.3 Methods
craps(self ) Resolves all current 1-roll bets. If the point is zero, this was a come out roll: Pass Line bets are an immediate loss, Dont Pass Line bets are an immediate win. If the point is non-zero, Come Line bets are an immediate loss; Dont Come Line bets are an immediate win. The state doesnt change. A future version will delegate responsibility to the craps() method of a current state object. natural(self ) Resolves all current 1-roll bets. If the point is zero, this was a come out roll: Pass Line bets are an immediate win; Dont Pass Line bets are an immediate loss. If the point is non-zero, Come Line bets are an immediate win; Dont Come bets are an immediate loss; the point is also reset to zero because the game is over. Also, hardways bets are all losses. A future version will delegate responsibility to the natural() method of a current state object. eleven(self ) Resolves all current 1-roll bets. If the point is zero, this is a come out roll: Pass Line bets are an immediate win; Dont Pass Line bets are an immediate loss.

146

Chapter 25. Throw Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

If the point is non-zero, Come Line bets are an immediate win; Dont Come bets are an immediate loss. The game state doesnt change. A future version will delegate responsibility to the eleven() method of a current state object. point(self, point ) Parameter point (integer) The point value to set. Resolves all current 1-roll bets. If the point was zero, this is a come out roll, and the value of the dice establishes the point. If the point was non-zero and this throw matches the point the game is over: Pass Line bets and associated odds bets are winners; Dont Pass bets and associated odds bets are losers; the point is reset to zero. Finally, if the point is non-zero and this throw does not match the point, the state doesnt change; however, Come point and Dont come point bets may be resolved. Additionally, hardways bets may be resolved. A future version will delegate responsibility to the current states point() method to advance the game state. __str__(self ) An easy-to-read String output method is also very handy. This should return a String representation of the current state. The stub version of this class has no internal state object. This class can simply return a string representation of the point; and the string "Point Off" when point is zero.

25.8 Throw Deliverables

There are eleven deliverables for this exercise. A stub class for CrapsGame with the various methods invoked by the throws. The design information includes details on bet resolution that doesnt need to be fully implemented at the present time. For this stub class, the change to the point variable is required for unit testing. The other information should be captured as comments and output statements that help conrm the correct behavior of the game. The Throw superclass, and the four subclasses: PointThrow. CrapsThrow, NaturalThrow, ElventThrow,

Five classes which perform unit tests on the various classes of the Throw class hierarchy.

25.8. Throw Deliverables

147

Building Skills in Object-Oriented Design, Release 2.1.1-Python

148

Chapter 25. Throw Class

CHAPTER

TWENTYSIX

DICE CLASS
Unlike Roulette, where a single Bin could be identied by the number in the bin, dice use a pair of numbers. In this chapter, we design Dice, as well as designing an inner class that is used only to make a single key out of a composite object.

26.1 Dice Overview

The dice have two responsibilities: they are a container for the Throw s and they pick one of the Throws at random. We nd that we have a potential naming problem: both a Wheel and the Dice are somehow instances of a common abstraction. Looking forward, we may wind up wrestling with a deck of cards trying to invent a common nomenclature for the classes. They create random events, and this leads us to a possible superclass: Randomizer. Rather than over-engineer this, well hold o on adding this design element until we nd something else that is common among them. See Soapbox on Over-Engineering for some additional opinions on this subject. Container. Since the Dice have 36 possible Throws, it is a collection. We can review our survey of the collections in Design Decision Choosing A Collection for some guidance here. In this case, we note that the choice of Throw can be selected by a random numeric index. For Python programmers, this makes the a list very appealing. After selection a collection type, we must then deciding how to index each Throw in the Dice collection. Recall that in Roulette, we had 38 numbers: 1 to 36, plus 0 and 00. By using 37 for the index of the Bin that contained 00, we had a simple integer index for each Bin. For Craps it seems better to use a two-part index with the values of two independent dice. Index Choices. In this case, we have two choices for computing the index into the collection, We can rethink our use of a simple sequential structure. If we use a Map, we can use an object representing the pair of numbers as an index instead of a single int value. We have to compute a unique index position from the two dice values. Decision Forces. There are a number of considerations to choosing between these two representations. 1. If we create a new class to contain each unique pair of integers, we can then use that object to be the index for a Map. The Map associates this pair of numbers object with its Throw. In Python, a tuple does this job nicely. We dont need to dene a whole new class. We can pick one of the existing NumberPair objects at random from the collection of keys for the Map that contains the individual Throws.

149

Building Skills in Object-Oriented Design, Release 2.1.1-Python

2. We can transform the two numeric dice values to a single index value for the sequence. This is a technique called Key Address Transformation; we transform the keys into the address (or index) of the data. We create the index, i, from two dice, d1 , d2 , via a simple linear equation: i = 6(d1 1) + (d2 1). We can reverse this calculation to determine the two dice values from an index. d1 = i 6 + 1; d2 = (i mod 6) + 1. Because of encapsulation, the choice of algorithm is completely hidden within the implementation of Dice. While the numeric calculation seems simple, it doesnt obviously scale to larger collections of dice very well. While Craps is a two-dice game, we can imagine simulating a game with larger number of dice, making this technique complex. Solution. Our recommendation is to encapsulate the pair of dice in a tuple instance. We can use this object as index into a dict that associates a tuple with a Throw. More advanced students can create a class hierarchy for Dice that includes all of the various implementations as alternative subclasses. Random Selection. The random number generator in random.Random helps us locate a Throw at random. First, we can get the list of keys from the dict that associates a tuple of dice numbers with a Throw. Second, we use Random.choice() to pick one of these tuples. We use this randomly selected tuple to return the selected Throw.

26.2 Throw Rework

We need to update Throw to return an appropriate key object. There are two parts to this. First, we need a getter to return the key. Well add a method to Throw to return the tuple that is a key for this Throw. getKey(self ) Second, we need to update the Throw constructor to create the key when the Throw is being built. This will allow all parts of the application to share references to a single instance of the key.

26.3 NumberPair Design

class NumberPair() In Python, its often easiest to use a named tuple rather than use a simple tuple.
from collections import namedtuple NumberPair = namedtuple('NumberPair', ['d1','d2'])

This can make the tuple slightly easier to work with.

26.3.1 Fields
d1 Contains the face of one die.

150

Chapter 26. Dice Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

d2 Contains the face of the other die. These attributes are built for us by the collections.namedtuple function.

26.3.2 Constructors
The constructor are built for us by the collections.namedtuple function.

26.4 Dice Design

class Dice() Dice contains the 36 individual throws of two dice, plus a random number generator. It can select a Throw at random, simulating a throw of the Craps dice.

26.4.1 Fields
throws This is a dict that associates a NumberPair with a Throw. rng An instance of random.Random Generates the next random number, used to select a Throw from the throws collection.

26.4.2 Constructors
__init__(self, rng=None) Build the dictionary of Throw instances. Parameter rng (random.Random) The random number generator to use. At the present time, this does not do the full initialization of all of the Throws. Were only building the features of Dice related to random selection. Well extend this class in a future exercise.

26.4.3 Methods
addThrow(self, throw ) Parameter throw (Throw) The Throw to add. Adds the given Throw to the mapping maintained by this instance of Dice. The key for this Throw is available from the Throw.getKey() method. next(self ) Returns the randomly selected Throw. First, get the list of keys from the throws. The random.Random.choice() method will select one of the available keys from the the list. This is used to get the corresponding Throw from the throws Map.

26.4. Dice Design

151

Building Skills in Object-Oriented Design, Release 2.1.1-Python

getThrow(self, d1, d2 ) Parameters d1 The value of one die d2 The other die While not needed by the application, unit tests may need a method to return a specic Throw rather than a randomly selected Throw. This method takes a particular combination of dice, locates (or creates) a NumberPair, and returns the appropriate Throw.

26.5 Dice Deliverables

There are three deliverables for this exercise. In considering the unit test requirements, we note that we will have to follow the design of the Wheel class for convenient testability: we will need a way to get a particular Throw from the Dice, as well as replacing the random number generator with one that produces a known sequence of numbers. The Dice class. A class which performs a unit test of building the Dice class. The unit test should create several instances of Outcome, two instances of Throw, and an instance of Dice. The unit test should establish that Throws can be added to the Dice. A class which performs a demonstration of selecting non-random values from the Dice class. By setting a particular seed, the Throws will be returned in a xed order. To discover this non-random order, a demonstration should be built which includes the following. 1. Create several instances of Outcome. 2. Create two instances of Throw that use the available Outcomes. 3. Create one instance of Dice that uses the two Throws. 4. A number of calls to the next() method should return randomly selected Throws. Note that the sequence of random numbers is xed by the seed value. The default constructor for a random number generator creates a seed based on the system clock. If your unit test sets a particular seed value, you will get a xed sequence of numbers that can be used to get a consistent result.

26.6 Dice Optimization

First, we note that premature optimization is a common trap. We should forget about small eiciencies, say about 97% of the time: premature optimization is the root of all evil. Yet we should not pass up our opportunities in that critical 3%. A good programmer will not be lulled into complacency by such reasoning, he will be wise to look carefully at the critical code; but only after that code has been identied Donald Knuth Structured Programming with Goto Statements. Computing Surveys 6:4 (1974), 261-301. In this exercise, it appears that the NumberPair is superuous. We can work directly with a list of Throws, bypassing any mapping between NumberPair and Throw.

152

Chapter 26. Dice Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Removing this needless code is an exercise that should be considered, but not done at this time. In the Design Cleanup and Refactoring chapter, well clean up this design to remove several things which in retrospect will be poor design decisions.

26.6. Dice Optimization

153

Building Skills in Object-Oriented Design, Release 2.1.1-Python

154

Chapter 26. Dice Class

CHAPTER

TWENTYSEVEN

THROW BUILDER CLASS

This chapter identies some subtleties of bets in Craps, and proposes some rather involved design rework to resolve the issues that are raised. Specically, we didnt allow for an Outcome to have odds based on a throw of the dice. Well need to add this feature carefully. Well present two solutions: Design Heavy and Design Light. While we recommend Design Heavy, we nd that many programmers and project managers object to the apparent over-engineering of the approach. We have found that it is almost impossible justify the design heavy approach. See our Soapbox on Justication for more on this common blind spot. Instead, we provide the light alternative, which will become increasingly unwieldy throughout this part. At some point, we will eventually be forced to adopt the design heavy approach. Well present sidebars on the proper design of subclasses and the proper architecture for the packages that make up an application. Additionally, well provide a brief FAQ on the design issues raised.

27.1 Throw Builder Overview

Enumerating each Outcome in the 36 Throws could be a tedious undertaking. It may be slightly simpler to hand-code the various combinations., For complete exibility, well design a Builder to enumerate all of the Throws and their associated list of Outcomes. This will build the Dice, nishing the elements we deferred from Dice Class. The 36 ways the dice fall can be summarized into 15 kinds of Throw, with a xed distribution of probabilities. We could develop a Builder class that enumerates the 36 individual Throws, assigning the appropriate attribute values to each object. An alternative is for a Builder class to step through the 15 kinds of Throws, creating the proper number of instances of each kind. It seems slightly simpler to examine each pair of dice and determine which kind of Throw to build. The proposition bets dene eight one-roll Outcomes that need to be assigned to the various Throw instances we are building. We will share references to the following Outcome objects among the Throws: The number 2, with 30:1 odds. The number 3, with 15:1 odds. The number 7, with 4:1 odds. The number 11, with 15:1 odds. The number 12, with 30:1 odds. The any craps outcome, with 7:1 odds. This belongs to throws for 2, 3 and 12. The horn outcome, with two sets of odds. This belongs to throws for 2, 3, 11 and 12. For 2 and 12, the odds are 27:4; for 3 and 11, the odds are 3:1. 155

Building Skills in Object-Oriented Design, Release 2.1.1-Python

The eld outcome, with two sets of odds. This belongs to throws for 2, 3, 4, 9, 10, 11 and 12. For 2 and 12, this pays 2:1, all others pay even money (1:1). We can use the following algorithm for building the Dice.

Building Dice
For All Faces Of Die 1. For all d1 , such that 1 d1 < 7: For All Faces Of A Die 2. For d2 , such that 1 d2 < 7: Sum the Dice. Compute the sum, s d1 + d2 . Craps? If s is in 2, 3, and 12, we create a CrapsThrow instance. This will include a reference to one of the 2, 3 or 12 Outcome s, plus references to the Any Craps, Horn and Field Outcomes. Point? For s in 4, 5, 6, 8, 9, and 10 we will create a PointThrow instance. Hard? When d1 = d2 , this is a hard 4, 6, 8 or 10. Easy? Otherwise, d1 = d2 , this is an easy 4, 6, 8 or 10. Field? For s in 4, 9 and 10 we include a refrence to the Field Outcome. Note that 2, 3, and 12 Field outcomes where handled above under Craps. Natural? For s of 7, we create a NaturalThrow instance. This will also include a reference to the 7 Outcome. Eleven? For s of 11, we create an ElevenThrow instance. This will include references to the 11, Horn and Field Outcomes.

27.2 Outcomes with Variable Odds

Our detailed examination of the bets has turned up an interesting fact about Field bets and Horn bets: these bets also have payos that depend on the number on the dice. In our earlier Outcome Class, we missed this nuance, and did not provide for a Dice.winAmount() method that depends on the Dice. Well present the solution to this as a fairly formal procedure that we nd helps to resolve these kind of design issues. Problem Statement. Unlike the Pass Line and Come Line bets, Field bets and Horn bets have payos that depend on the number currently showing on the dice. Unlike Come Line bets, these bets are not moved on the table to a new Outcome when the point is established. How do we compute the win amount for Field and Horn bets? Context. Our design objective is to have a Bet reference a single Outcome object. The Bet is compared with a Set of winning Outcomes. Wed like to have a single horn Outcome object and eld Outcome object shared by multiple instances of Throw to make this comparison work in a simple, general way. As an example, the player can place a bet on the Field Outcome, which is shared by all of the eld numbers (2, 3, 4, 9, 10, 11, 12). The problem we have is that for 2 and 12, the outcome pays 2:1 and for the other eld numbers it pays 1:1, and our design only has a single set of payout odds. Forces. In order to handle this neatly, we have two choices. One choice is to have two Outcomes bundled into a single bet. This allows us to create a Bet that includes both the low-odds eld outcome (3, 4, 9, 10 and 11) plus the high-odds eld outcome (2 and 12). One of the nice features of this is that it is a small expansion to Bet. However, further research shows us that there are casino-specic variations on the eld bet, including the possibility of three separate Outcomes for those casinos that pay 3:1 on 12. This makes construction of the Bet rather 156 Chapter 27. Throw Builder Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

complex, and dilutes the responsibility for creating a proper Bet. Once we put multiple Outcomes into a Bet, we need to assign responsibility for keeping the bundle of Field Outcomes together. Pursuing this further, we could expand Outcome to follow the Composite design pattern. We could introduce a subclass which was a bundle of multiple Outcomes. This would allow us to keep Bet very simple, but we still have to construct appropriate composite Outcome instances for the purpose of creating Bets. Rather than dive into allocating this responsibility, well look at other alternatives, and see if something turns up that doesnt add as much complexity. Another approach is to add an optional argument to Outcome that uses the current Throw to calculate the win amount. This allows us to have a single eld bet Outcome with dierent odds for the various numbers in the eld. This further allows us to create slightly dierent eld bet Outcome class denitions for the casino-specic variations on the rules. Solution. Our rst design decision, then, is to add an additional method to Outcome that calculates the win amount given the current Throw. Consequences. There are a number of consequences of this design decision. The next design problem we have to solve is where in the Outcome class hierarchy do we add this additional winAmount() method? After that, we will need to determine the signature of the revised winAmount() method. This leads us to two rounds of additional problem-solving.

27.3 Refactoring The Outcome Hierarchy

Consequent Problem: Class Hierarchy. While it appears simplest to add a variable odds subclass of Outcome with a new method that uses the number on the dice, we nd that there are some additional considerations. Our design depends on polymorphism among objects of the Outcome class: all instances have the same interface. In order to maintain this polymorphism, we need to add this new method to the superclass. The superclass version of the new winAmount() based on the Craps Throw can return an answer computed by the original winAmount() method. We can then override this in a subclass for Field and Horn bets in Craps. An alternative is to break polymorphism and create a Craps-specic Outcome subclass. This would ripple out to Throw, Bet, Table, Player. This is an unpleasant cascade of change, easily avoided by assuring that the entire Outcome class hierarchy is polymorphic. Solution. Our second design decision, then, is to insert the change at the top of the Outcome class hierarchy, and override this new winAmount() method in just the few subclasses that we use to create Horn and Field Outcomes. The Horn bets winAmount() method applies one of two odds, based on the events value. The Field bet may have any of two or three odds, depending on the casinos house rules. It is diicult to identify a lot of commonality between Horn bets and Field bets. Faced with these irreconcilable dierences, we will need two dierent winAmount() methods, leading us to create two subclasses: OutcomeField and OutcomeHorn. The dierences are minor, merely a list of numbers and odds. However, our design objective is to minimize if-statements. We prefer many simple classes over a single class with even a moderately complex method. Consequent Problem: Dependencies. Weve decided to add a dependency to the Outcome.winAmount(); specically, weve made it dependent on a Throw object. While this works well for Craps, it makes very little sense for Roulette.

27.3. Refactoring The Outcome Hierarchy

157

Building Skills in Object-Oriented Design, Release 2.1.1-Python

To allow the games to evolve independently, we should not have any dependencies between games. This means that a general-purpose class like Outcome cant depend on a game-specic class like Throw. A generalpurpose class has to depend on some a superclass (or interface) that encompasses the Craps-specic Throw as well as the Roulette-specic Bin. Additional Classes. To break the dependency between a general-purposes class and a game-specic class, we need introduce a superclass that includes both Throw and Bin as subclasses. This will Outcome to work with either Craps and Roulette; keeping them independent of each other. We could call the parent class a RandomEvent. This new class would have an integer event identier: either the wheels bin number or the total of the two dice. Given this new superclass, we could then rearrange both Throw and Bin to be subclasses of RandomEvent. This would also force us to rework parts of Wheel that creates the Bins. A benet of creating a RandomEvent class hierarchy is that we can change the new winAmount() method to compute the win amount given a RandomEvent instead of a highly Craps-specic Throw. This makes the winAmount() method far more generally useful, and keeps Craps and Roulette separate from each other. This technique of reworking Throw and Bin to be subclasses of a common superclass is a fairly common kind of generalization refactoring. We observe that more experienced designers develop an ability to locate this kind of commonality. Less experienced designers tend to conate too many nearly-common features into a single class, leading to a brittle design that cannot easily be reworked. In our example, we considered lifting one common attribute to the superclass so that a related class (Outcome) could operate on instances of these two classes in a uniform manner. For more information on this rework, see Soapbox on Subclasses. Approaches. We will present two alternative designs paths: minimal rework, and a design that is at the fringe of over-engineering. The minimal design eort has one unpleasant consequence: Roulettes Outcome instances depend on the Craps-related Throw class. This leads to an entanglement between Roulette and Craps around a feature that is really a special case for Craps only. This kind of entanglement may limit our ability to successfully reuse these classes. See Soapbox on Architecture for a discussion on this issue. The possible over-engineering prevents this entanglement. We consider this separation very desirable, even in an application as small as this exercise.

27.4 Soapbox on Subclasses

Designers new to OO techniques are sometimes uncomfortable with the notion of highlyspecialized subclasses. Well touch on two reasons why specialized subclasses are far superior to the alternatives. One approach to creating common features is to add nested if-statements instead of creating subclasses. In our example, we might have elected to add if-statements that would determine if this was a variable-odds outcome, and then determine which of the available odds would be used. The rst test (for being a variable-odds outcome) is, in eect, a determination of the subclass of Outcome. Since an objects membership in a class determines the available methods, theres no reason to test for membership. In most cases, a test for membership in a class is also done at construction time. If we use that initial decision to select the subclass (with appropriate subclass-specic methods) we do not repeat that decision every time a method is invoked. This is the eiciency rationale for introducing a subclass to handle these special cases. The more fundamental reason is that specialized subclasses usually represent distinct kinds of real-world things, which we are modeling in software-world. In contrast, a procedural viewpoint is that the variant behavior is a special case: a condition or situation that requires unique processing. This view often confuses the implementation of the special case (via an if-statement) with the nature of the specialization. In our case, we have a number of distinct things, some

158

Chapter 27. Throw Builder Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

of which are related because they have common attributes and behavior. The Outcome is fairly intangible, so the notion of commonality can be diicult to see. Contrast this with Dice and Wheel, which are tangible, and are obviously dierent things, however they have common behavior and a common relationship with a casino game. Design Aid. Sometimes it helps to visualize this by getting pads of dierent-colored sticky paper, and making a mockup of the object structure on whiteboard. Each class is represented by a dierent color of paper. Each individual object is an individual slip of sticky paper. To show the relationship of Dice, Throw and Outcome, we draw a large space on the board for an instance of Dice which has smaller spaces for 36 individual Throws. In one Throw instance, we put a sticky for Outcome s 2, Field, Horn, and Any Craps. We use three colors of stickies to show that 2 and Any Craps are ordinary Outcome s, Field is one subclass and Horn is another subclass. In another Throw instance, we put a sticky for Outcome 7, using the color of sticky for ordinary Outcomes. This can help to show what the nal game object will be examining to evaluate winning bets. The game object will have a list of winning Outcomes and bet Outcome s actually on the table. When a 2 is thrown, the game process will pick up each of the stickies, compare the winning Outcome s to the bets, and then use the method appropriate to the color of the sticky when computing the results of the bet.

27.5 Soapbox on Architecture

There are a number of advanced considerations behind the Design Heavy section. This is a digression on architecture and packages of classes. While this is beyond the basics of OO design, it is a kind of justication for the architecture weve chosen. A good design balances a number of forces. One example of this is our use of a class hierarchy to decompose a problem into related class descriptions, coupled with the collaboration among individual objects to compose the desired solution. The desired behavior emerges from this tension between decomposition of the class design and composition of the objects to create the desired behavior. Another example of this decomposition vs. composition is the organization of our classes into packages. We have, in this book, avoided discussion of how we package classes. It is a more subtle aspect of a good design, consequently we nd it challenging to articulate sound principles behind the layers and partitions of a good collection of packages. There are some design patterns that give us packaging guidance, however. Design Patterns. One packaging pattern is the 5-Layer Design, which encourages us to separate our design into layers of view, control, model, access and persistence. For our current application, the view is the output log written to System.out, the control is the overall main method and the Simulation class, the model is the casino game model. We dont have any data access or data persistence issues, but these are often implemented with JDBC and a relational database. While one of the most helpful architectural patterns, this version of the 5-Layer Design still leaves us with some unsatisfying gaps. For example, common or infrastructure elements dont have a proper home. They seem to form another layer (or set of layers). Further, the model layer often decomposes into domain elements, plus elements which are specializations focused on unique features of the business, customer, vendor or product. Another packaging pattern is the Sibling Partition, which encourages us to separate our application-specic elements to make them parallel siblings of a superclass so that we can more 27.5. Soapbox on Architecture 159

Building Skills in Object-Oriented Design, Release 2.1.1-Python

easily add new applications or remove obsolete applications. In this case, each casino game is a separate application of our casino game simulator. At some point, we may want to isolate one of the games to reuse just the classes of that game in another application. By making the games proper siblings of each other, and children of an abstract parent, they can be more easily separated. General vs. Specic. Applying these layered design and application partitioning design patterns causes us to examine our casino game model more closely and further sub-divide the model into game-specic and game-independent elements. If necessary, we can further subdivide the general elements into those that are part of the problem domain (casino games) and those that are even more general application infrastructure (e.g., simulation and statistics). Our ideal is to have a tidy, short list of classes that provides a complete game simulation. We can cut our current design into three parts: Roulette, Craps and application infrastructure. This allows us to compose Roulette from the Roulette-specic classes and the general infrastructure classes, without including any of the Craps-specic classes. The following architecture diagram captures a way to structure the packages of these applications.

Our class denitions have implicitly followed this architecture, working from general to gameand player-specic classes. Note that our low-level classes evolved through several increments. We nd this to be superior to attempting to design the general classes from the outset: it avoids any over-engineering of the supporting infrastructure. Additionally, we we careful to assure that our top-level classes contain minimal processing, and are are compositions of lower-level object instances. Dependencies. A very good design could carefully formalize this aspect of the architecture by assuring that there are minimal references between layers and partitions, and all references are downward references from application-specic to general infrastructure packages. In our case, the Simulator should have access only to Player and Game layers. Two Game partitions should be separate with no references between these packages. Finally, we would like to assure that the Player and Game dont have invalid upward references to the Simulator. There are some tools that can enforce this architectural separation aspect of the design. Lacking tools, we need to exercise some discipline to honor the layers and partitions.

160

Chapter 27. Throw Builder Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

27.6 Throw Builder Questions and Answers

Why is do we need RandomEvent? Isnt this overengineering? Clean separation between Craps and Roulette isnt necessary, but is highly desirable. We prefer not to have Roulette classes depend in any way on Craps classes. Instead of having them entangled, we factor out the entanglement and make a new class from this. This is also called reducing the coupling between classes. We prefer the term entanglement because it has a suitably negative connotation. Didnt we notice the need for this RandomEvent class back in Soapbox on Overengineering? Why was it a bad idea then and a good idea now? Some experienced designers immediately notice this kind of commonality between Throw and Bin, and can handle it without getting badly side-tracked. However, some beginning designers can spend too much time searching for this kind of commonality. Weve seen examples of improper factoring where classes were combined in an early phase of design, only to lead to expensive rework when distinctions were uncovered later, invalidating large parts of the design. We prefer to wait until we are sure weve understood the problem and the solution before committing to a particular class design. Isnt the goal to leave Roulette alone? Isnt the ideal to extend the design with subclasses, leaving the original design in place? Yes, the goal is to extend a design via subclasses. But, this is only possible if the original design is suitable for extension by subclassing. We nd that it is very diicult to create a design that both solves a problem and can be extended to solve a number of related problems. Note that a general, extensible design has two independent feature sets. On one level it solves a useful problem. Often, this is a diicult problem in its own right, and requires considerable skill merely to ferret out the actual problem and craft a usable solution within budget, time and skill constraints. On another, deeper level, our ideal design can be extended. This is a dierent kind of problem that requires us to consider the various kinds of design mutations that may occur as the software is maintained and adapted. This requires some in-depth knowledge of the problem domain. We need to know how the current problem is a specialization of other more general problems. We also need to note how our solution is only one of many solutions to the current problem. We have two dimensions of generalization: problem generalization as well as solution generalization. Our initial design for roulette just barely provided the rst level of solution. We didnt make any eort to plan for generalization. The Design Heavy solution generalizes Roulette to make it more suitable for Craps, also. Looking forward, well have to make even more adjustments before we have a very tidy, general solution.

27.7 Soapbox on Justication

It is very diicult to justify design rework. Most managers and designers share a common blindspot on the amount of evidence required to justify rework. The conversations have the following form. Architect. We need to disentangle Roulette and Craps so that Craps is not a subclass of Roulette. Ive got a revised design that will take X hours of eort to implement. Manager. Ill need some justication. Why do we have to x it?

27.6. Throw Builder Questions and Answers

161

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Architect. The structure is illogical: Craps isnt a special case of Roulette, theyre independent specializations of something more general. Manager. Illogical isnt a good enough justication. Our overall problem domain always contains illogical special cases and user-oriented considerations. Youll have to provide something more concrete. Architect. Okay, in addition to being illogical, it will become too complex: in the future, well probably have trouble implementing other games. Manager. How much trouble? Will it be more than X hours of eort? Architect. When we include maintenance, adaptation and debugging, the potential future cost is probably larger than 2X hours of eort. And its illogical. Manager. Probably larger? We cant justify rework based on probable costs. Youll need something tangible. Will it save us any lines of code? Architect. No, it will add lines of code. But it will reduce maintenance and adaptation costs because it will be more logical. Manager. I conclude that the change is unjustied. In many cases, we have a manager whos mind is made up and who wont be swayed by facts. This is generally a symptom of an organization that is thinking-impaired. Typically, it is impossible to engage project managers further than this. However, in some cases, conversation continues in the following vein. Architect. Unjustied? Okay, please dene what constitutes adequate justication for improvements? Manager. Real savings of eort is the only justication for disrupting the schedule. Architect. And future eort doesnt count? Manager. The probability of savings in the future isnt tangible. Architect. And more logical doesnt count? Manager. If course not; it doesnt result in real schedule savings. Architect. Real schedule savings? Thats absurd. The schedule is a notional projection of possible eort. A possible reduction in the possible eort is just as real as the schedule. Manager. Wouldnt it be simpler to...? For reasons we dont fully understand, a schedule becomes a kind of established fact. Any change to the schedule requires other established facts, not the conjecture of design. For some reason, the idea that the schedule is only a conjecture, based on a previously conjectured design doesnt seem to sway managers from clinging to the schedule. This makes it nearly impossible to justiy making a design change. The only way to accumulate enough evidence is to make the design change and then measure the impact of the change. In eect, no level of proof can ever override the precedence-setting fact of the schedule. To continue this rant, the same kind of inadequate evidence issue seems to surround all technology changes. We have had conversations like the one shown above regarding object-oriented design, the use of objects in relational databases, the use of object-oriented databases, the use of open-source software, and the use of the star-schema data model for reporting and analysis. For many people, it seems that change can only be considered based on established facts; we observe that these facts can only be established by making a change. While this chicken-and-egg problem is easily resolved by recognizing that the current architecture or schedule is actually not an established fact, this is a diicult mental step to take.

162

Chapter 27. Throw Builder Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

As a nal complaint, we note that wouldnt it be simpler is best described as a management trump card. The point is rarely an eort to reduce code complexity, but to promote management understanding of the design. While this is a noble eort, there is sometimes a communication gap between designers and managers. It is incumbent both on designers to communicate fully, and on managers to provide time, budget and positive reinforcement for detailed communication of design considerations and consequences.

27.8 Design Light

In order to get the Craps game to work, we can minimize the amount of design. This minimal rework is a revision to Outcome. This is followed by Common Design: the two subclasses of Outcome (OutcomeField, and OutcomeHorn), and the initializer for Dice. This minimal design eort has one unpleasant consequence: Roulettes Outcome instances will depend on the Craps-specic Throw class. This entangles Roulette and Craps around a feature that is really a special case for Craps only. This kind of entanglement often limits our ability to successfully package and reuse these classes.

27.8.1 Outcome Rework

The class Outcome needs a method to compute the win amount based on a Throw. In Python, we use optional parameters to achieve the same degree of exibility. winAmount(self, throw=None) Returns the product this Outcomes odds numerator by the given amount, divided by the odds denominator. Parameter throw (Throw) An optional Throw that determines the actual odds to use. If not provided, this Outcomes odds are used. For Craps Horn bet and Field bets, a subclass will override this method to check the specic value of the throw and compute appropriate odds.

27.9 Design Heavy

In order to produce a solution that has a better architecture with more reusable components, we need to do some additional generalization. This design eort disentangles Roulette and Craps; they will not share the Throw class that should only be part of Craps. Instead, the highly reused Outcome class depends only on a new superclass, RandomEvent, which is not specic to either game. Given the new generalization, RandomEvent, we can rework the Outcome to use this for computing win amounts. We will have to rework Bin, Wheel, and Throw to make proper use of this new superclass. Then we can move to the Common Design features: the craps-specic subclasses (OutcomeField, and OutcomeHorn), and the initializer for Dice.

27.9.1 RandomEvent class

class RandomEvent()

27.8. Design Light

163

Building Skills in Object-Oriented Design, Release 2.1.1-Python

The class RandomEvent is the superclass for the random events on which a player bets. This includes Bin of a Roulette wheel and Throw of Craps dice. Fields The most notable common feature between a Bin and a Throw is that both have a collection of Outcomes. This common set can be moved to this superclas from the subclasses. outcomes A frozenset that holds the collection of individual Outcomes.

27.9.2 Bin Rework

The class Bin needs to be a subclass of RandomEvent. The set of outcomes is removed from Bin; its dened in RandomEvent.

27.9.3 Throw Rework

The class Throw needs to be a subclass of RandomEvent. The set of outcomes is removed from Throw; its dened in RandomEvent.

27.9.4 Outcome Rework

The class Outcome needs a method to compute the win amount based on a RandomEvent. In Python, we use optional parameters to achieve the same degree of exibility. winAmount(self, RandomEvent=None) Returns the product this Outcomes odds numerator by the given amount, divided by the odds denominator. Parameter event (Throw) An optional RandomEvent that determines the actual odds to use. If not provided, this Outcomes odds are used. For Craps Horn bet and Field bets, a subclass will override this method to check the specic value of the event and compute appropriate odds.

27.10 Common Design

27.10.1 OutcomeField Design
class OutcomeField() OutcomeField contains a single outcome for a eld bets that has a number of dierent odds, and the odds used depend on a RandomEvent.

164

Chapter 27. Throw Builder Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Methods winAmount(self, throw=None) Returns the product this Outcomes odds numerator by the given amount, divided by the odds denominator. Parameter throw (Throw) An optional Throw that determines the actual odds to use. If not provided, this Outcomes odds are used. __str__(self ) This should return a String representation of the name and the odds. A form that looks like Field (1:1, 2 and 12 2:1) works nicely.

27.10.2 OutcomeHorn Design

class OutcomeHorn() OutcomeHorn contains a single outcome for a Horn bet that has a number of dierent odds, and the odds used depend on a RandomEvent. Methods winAmount(self, throw=None) Returns the product this Outcomes odds numerator by the given amount, divided by the odds denominator. Parameter throw (Throw) An optional Throw that determines the actual odds to use. If not provided, this Outcomes odds are used. __str__(self ) This should return a String representation of the name and the odds. A form that looks like Horn (27:4, 3:1) works nicely.

27.10.3 ThrowBuilder class

class ThrowBuilder() ThrowBuilder initializes the 36 Throws, each initialized with the appropriate Outcomes. Subclasses can override this to reect dierent casino-specic rules for odds on Field bets. Constructors __init__(self ) Initializes the ThrowBuilder. Methods buildThrows(self, dice) Creates the 8 one-roll Outcome instances (2, 3, 7, 11, 12, Field, Horn, Any Craps). It then creates each of the 36 Throws, each of which has the appropriate combination of Outcomes. The Throws are assigned to dice.

27.10. Common Design

165

Building Skills in Object-Oriented Design, Release 2.1.1-Python

27.11 Throw-Builder Deliverables

There are two deliverables for the light version of this exercise. Rework the Outcome class to add the new winAmount() method that uses a Throw. Rework the Outcome class unit test to exercise the new winAmount() method that uses a Throw. For all current subclasses of Outcome, the results of both versions of the winAmount() method produce the same results. There are ve deliverables for the heavy version of this exercise. Create the RandomEvent class. Rework the Bin class to be a subclass of RandomEvent. The existing unit tests for Bin should continue to work correctly. Rework the Throw class to be a subclass of RandomEvent. The existing unit tests should continue to work correctly. Rework the Outcome class to add the new winAmount() method that uses a RandomEvent. Rework the Outcome class unit test to exercise the new winAmount() method that uses a RandomEvent. For all current subclasses of Outcome, the results of both versions of the winAmount() method produce the same results. There a six common deliverables no matter which approach you take. Create the OutcomeField class. Create a unit test for the OutcomeField class. Two instances of Throw are required: a 2 and a 3. This should conrm that there are dierent values for winAmount() for the two dierent Throw instances. Create the OutcomeHorn class. Create a unit test for the OutcomeHorn class. Two instances of Throw are required: a 2 and a 3. This should conrm that there are dierent values for winAmount() for the two dierent Throw instances. Create the ThrowBuilder. This was our objective, after all. Rework the unit test of the Dice class. The unit test should create and initialize a Dice. It can use the getThrow() method to check selected Throws for the correct Outcomes. The correct distribution of throws is as follows. ThrowBuilder. Throw 2 3 easy 4 hard 4 5 easy 6 hard 6 7 easy 8 hard 8 9 easy 10 hard 10 11 12 166 Frequency 1 2 2 1 4 4 1 6 4 1 4 2 1 2 1 Chapter 27. Throw Builder Class This information will help conrm the results of

CHAPTER

TWENTYEIGHT

BET CLASS
This chapter will examine the Bet class, and its suitability for the game of Craps. Well expand the design to handle additional complications present in real casino games.

28.1 Bet Overview

A Bet is an amount that the player has wagered on a specic Outcome. This is a simple association of an amount, an Outcome, and a specic Player. When considering the various line bet outcomes (Pass Line, Come Line, Dont Pass and Dont Come), we noted that when a point was established the bet was either a winner or a loser, or it was moved from the line to a particular number based on the throw of the dice. Well need to add this responsibility to our existing denition of Bet. This responsibility can be implemented as a setOutcome() method that leaves the amount intact, but changes the Outcome from the initial Pass Line or Come Line to a specic point outcome. A complexity of placing bets in Craps is the commission (or vigorish) required for Buy bets and Lay bets. This is a 5% fee, in addition to the bet amount. A player puts $21 down, which is a $20 bet and a $1 commission. Well need to add a a commission or vig responsibility to our denition of Bet. This price to place a bet generalizes nicely to all other bets. For most bets, the price is simply the amount of the bet. For Buy bets, the price is 5% higher than the amount of the bet; for Lay bets, the price depends on the odds. This adds a new method to Bet that computes the price of the bet. This has a ripple eect throughout our Player hierarchy to reect this notion of the price of a bet. We will have to make a series of updates to properly deduct the price from the players stake instead of deducting the amount of the bet. There are two parts to creating a proper Craps bet: a revision of the base Bet to separate the price from the amount bet, and a CommissionBet subclass to compute prices properly for the more complex Craps bets.

28.2 Bet Rework

Bet associates an amount and an Outcome. The Game may move a Bet to a dierent Outcome to reect a change in the odds used to resolve the Bet. In a future round of design, we can also associate a it with a Player.

28.2.1 Methods
setOutcome(self, outcome) Parameter outcome (Outcome) The new outcome for this bet amount

167

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Sets the Outcome for this bet. This has the eect of moving the bet to another Outcome. price(self ) Computes the price for this bet. For most bets, the price is the amount. Subclasses can override this to handle buy and lay bets where the price includes a 5% commission on the potential winnings. For Buy and Lay bets, a $20 bet has a price of $21.

28.3 CommissionBet Design

class CommissionBet() CommissionBet is a Bet with a commission payment (or vigorish) that determines the price for placing the bet.

28.3.1 Fields
vig Holds the amount of the vigorish. This is almost universally 5%.

28.3.2 Methods
price(self ) Computes the price for this bet. There are two variations: Buy bets and Lay bets. A Buy bet is a right bet; it has a numerator greater than or equal to the denominator (for example, 2:1 odds, which risks 1 to win 2), the price is 5% of the amount bet. A $20 Buy bet has a price of $21. A Lay bet is a wrong bet; it has a denominator greater than the numerator (for example, 2:3 odds, which risks 3 to win 2), the price is 5% of 2/3 of the amount. A $30 bet Layed at 2:3 odds has a price of $31, the $30 bet, plus the vig of 5% of $20 payout.

28.4 Bet Deliverables

There are three deliverables for this exercise. The revised Bet class. The new CommissionBet subclass. This computes a price that is 5% of the bet amount. A class which performs a unit test of the various Bet classes. The unit test should create a couple instances of Outcome, and establish that the winAmount() and price() methods work correctly. It should also reset the Outcome associated with a Bet We could rework the entire Player class hierarchy for Roulette to compute the Bets price in the placeBets(), and deduct that price from the players stake. For Roulette, however, this subtlety is at the fringe of over-engineering, as no bet in Roulette has a commission.

168

Chapter 28. Bet Class

CHAPTER

TWENTYNINE

CRAPS TABLE CLASS

In Roulette, the table was a passive repository for Bets. In Craps, however, the table and game must collaborate to accept or reject bets based on the state of the game. This validation includes rules based on the total amount bet as well as rules for the individual bets. In Throw Class, we roughed out a stub version of CrapsGame that could be used to test Throw. In this section, we will extend this stub with additional features required by the table.

29.1 Craps Table Overview

The Table is where the Bets are placed. The money placed on Bets on the Table is at risk; these bets either win an amount based on the odds, or lose the amount placed by the player. The Dont Come and Dont Pass bets may be returned, called a push. The Buy and Lay bets include a commission (or vigorish) to place the bet; the commission is lost money; the balance of the bet, however, may win or lose. The responsibility for a push is something we can allocate to Game, the commission price belongs to Bet. Some Bets (specically Pass, Dont Pass, Come and Dont Come) may have their Outcome changed. The use case works like this. 1. The bet is created by the Player with one Outcome, for example, Pass. The Table accepts this bet. 2. That bet may be resolved as an immediate winner or loser. The Game and Throw will determine if the Bet is a winner as placed. More commonly, the Bet may be changed to a new Outcome, possibly with dierent odds. In a casino, the chips initially placed on Come Line and Dont Come bets are relocated to a point number box to show this change. In the case of Pass Line and Dont Pass bets, the On marker is placed on the table to show an implicit movement of all of those line bets. The change is the responsibility of the Game; however, the Table must provide an iterator over the line bets that the Game will move. Game State. Each change to the game state changes the allowed bets as well as the active bets. When the point is o, most of the bets on the table are not allowed, and some others are inactive, or not working. When a point is established, all bets are allowed, and all bets are active. Well examine the rules in detail, below. The Table must be able to reject bets which are inappropriate for the current Game state.

169

Building Skills in Object-Oriented Design, Release 2.1.1-Python

29.2 Design Decision Table vs. Game Responsibility

Weve identied two responsibilities that are part of handling Craps: changing Bets, inactivating outcomes and rejecting bets based on Game state. Clearly, these require additional collaboration between Game and Table. We will have to add methods to CrapsGame that will allow or deny some bets, as well as methods that will active or deactive some bets. We have to choose where in the class hierarchy we will retrot this additional collaboration. Problem. Should we put these new responsibilities at a high-enough level that well add table and game collaboriation to the Table class used for Roulette? Forces. If we do add this for Roulette, we could simply return true from the method that validates the allowed bets, since all bets are allowed in Roulette. However, our overall application design does not depend on all subclasses of Game and Table being polymorphic; we will never mix and match dierent combinations of Craps Table and Roulette Game. Solution. Because we dont need polymorphism between Craps and Roulette, we can create a subclass of Table with a more complex interface and leave Roulette untouched. Perhaps well call it CrapsTable.

29.3 Design Decision Allowable Outcomes

After deciding to create a CrapsTable subclass, we have several consequent decisions. First, we turn the interesting question of how best to allocate responsibility for keeping the list of Outcomes which change with the game state. Problem. Which class determines the valid and invalid Outcomes? Forces. We can see three places to place this responsibility. 1. We could make CrapsTable responsible; it could have methods to return the lists of Outcomes that are allowed or not allowed. CrapsGame can make a call to get the list of Outcomes and make the changes. Making each change would involve the CrapsTable a second time to mark the individual Outcomes. This information is then used by the CrapsTable to validate individual Bets. 2. We could CrapsGame responsible; it could invoke a method of CrapsTable that changes a single Outcome s state to makt it inactive. This information is then used by the CrapsTable to validate individual Bets. 3. An appealing choice is to have the validBet() method of CrapsTable depend on CrapsGame to determine which bets are allowed or denied. In this case, CrapsGame has the responsibility to respond to requests from either CrapsTable or Player regarding a specic Outcomes. The third choice seems to focus bet-handling responsibility mostly on CrapsGame. Solution We need to place a valid method in the CrapsTable; but the Table simply delegates the details to the CrapsGame. This allows the Player to deal directly with the Table. But it centralizes the actual decision-making on the Game. Consequences. The game must move Outcomes for certain kinds of bets. Additionally, the CrapsTables isValid() method will use the CrapsGame to both check the validity of individual bets as well as the entire set of bets created by a player. The rst check allows or denies individual bets, something CrapsTable must do in collaboration with CrapsGame. For the second check, the CrapsTable assures that the total of the bets is within the table limits; something for which only the table has the information required.

170

Chapter 29. Craps Table Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Allowed Bets. The rule for allowed and non-allowed bets is relatively simple. When the game state has no point (also known as the come out roll), only Pass Line and Dont Pass bets are allowed, all other bets are not allowed. When the point is on, all bets are allowed. Well have to add an isAllowed() to CrapsGame, which CrapsTable will use when the player attempts to place a bet. Working Bets. The rule for working and non-working bets is also relatively simple. On the come out roll, all odds bets placed behind any of the six Come Point numbers are not working. This rule only applies to odds behind Come Point bets; odds behind Dont Come bets are always working. Well have to add an isWorking() to CrapsGame, which CrapsTable will use when iterating through working bets. The sequence of events that can lead to this condition is as follows. First, the player places a Come Line bet, the dice roll is 4, 5, 6, 8, 9 or 10, and establishes a point; the bet is moved to one of the six come points. Second, the player creates an additional odds bet placed behind this come point bet. Third, the main game point is a winner, changing the game state so the next roll is a come out roll. In this state, any additional odds behind a come point bet will be non-working bets on the come-out roll. This kind of exceptional subtlety is one of the important reasons why object-oriented programming can be more successful than procedural programming. In this case, we can isolate this state-specic processing to the CrapsGame. We can also provide the interface to the CrapsTable that makes this responsibility explicit and easy to use.

29.4 CrapsGame Stub

CrapsGame is a preliminary design for the game of Craps. In addition to features required by the Throw, this version includes features required by the CrapsTable class.

29.4.1 Methods
isAllowed(self, outcome) Parameter outcome (Outcome) An Outcome that may be allowed or not allowed, depending on the game state. Determines if the Outcome is allowed in the current state of the game. When the point is zero, it is the come out roll, and only Pass, Dont Pass, Come and Dont Come bets are allowed. Otherwise, all bets are allowed. isWorking(self, outcome) Parameter outcome (Outcome) An Outcome that may be allowed or not allowed, depending on the game state. Determines if the Outcome is working in the current state of the game. When the point is zero, it is the come out roll, odds bets placed behind any of the six come point numbers are not working.

29.5 CrapsTable Design

CrapsTable is a subclass of Table that has an association with a CrapsGame object. As a Table, it contains all the Bets created by the Player. It also has a betting limit, and the sum of all of a players bets must be less than or equal to this limit. We assume a single Player in the simulation.

29.4. CrapsGame Stub

171

Building Skills in Object-Oriented Design, Release 2.1.1-Python

29.5.1 Fields
game The CrapsGame used to determine if a given bet is allowed or working in a particular game state.

29.5.2 Constructors
__init__(self, game) Parameter game (CrapsGame) The CrapsGame instance that controls the state of this table Uses the superclass for initialization of the empty LinkedList of bets.

29.5.3 Methods
isValid(self, bet ) Parameter bet (Bet) The bet to validate. Validates this bet by checking with the CrapsGame to see if the bet is valid; it returns true if the bet is valid, false otherwise. allValid(self ) This uses the superclass to see if the sum of all bets is less than or equal to the table limit. If the individual bet outcomes are also valid, return true. Otherwise, return false.

29.6 Craps Table Deliverables

There are three deliverables for this exercise. A revision of the stub CrapsGame class to add methods for validating bets in dierent game states. In the stub, the point value of 0 means that only the Pass Line and Dont Pass Line bets are valid, where a point value of non-zero means all bets are valid. The CrapsTable subclass. A class which performs a unit test of the CrapsTable class. The unit test should create a couple instances of Bet, and establish that these Bets are managed by the table correctly. For testing purposes, it is easiest to have the test method simply set the the point variable in the CrapsGame instance to force a change in the game state. While public instance variables are considered by some to be a bad policy, they facilitate the creation of unit test classes.

172

Chapter 29. Craps Table Class

CHAPTER

THIRTY

CRAPSGAME CLASS
In Throw Class, we roughed out a stub version of CrapsGame that could be used to test Throw. We extended that stub in Craps Table Class. In this chapter, we will revise the game to provide the complete process for Craps. This involves a number of features, and we will have a state hierarchy as well as the Game class itself. In the process of completing the design for CrapsGame, we will uncover another subtlety of craps: winning bets and losing bets. Unlike Roulette, where a Bin contained winning Outcomes and all other Outcomes where losers, Craps includes winning Outcomes, losing Outcomes, and unresolved Outcomes. This will lead to some rework of previously created Craps classes. We can see three necessary features to the CrapsGame: Game State, Resolving Bets, Moveable Bets when a point is established. Also, we will discover some additional design features to add to other classes.

30.1 Game State

A CrapsGame object cycles through the various steps of the Craps game; this sequence is shown in Game State. For statistical sampling purposes, we dont want to process complete games, since they have an arbitrary number of dice throws, and each throw oers additional betting opportunities. Because of all the betting opportunities, we will gather data from each individual throw of the dice. Since the dice are thrown at a predictable average rate, the length of a session depends on the number of throws and has little to do with the number of games. Since we will follow the State design pattern, we have three basic design decisions. First, we have to design the state class hierarchy to own responsibilities for the unique processing of the individual states. Second, we have to design an interface for the game state objects to interact with the overall CrapsGame. Additionally, we will need to keep an object in the CrapsGame which contains the current state. Each throw of the dice will update the state, and possibly resolve game bets. To restart the game, we can create a fresh object for the initial point-o state. The following procedure provides the detailed algorithm for the game of Craps.

173

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A Single Game of Craps

Point O State. The rst throw of the dice is made with no point. This game may be resolved in a single throw of the dice, no point will be established. If a point is established the game transitions to the Point On State. 1. Place Bets. The point is o; this is the come out roll. Notify the Player to create Bets. The real work of placing bets is delegated to the Player class. Only Pass and Dont Pass bets will be allowed by the current game state. 2. Odds Bet O? Optional, for some casinos only. For any odds bets behind a come point, interrogate the player to see if the bet is on or o. 3. Come-Out Roll. Get the next throw of the Dice, giving the winning Throw, t. The Throw contains the individual Outcomes that can be resolved on this throw. 4. Resolve Proposition Bets. For each Bet, b, placed on a one-roll proposition: Proposition Winner? If Bet bs Outcome is in the winning Throw, t, then notify the Player that Bet b was a winner and update the Player s stake. Note that the odds paid for winning eld bets and horn bets depend on the Throw. Proposition Loser? If Bet bs Outcome is not in the winning Throw, t, then notify the Player that Bet b was a loser. This allows the Player to update their betting amount for the next round. 5. Natural? If the throw is a 7 or 11, this game is an immediate winner. The game state must provide the Pass Line Outcome as a winner. For each Bet, b : Come-Out Roll Natural Winner? If Bet bs Outcome is in the winning game state, then notify the Player that Bet b was a winner and update the Player s stake. A Pass Line bet is a winner, and a Dont Pass bet is a loser. Come-Out Roll Natural Loser? If Bet bs Outcome is not in the winning game state, then notify the Player that Bet b was a loser. This allows the Player to update the betting amount for the next round. A Pass Line bet is a loser, and a Dont Pass bet is a winner. 6. Craps? If the throw is a 2, 3, or 12, this game is an immediate loser. The game state must provide the Dont Pass Line Outcome as a winner; note that 12 is a push in this case, requiring special processing by the Bet or the Outcome: the bet amount is simply returned. For each Bet, b: (a) Come-Out Roll Craps Winner? If Bet b s Outcome is in the winning game state and the bet is working, then notify the Player that Bet b was a winner and update the Players stake. If the bet is not working, it is ignored. (b) Come-Out Roll Craps Loser? If Bet b s Outcome is not in the winning game state and the bet is working, then notify the Player that Bet b was a loser. If the bet is not working, it is ignored. 7. Point Established. If the throw is a 4, 5, 6, 8, 9 or 10, a point is established. The game state changes, to reect the point being on. The Pass Line and Dont Pass Line bets have a new Outcome assigned, based on the point. Point On State. While the game remains unresolved, the following steps are performed. The game is resolved when the point is made or a natural is thrown. 1. Place Bets. Notify the player to place any additional bets. The game state will allow all bets. 2. Point Roll. Get the next throw of the dice.

174

Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

3. Resolve Proposition Bets. Resolve any one-roll proposition bets. This is the procedure described above for iterating through all one-roll propositions. See Resolve Proposition Bets. 4. Natural? If the throw was 7, the game is a loser. Resolve all bets; the game state will show that all bets are active. The game state will include Dont Pass and Dont Come bets as winners, as will any of the six point bets created from Dont Pass and Dont Come Line bets. All other bets will lose, including all hardways bets. This throw resolves the game, changing the game state. The point is o. 5. Point Made? If the throw was the main game point, the game is a winner. Resolve Pass Line and Dont Pass Line bets, as well as the point and any odds behind the point. Come Point and Dont Come Point bets (and their odds) remain for the next game. A Come Line or Dont Come Line bet will be moved to the appropriate Come Point. This throw ends the game; changing the game state. The point is o; odds placed behind Come Line bets are not working for the come out roll. 6. Other Point Number? If the throw was any of the come point numbers, come bets on that point are winners. Resolve the point come bet and any odds behind the point. Also, any buy or lay bets will be resolved as if they were odds bets behind the point; recall that the buy and lay bets involved a commission, which was paid when the bet was created. 7. Hardways? For 4, 6, 8 and 10, resolve hardways bets. If the throw was made the hard way (both dice equal), a hardways bet on the thrown number is a winner. If the throw was made the easy way, a hardways bet on the thrown number is a loser. If the throw was a 7, all hardways bets are losers. Otherwise, the hardways bets remain unresolved.

30.2 Game State Class Hierarchy

We have identied some processing that is unique to each game state. Both states will have a unique list of allowed bets, a unique list of non-working bets, a unique list of throws that cause state change and resolve game bets, and throws that resolve hardways bets. In the Craps Table (Craps Table Overview ), we allocated some responsibilities to CrapsGame so that a CrapsTable could validate bets and determine which bets were working. Our further design details have shown that the work varies by state. Therefore, the methods in CrapsGame will delegate the real work to each states methods. The current stub implementation checks the value of the point variable to determine the state. Well replace this with simply calling an appropriate method of the current state object. State Responsibilities. Each CrapsGameState subclass, therefore, will have an isValid() method that implements the state-specic bet validation rules. In this case, a point-o state object only allows the two Pass Line bets: Pass Line, Dont Pass Line. The point-on state allows all bets. Additionally, weve assigned to the CrapsTable has to determine if the total amount of all a players bets meets or exceeds the table limits. Each subclass of CrapsGameState will override an isWorking() method with one that validates the statespecic rules for the working bets. In this case, a point-o state object will identify the the six odds bets placed behind the come point numbers (4, 5, 6, 8, 9 and 10) as non-working bets, and all other bets will be working bets. A point-on state object will simply identify all bets as working. The subclasses of CrapsGameState will need methods with which to collaborate with a Throw object to update the state of the CrapsGame. Changing Game State. We have identied two game states: point-o (also know as the come out roll) and point-on. We have also set aside four methods that the various Throw objects will use to change 30.2. Game State Class Hierarchy 175

Building Skills in Object-Oriented Design, Release 2.1.1-Python

the game state. The interaction between CrapsGame, the four kinds of Throws and the two subclasses of CrapsGameStates works as follows: 1. There are 36 instances of Throw, one of which was selected at random to be the current throw of the dice. The Game object calls the Throw objects updateGame() method. Each of the subclasses of Throw have dierent implementations for this method. 2. The Throw object calls one of the Games methods to change the state. There are four methods available: craps(), natural(), eleven(), and point(). Dieren subclasses of Throw will call an appropriate method for the kind of throw. 3. The Game has a current state, embodied in a CrapsGameState object. The Game will delegate each of the four state change methods (craps(), natural(), eleven(), and point()) to the current CrapsGameState object. There are two subclasses, depending on the state of the point: point-on and point-o. 4. In parallel with Game, each CrapsGameState object has four state change methods (craps(), natural(), eleven(), and point()). Each state provides dierent implementations for these methods. In eect, the two states and four methods create a kind of table that enumerates all possible state change rules. Complex? At rst glance the indirection and delegation seems like a lot of overhead for a simple state change. When we consider the kinds of decision-making this embodies, however, we can see that this is an eective solution. When one of the 36 available Throws has been chosen, the CrapsGame calls a single method to update the game state. Because the various subclasses of Throw are polymorphic, they all respond with unique, correct behavior. Similarly, each of the subclasses of Throw simply uses one of four methods to update the CrapsGame, without having to discern the current state of the CrapsGame. We can consider CrapsGame as a kind of faade over the methods of the polymorphic CrapsGameState. Our objective is to do the decision-making once when the object is created; this makes all subsequent processing free of complex decision-making (i.e., simple) but indirect. Whats important about this design is that there are no if-statements required to make it work. Instead, objects simply invoke methods.

30.3 Resolving Bets

The CrapsGame class also has the responsibility for matching the Outcomes in the current Throw with the Outcomes of the Bets on the CrapsTable. In addition to matching Outcomes in the Throw, we also have to match the Outcomes of the current game state. Finally, the CrapsGame class must also resolve hardways bets, which are casually tied to the current game state. Well look at each of these three resolution procedures in some detail before making any nal design decisions. Resolving Bets on Proposition Outcomes. Well need a bet resolution method that handles one-roll propositions. This is similar to the bet resolution in the Roulette game class. The current Throw contains a collection of Outcomes which are resolved as winners. All other Outcomes will be losers. While appropriate for the one-roll propositions, well see that this doesnt generalize for other kinds of bets. Resolving Bets on Game Outcomes. The second, and most complex bet resolution method handles game outcomes. Bets on the game as a whole have three groups of Outcomes: winners, losers and unresolved. 176 Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This unresolved outcome is fundamentally dierent from Roulette bet resolution and proposition bet resolution. Consider a Pass Line bet: in the point-o state, a roll of 7 or 11 makes this bet a winner, a roll of 2, 3 or 12 makes this bet a loser, all other numbers leave this bet unresolved. After a point is established, this Pass Line bet has the following resolutions: a roll of 7 makes this bet a loser, rolling the point makes this bet is a wiunner, all other numbers leave this bet unresolved. In addition to this three-way decision, we have the additional subtlety of Dont Pass Line bets that can lead to a fourth resolution: a push when the throw is 12 on a come out roll. We dont want to ignore this detail because it changes the odds by almost 3%. Hardways Bets. We have several choices for implementation of this multi-way decision. This is an overview, well dive into details below. We can keep separate collections of winning and losing Outcomes in each Throw. This will obligate the game to check a set winners and a set of losers for bet resolution. We can add a method to the Bet class that will return a code for the eect of a win, lose or wait for each Outcome. A win would add money to the Player; a lose would subtract money from the Player. This means that the game will have to decode this win-lose response as part of bet resolution. We can make each kind of resolution into a Command class. Each subclass of BetResolution would peform the pay a winner, collect a loser or leave unresolved procedure based on the Throw or class:CrapsGameState.

30.4 Moveable Bets

In the casino, the Come (and Dont Come) Line bets start on the given line. If a come point is established, the come line bet is moved to a numbered box. When you add behind the line odds bets, you place the chips directly on the numbered box for the Come Point number. This procedure is dierent from the Pass (and Dont Pass) Line bet. The bet is is placed on the line. If a point is established, a large white On token shows the numbered box where, in eect, the behind the line odds chips belong. Note that the net eect of both bets is identical. The pass line and behind-the-line odds bets have a payout that depends on the On token. The come line bets are moved and odds a place in a box on which the payout depends. Moving Bets. One of the things the CrapsGame does is change the Outcome of the Come and Dont Come Line bets. If a Come or Dont Come Line bet is placed and the throw is a point number (4, 5, 6, 8, 9, or 10), the bet is not resolved on the rst throw; it is moved to one of the six point number Outcomes. When designing the Bet class, in the Craps Bet Bet Overview , we recognized the need to change the Outcome from a generic Pass Line Odds to a specic point with specic odds of 2:1, 3:2 or 6:5. Well develop a moveToThrow() method that accepts a Bet and the current Throw and move that bet to an appropriate new Outcome. In addition to moving bets, we also need to create bets based on the currently established point. We also need to deactivate bets based on the established point. As an example, the Pass Line Odds and Dont Pass Odds are created after the point is established. Since the point is already known, creating these bets is best done by adding a CrapsGame.pointOutcome() method that returns an Outcome based on the current point. This allows the CrapsPlayer to get the necessary Outcome object, create a Bet and give that Bet to the CrapsTable.

30.4. Moveable Bets

177

Building Skills in Object-Oriented Design, Release 2.1.1-Python

30.5 Design Decision Win, Lose, Wait

Bet resolution in Craps is more complex than simply paying winners and collecting all other bets as losers. In craps, we have winners, losers and unresolved bets. Further, some bets have a resolution in which only the original price of the bet is returned. This is a kind of 1:1 odds special case. This leaves us with a design decision on how best to handle the wide variety of bets that will be on the CrapsTable. Problem. Whats the best way to retain a collection of Outcomes that are resolved as a mixture of winning, losing, unresolved and pushes. Note that if we elect to introduce a more complex multi-way bet resolution, we have to decide if we should reimplement the bet resolution in Roulette. Using very dierent bet resolution algorithms for Craps and Roulette would increase the cost of maintenance and adaptation. While a uniform approach is benecial, it would involve some rework of the Roulette game to make use of a more sophisticated design. Alternatives Well look at three altenative responsibility assignments in some depth. Winning and Losing Collections. We could expand the Throw to keep separate collections of winners and losers. We could expand the CrapsGameState to detail the winning and losing Outcomes for each state. All other Outcomes would be left unresolved. This is a minor revision to Dice and Throw to properly create the two groups of Outcomes. Consequently ThrowBuilder will have to be expanded to identify losing Outcomes in addition to the existing winning Outcomes. This will require the CrapsGame to make two passes through the bets. It must match all active Bets on the CrapsTable against the winning Outcomes in the current CrapsGameState; the matches are paid a winning amount and removed. It must also match match all active Bets on the CrapsTable against the losing Outcomes in the current CrapsGameState; these are removed as losers. Winning and Losing Codes, Evaluated by CrapsGame. We could enumerate three code values that represent actions to take: these actions are win, lose, and unresolved. The class Game and each subclass of GameState would have a resolution method that examines a Bet and returns the appropriate code value. This is a minor revision to Dice and Throw to properly associate a special code with each Outcome. Consequently ThrowBuilder will have to be expanded to identify losing Outcomes in addition to the existing winning Outcomes. Each GameState would also need to respond with appropriate codes. This will require the CrapsGame to make one pass through the Bets, passing each each bet to the GameState resolution method. Based on the code returned, the CrapsGame would then have an ifstatement to decide to provide bets to the Player.win() or Player.lose() method. Wining and Losing Commands. We could dene a hierarchy of three subclasses. Each subclass implements winning, losing or leaving a bet unresolved. This is a minor revision to Dice and Throw to properly associate a special object with each Outcome. We would create single objects of each resolution subclass. The ThrowBuilder will have to be expanded to associate the loser Command or winner command with each Outcome. Further, the unresolved Command would have to be associated with all Outcomes that are not resolved by the Throw or GameState. This will require the CrapsGame to make one pass through the Bets, using the associated resolution object. The resolution object would then handle winning, losing and leaving the bet unresolved. Before making a determination, well examine the remaining bet resolution issue to see if a single approach can cover single-roll, game and hardways outcomes. 178 Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Resolving Bets on Hardways Outcomes. In addition to methods to resolve one roll and game bets, we have to resolve the hardways bets. Hardways bets are similar to game bets. For Throws of 4, 6, 8 or 10 there will be one of three outcomes: when the number is made the hard way, the matching hardways bet is a winner; when the number is made the easy way, the matching hardways bet is a loser; otherwise the hardways bet is unresolved; on a roll of seven, all hardways bets are losers. Since this parallels the game rules, but applies to a Throw, it leads us to consider the design of Throw to be parallel to the design of CrapsGame. We can use either a collection of losing Outcomes in addition to the collection of winning Outcomes, or create a multi-way discrimination method, or have the Throw call appropriate methods of CrapsTable to resolve the bet. Solution. A reasonably exible design for Bet resolution that works for all three kinds of bet resolutions is to have Throw and CrapsGameState call specic bet resolution methods in CrapsPlayer. This unies one-roll, game and hardways bets into a single mechanism. It requires us to provide methods for win, lose and push in the CrapsPlayer. We can slightly simplify this to treat a push as a kind of win that returns the bet amount. Consequences. The CrapsGame will iterate through the the active Bets. Each Bet and the Player will be provided to the current Throw for resolving one-roll and hardways bets. Each Bet and the Player will also be provided to the :class:`CrapsGameState to resolve the winning and losing game bets. We can further simplify this if each Bet carries a reference to the owning Player. In this way, the Bet has all the information necessary to notify the Player. In the long run, this reects the reality of craps table where the table operators assure that each bet has an owning player.

30.6 Additional Craps Design

We will have to rework our design for Throw to have both a one-roll resolution method and a hardways resolution method. Each of these methods will accept a single active Bet. Each resolution method could use a Set of winner Outcomes and a Set of loser Outcomes to attempt to resolve the Bet. We will also need to rework our design for Dice to correctly set both winners and losers for both one-roll and harways bets when constructing the 36 individual Throw instances. We can use the following expanded algorithm for building the Dice. This is a revision to Throw Builder Overview to include lists of losing bets as well as winning bets.

Building Dice With Winning and Losing Outcomes

For All Faces Of Die 1. For all d1 , such that 1 d1 < 7: For All Faces Of A Die 2. For d2 , such that 1 d2 < 7: Sum the Dice. Compute the sum, s d1 + d2 . Craps? If s is in 2, 3, and 12, we create a CrapsThrow instance. The winning bets include one of the 2, 3 or 12 number Outcome, plus all craps, horn and eld Outcomes. The losing bets include the other number Outcomes. This throw does not resolve hardways bets. Point? For s in 4, 5, 6, 8, 9, and 10 we will create a PointThrow instance. 30.6. Additional Craps Design 179

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Hard way? When d1 = d2 , this is a hard 4, 6, 8 or 10. The appropriate hard number Outcome is a winner. Easy way? Otherwise, d1 = d2 , this is an easy 4, 6, 8 or 10. The appropriate hard number Outcome is a loser. Field? For s in 4, 9 and 10 we include the eld Outcome as a winner. Otherwise the eld Outcome is a loser. Note that 2, 3, and 12 Field outcomes where handled above under Craps. Losing Propositions. Other one-roll Outcomes, including 2, 3, 7, 11, 12, Horn and Any Craps are all losers for this Throw. Natural? If s is 7, we create a NaturalThrow instance. This will also include a 7 Outcome as a winner. It will have numbers 2, 3, 11, 12, Horn, Field and Any Craps Outcomes as losers for this Throw. Also, all four hardways are losers for this throw. Eleven? If s is 11, we create an ElevenThrow instance. This will include 11, Horn and Field Outcomes as winners. It will have numbers 2, 3, 7, 12 and Any Craps Outcomes as losers for this Throw. There is no hardways resolution. Craps Player Class Hierarchy. We have not designed the actual CrapsPlayer. This is really a complete tree of classes, each of which provides a dierent betting strategy. We will defer this design work until later. For the purposes of making the CrapsGame work, we can develop our unit tests with a kind of stub for CrapsPlayer which simply places a single Pass Line Bet. In several future exercises, well revisit this design to make more sophisticated players. See Some Betting Strategies for a further discussion on an additional player decision oered by some variant games. Our design can be expanded to cover this. Well leave this as an exercise for the more advanced student. This involves a level of collaboration between CrapsPlayer and CrapsGame that is over the top for this part. Well address this kind of very rich interaction in Blackjack.

30.7 Craps Game Implementation Overview

Changing the Throw to include both winning and losing Outcomes is an important change. Once we have xed the Throw class, we can update the ThrowBuilder class to do a correct initialization using both winners and losers. Note that we have encapsulated this information so that there is no change to Dice. We will also update Bet to carry a reference to the Player to make it easier to post winning and losing information directly to the player object. We will need to create a stub CrapsPlayer for testing purposes. We will also need to create our CrapsGameState class hierarchy to represent the two states of the game. Once the preliminary work is complete, we can then transform the CrapsGame we started in CrapsGame Stub into a nal version of CrapsGame. It will collaborate with a CrapsPlayer and maintain a correct CrapsGameState. It will be able to get a random Throw and resolve Bets.

30.8 Throw Rework

Throw is the superclass for the various throws of the dice. A Throw identies two sets of Outcomes: immediate winners and immediate losers. Each subclass is a dierent grouping of the numbers, based on the state-change rules for Craps.

180

Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

30.8.1 Fields
win1Roll A set of of one-roll Outcomes that win with this throw. lose1Roll A set of one-roll Outcomes that lose with this throw. winHardway A set of hardways Outcomes that win with this throw. Not all throws resolve hardways bets, so this and the loseHardway Set may both be empty. loseHardway A set of hardways Outcomes that lose with this throw. Not all throws resolve hardways bets, so this and the winHardway Set may both be empty. d1 One of the two die values, from 1 to 6. d2 The other of the two die values, from 1 to 6.

30.8.2 Constructors
__init__(d1, d2, winners=None, losers=None) Parameters d1 (int) One die value. d2 (int) The other die value. winners (set of Outcomes) All the outcomes which will be paid as winners for this Throw. losers All the outcomes which will be collectsd as winners for this Throw. Creates this throw, and associates the two given Set s of Outcomes that are winning one-roll propositions and losing one roll propositions.

30.8.3 Methods
add1Roll(self, winners, losers) Parameters winners (set of Outcomes) All the outcomes which will be paid as winners for this Throw. losers All the outcomes which will be collectsd as winners for this Throw. Adds outcomes to the one-roll winners and one-roll losers Sets. addHardways(self, winners, losers) Parameters winners (set of Outcomes) All the outcomes which will be paid as winners for this Throw. losers All the outcomes which will be collectsd as winners for this Throw.

30.8. Throw Rework

181

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Adds outcomes to the hardways winners and hardways losers Sets. hard(self ) Returns true if d1 is equal to d2. This helps determine if hardways bets have been won or lost. updateGame(self, game) Parameter game (CrapsGame) CrapsGame instance to be updated with the results of this throw Calls one of the Game state change methods: craps(), natural(), eleven(), point(). This may change the game state and resolve bets. resolveOneRoll(self, bet ) Parameter bet The bet to to be resolved If this Bets Outcome is in the Set of one-roll winners, pay the Player that created the Bet. Return true so that this Bet is removed. If this Bets Outcome is in the Set of one-roll losers, return true so that this Bet is removed. Otherwise, return false to leave this Bet on the table. resolveHardways(self, bet ) Parameter bet The bet to to be resolved If this Bets Outcome is in the Set of hardways winners, pay the Player that created the Bet. Return true so that this Bet is removed. If this Bets Outcome is in the Set of hardways losers, return true so that this Bet is removed. Otherwise, return false to leave this Bet on the table. __str__(self ) This should return a String representation of the dice. A form that looks like 1,2 works nicely.

30.9 ThrowBuilder Rework

ThrowBuilder initializes the 36 Throws, each initialized with the appropriate Outcomes. Subclasses can override this to reect dierent casino-specic rules for odds on Field bets.

30.9.1 Methods
buildThrows(self, dice) Para dice The Dice to build Creates the 8 one-roll Outcome instances (2, 3, 7, 11, 12, Field, Horn, Any Craps), as well as the 8 hardways Outcome instances (easy 4, hard 4, easy 6, hard 6, easy 8, hard 8, easy 10, hard 10). It then creates each of the 36 Throws, each of which has the appropriate combination of Outcomes for one-roll and hardways. The various Throws are assigned to dice.

182

Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

30.10 Bet Rework

Bet associates an amount, an Outcome and a Player. The Game may move a Bet to a dierent Outcome to reect a change in the odds used to resolve the Bet.

30.10.1 Constructors
__init__(self, amount, outcome, player=None) This replaces the existing constructor and adds an optional parameter. Parameters amount (int) The amount being wagered. outcome (Outcome) The specic outcome on which the wager is placed. player (CrapsPlayer) The player who will pay a losing bet or be paid by a winning bet. Initialize the instance variables of this bet. This works by saving the additional player information, then using the existing Bet.Bet() constructor.

30.11 CrapsPlayer Class Stub

CrapsPlayer constructs a Bet based on the Outcome named "Pass Line". This is a very persistent player.

30.11.1 Fields
passLine This is the Outcome on which this player focuses their betting. It will be an instance of the "Pass Line" Outcome, with 1:1 odds. workingBet This is the current Pass Line Bet. Initially this is None. Each time the bet is resolved, this is reset to None. This assures that only one bet is working at a time. table That Table which collects all bets.

30.11.2 Constructors
__init__(self, table) Parameter table (Table) The Table. Constructs the CrapsPlayer with a specic table for placing bets. The player creates a single "Pass Line" Outcome, which is saved in the passLine variable for use in creating Bets.

30.10. Bet Rework

183

Building Skills in Object-Oriented Design, Release 2.1.1-Python

30.11.3 Methods
If workingBet is null, create a new Pass Line Bet, and use Table placeBet() to place that bet. If workingBet is not null, the bet is still working. Do not place any more bets. Notication from the Game that the Bet was a winner. The amount of money won is available via theBet winAmount(). Notication from the Game that the Bet was a loser.

30.12 CrapsGameState Class

CrapsGameState denes the state-specic behavior of a Craps game. Individual subclasses provide methods used by CrapsTable to validate bets and determine the active bets. Subclasses provide state-specic methods used by a Throw to possibly change the state and resolve bets.

30.12.1 Fields
game The overall CrapsGame for which this is a specic state. From this object, the various next state-change methods can get the CrapsTable and an Iterator over the active Bets.

30.12.2 Constructors
__init__(self, game) Parameter game (Game) The game to which this state applies Saves the overall CrapsGame object to which this state applies.

30.12.3 Methods
isValid(self, outcome) Parameter outcome (Outcome) The outcome to be tested for validity Returns true if this is a valid outcome for creating bets in the current game state. Each subclass provides a unique denition of valid bets for their game state. isWorking(self, outcome) Parameter outcome (Outcome) The outcome to be tested for if its working Returns true if this is a working outcome for existing bets in the current game state. Each subclass provides a unique denition of active bets for their game state. craps(self, throw ) Parameter throw (Throw) The throw that is associated with craps. Return an appropriate state when a 2, 3 or 12 is rolled. It then resolves any game bets. Each subclass provides a unique denition of what new state and what bet resolution happens. natural(self, throw )

184

Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Parameter throw (Throw) The throw that is associated with a natural seven. Returns an appropriate state when a 7 is rolled. It then resolves any game bets. Each subclass provides a unique denition of what new state and what bet resolution happens. eleven(self, throw ) Parameter throw (Throw) The throw that is associated an eleven. Returns an appropriate state when an 11 is rolled. It then resolves any game bets. Each subclass provides a unique denition of what new state and what bet resolution happens. point(self, throw ) Parameter throw (Throw) The throw that is associated with a point number. Returns an appropriate state when the given point number is rolled. It then resolves any game bets. Each subclass provides a unique denition of what new state and what bet resolution happens. pointOutcome(self ) Returns the Outcome based on the current point. This is used to create Pass Line Odds or Dont Pass Odds bets. This delegates the real work to the current CrapsGameState object. moveToThrow(self, bet, throw ) Parameters bet (Bet) The Bet to update based on the current Throw throw (Throw) The Throw to which the outcome is changed Moves a Come Line or Dont Come Line bet to a new Outcome based on the current throw. If the value of theThrow is 4, 5, 6, 8, 9 or 10, this delegates the move to the current CrapsGameState object. For values of 4 and 10, the odds are 2:1. For values of 5 and 9, the odds are 3:2. For values of 6 and 8, the odds are 6:5. For other values of theThrow, this method does nothing. __str__(self ) In the superclass, this doesnt do anything. Each subclass, however, should display something useful.

30.13 CrapsGamePointO Class

CrapsGamePointOff denes the behavior of the Craps game when the point is o. It denes the allowed bets and the active bets. It provides methods used by a Throw to change the state and resolve bets. All four of the game update methods (craps, natural, eleven and point) use the same basic algorithm. The method will get the CrapsTable from theGame. From the CrapsTable, the method gets the Iterator over the Bets. It can then match each Bet against the various Outcomes which win and lose, and resolve the bets.

30.13.1 Constructors
__init__(self, game) Parameter game (CrapsGame) The game to which this state applies. Uses the superclass constructor to save the overall CrapsGame object.

30.13. CrapsGamePointO Class

185

Building Skills in Object-Oriented Design, Release 2.1.1-Python

30.13.2 Methods
isValid(self, outcome) Parameter outcome (Outcome) The outcome to be tested for validity There are two valid Outcomes: Pass Line, Dont Pass Line. All other Outcomes are invalid. isWorking(self, outcome) Parameter outcome (Outcome) The outcome to be tested to see if its working There are six non-working Outcomes: Come Odds 4, Come Odds 5, Come Odds 6, Come Odds 8, Come Odds 9 and Come Odds 10. All other Outcomes are working. craps(self, throw ) Parameter throw (Throw) The throw that is associated with craps. When the point is o, a roll of 2, 3 or 12 means the game is an immediate loser. The Pass Line Outcome is a loset. If the Throw value is 12, a Dont Pass Line Outcome is a push, otherwise the Dont Pass Line Outcome is a winner. The next state is the same as this state, and the method should return this. natural(self, throw ) Parameter throw (Throw) The throw that is associated with a natural seven. When the point is o, 7 means the game is an immediate winner. The Pass Line Outcome is a winner, the Dont Pass Line Outcome is a loser. The next state is the same as this state, and the method should return this. eleven(self, throw ) Parameter throw (Throw) The throw that is associated an eleven. When the point is o, 11 means the game is an immediate winner. The Pass Line Outcome is a winner, the Dont Pass Line Outcome is a loser. The next state is the same as this state, and the method should return this. point(self, throw ) Parameter throw (Throw) The throw that is associated with a point number. When the point is o, a new point is established. This method should return a new instance of CrapsGamePointOn created with the given Throws value. Note that any Come Point bets or Dont Come Point bets that may be on this point are pushed to player: they cant be legal bets in the next game state. pointOutcome(self ) Returns the Outcome based on the current point. This is used to create Pass Line Odds or Dont Pass Odds bets. This delegates the real work to the current CrapsGameState object. Since no point has been established, this returns null. __str__(self ) The point-o state should simply report that the point is o, or that this is the come out roll.

30.14 CrapsGamePointOn Class

CrapsGamePointOn denes the behavior of the Craps game when the point is on. It denes the allowed bets and the active bets. It provides methods used by a Throw to change the state and resolve bets. 186 Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

30.14.1 Fields
point The point value.

30.14.2 Constructors
__init__(self, point, game) Saves the given point value. Uses the superclass constructor to save the overall CrapsGame object.

30.14.3 Methods
isValid(self, outcome) Parameter outcome (Outcome) The outcome to be tested for validity It is invalid to Buy or Lay the Outcomes that match the point. If the point is 6, for example, it is invalid to buy the Come Point 6 Outcome. All other Outcomes are valid. isWorking(self, outcome) Parameter outcome (Outcome) The outcome to be tested to see if its working All Outcomes are working. craps(self, throw ) Parameter throw (Throw) The throw that is associated with craps. When the point is on, 2, 3 and 12 do not change the game state. The Come Line Outcome is a loser, the Dont Come Line Outcome is a winner. The next state is the same as this state, and the method should return this. natural(self, outcome) Parameter throw (Throw) The throw that is associated with a natural seven. When the point is on, 7 means the game is a loss. Pass Line Outcomes lose, as do the pass-line odds Outcome s based on the point. Dont Pass Line Outcomes win, as do all Dont Pass odds Outcome based on the point. The Come Line Outcome is a winner, the Dont Come Line Outcome is a loser. However, all Come Point number Outcomes and Come Point Number odds Outcome are all losers. All Dont Come Point number Outcomes and Dont Come Point odds Outcomes are all winners. The next state is a new instance of the CrapsGamePointOff state. Also note that the Throw of 7 also resolved all hardways bets. A consequence of this is that all Bets on the CrapsTable are resolved. eleven(self, throw ) Parameter throw (Throw) The throw that is associated an eleven. When the point is on, 11 does not change the game state. The Come Line Outcome is a winner, and the Dont Come Line Outcome is a loser. The next state is the same as this state, and the method should return this. point(self, throw ) Parameter throw (Throw) The throw that is associated with a point number.

30.14. CrapsGamePointOn Class

187

Building Skills in Object-Oriented Design, Release 2.1.1-Python

When the point is on and the value of throw doesnt match point, then the various Come Line bets can be resolved. Come Point Outcome s for this number (and their odds) are winners. Dont Come Line Outcome s for this number (and their odds) are losers. Other Come Point number and Dont Come Point numbers remain, unresolved. Any Come Line bets are moved to the Come Point number Outcomes. For example, a throw of 6 moves the Outcome of the Come Line Bet to Come Point 6. Dont Come Line bets are moved to be Dont Come number Outcomes. The method should return this. When the point is on and the value of throw matches point, the game is a winner. Pass Line Outcomes are all winners, as are the behind the line odds Outcomes. Dont Pass line Outcomes are all losers, as are the Dont Pass Odds Outcomes. Come Line bets are moved to thee Come Point number Outcomes. Dont Come Line bets are moved to be Dont Come number Outcomes. The next state is a new instance of the CrapsGamePointOff state. pointOutcome(self ) Returns the Outcome based on the current point. This is used to create Pass Line Odds or Dont Pass Odds bets. This delegates the real work to the current CrapsGameState object. For points of 4 and 10, the Outcome odds are 2:1. For points of 5 and 9, the odds are 3:2. For points of 6 and 8, the odds are 6:5. __str__(self ) The point-o state should simply report that the point is o, or that this is the come out roll.

30.15 CrapsGame Class

CrapsGame manages the sequence of actions that denes the game of Craps. This includes notifying the Player to place bets, throwing the Dice and resolving the Bet s actually present on the Table. Note that a single cycle of play is one throw of the dice, not a complete craps game. The state of the game may or may not change.

30.15.1 Fields
dice Contains the dice that returns a randomly selected Throw with winning and losing Outcomes. This is an instance of Dice. table The CrapsTable contains the bets placed by the player. player The CrapsPlayer who places bets on the CrapsTable.

30.15.2 Constructors
We based this constructor on an design that allows any of these objects to be replaced. This is the Strategy design pattern. Each of these objects is a replaceable strategy, and can be changed by the client that uses this game. Additionally, we specically do not include the Player instance in the constructor. The Game exists independently of any particular Player, and we defer binding the Player and Game until we are gathering statistical samples. __init__(self, dice, table) 188 Chapter 30. CrapsGame Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Parameters dice (Dice) The dice to use table The table to use for collecting bets CrapsTable Constructs a new CrapsGame, using a given Dice and CrapsTable. The player is not dened at this time, since we may want to run several simulations with dierent players.

30.15.3 Methods
__init__(self, player ) Parameter player (CrapsPlayer) The player who will place bets on this game This will execute a single cycle of play with a given Player. 1. It will call thePlayer placeBets() to get bets. It will validate the bets, both individually, based on the game state, and collectively to see that the table limits are met. 2. It will call theDice next() to get the next winning Throw. 3. It will use the Throws updateGame() to advance the game state. 4. It will then call theTable bets() to get an Iterator ; stepping through this Iterator returns the individual Bet objects. It will use the Throws resolveOneRoll() method to check one-roll propositions. If the method returns true, the Bet is resolved and should be deleted. It will use the Throws resolveHardways() method to check the hardways bets. If the method returns true, the Bet is resolved and should be deleted. pointOutcome(self ) Returns the Outcome based on the current point. This is used to create Pass Line Odds or Dont Pass Odds bets. This delegates the real work to the current CrapsGameState object. moveToThrow(self, bet, throw ) Parameters bet (Bet) The Bet to move based on the current throw throw (Throw) The Throw to which to move the Bets Outcome Moves a Come Line or Dont Come Line bet to a new Outcome based on the current throw. This delegates the move to the current CrapsGameState object. This method should just as a precaution assert that the value of theThrow is 4, 5, 6, 8, 9 or 10. These point values indicate that a Line bet can be moved. For other values of theThrow, this method should raise an exception, since theres no reason for attempting to move a line bet on anything but a point throw. reset(self ) This will reset the game by setting the state to a new instance of GamePointOff. It will also tell the table to clear all bets.

30.15. CrapsGame Class

189

Building Skills in Object-Oriented Design, Release 2.1.1-Python

30.16 Craps Game Deliverables

There are over a dozen deliverables for this exercise. This includes signicant rework for Throw and Dice. It also includes development of a stub CrapsPlayer, the CrapsGameState hierarchy and the rst version of the CrapsGame. We will break the deliverables down into two groups. Rework. The rst group of deliverables includes the rework for Throw and Dice, and all of the associated unit testing. The revised and expanded Throw class. This will ripple through the constructors for all four subclasses, NaturalThrow, CrapsThrow, ElevenThrow, PointThrow. Five updated unit tests for the classes in the Throw class hierarchy. This will conrm the new functionality for holding winning as well as losing Outcomes. The revised and expanded ThrowBuilder. This will construct Throws with winning as well as losing Outcomes. A unit test for the Dice class that conrms the new initializer that creates winning as well as losing Outcomes. New Development. The second group of deliverables includes development of a stub CrapsPlayer, the CrapsGameState hierarchy and the rst version of the CrapsGame. This also includes signicant unit testing. The CrapsPlayer class stub. We will rework this design later. This class places a bet on the Pass Line when there is no Pass Line Bet on the table. One consequence of this is that the player will be given some opportunities to place bets, but will decline. Since this is simply used to test CrapsGame, it doesnt deserve a very sophisticated unit test of its own. It will be replaced in a future exercise. A revised Bet, which carries a reference to the Player who created the Bet. This will ripple through all subclasses of Player, forcing them to all add the this parameter when constructing a new Bet. The CrapsGame class. A class which performs a demonstration of the CrapsGame class. This demo program creates the Dice, the stub CrapsPlayer and the CrapsTable. It creates the CrapsGame object and cycles a few times. Note that we will need to congure the Dice to return non-random results. We could, with some care, refactor our design to create some common superclasses between Roulette and Craps to extract features of Throw so they can be shared by Throw and Bin. Similarly, there may be more common features between RouletteGame and CrapsGame. Well leave that as an exercise for more advanced students.

30.17 Optional Working Bets

Some casinos may give the player an option to declare the odds bet behind a come point as on or o. This is should not be particularly complex to implement. There are a number of simple changes required if we want to add this interaction between CrapsPlayer and CrapsGame. 1. We must add a method to the CrapsPlayer to respond to a query from the CrapsGame that determines if the player wants their come point odds bet on or o. 2. We need to update Bet to store the Player who created the Bet. Third, the CrapsGame gets the relevant Bets from the Table, and interrogates the Player for the disposition of the Bet.

190

Chapter 30. CrapsGame Class

CHAPTER

THIRTYONE

CRAPSPLAYER CLASS
The variations on Player, all of which reect dierent betting strategies, is the heart of this application. In Roulette Game Class, we roughed out a stub class for Player, and rened it in Player Class. We will further rene this denition of Player for use in Craps.

31.1 Craps Player Overview

We have now built enough infrastructure that we can begin to add a variety of players and see how their betting strategies work. Each player is betting algorithm that we will evaluate by looking at the players stake to see how much they win, and when they stop playing because theyve run out of time or gone broke. The Player has the responsibility to create bets and manage the amount of their stake. To create bets, the player must create legal bets from known Outcome s and stay within table limits. To manage their stake, the player must deduct the price of a bet when it is created, accept winnings or pushes, report on the current value of the stake, and leave the table when they are out of money. We have an interface that was roughed out as part of the design of CrapsGame and CrapsTable. In designing CrapsGame, we put a placeBets() method in CrapsPlayer to place all bets. We expected the CrapsPlayer to create Bet s and use the placeBet() method of CrapsTable class to save all of the individual Bets. In an earlier exercise, we built a stub version of CrapsPlayer in order to test Game. See CrapsPlayer Class Stub. When we nish creating the nal superclass, CrapsPlayer, we will also revise our CrapsPlayerStub to be more complete, and rerun our unit tests to be sure that our more expanded design still handles the basic test cases correctly. Our objective is to have a new abstract class, CrapsPlayer, with a concrete subclass that follows the Martingale system, using simple Pass Line bets and behind the line odds bets. Well defer some of the design required to collect detailed measurements for statistical analysis. In this rst release, well simply place bets. Most of the Simulator class that we built for Roulette should be applicable to Craps without signicant modication. Some Basic Features. Our basic CrapsPlayer will place a Pass Line bet and a Pass Line Odds bet. This requires the player to interact with the CrapsTable or the CrapsGame to place bets legally. On a come out roll, only the Pass Line will be legal. After that, a single Pass Line Odds bet can be placed. This leads to three betting rules: Come Out Roll. Condition: No Pass Line Bet is currently placed and only the Pass Line bet is legal. Action: Place a Pass Line bet. First Point Roll. Condition: No odds bets is currently placed and odds bets are legal. Action: Place a Pass Line Odds bet. Other Point Roll. Condition: An odds bets is currently placed. Action: Do Nothing. 191

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Beyond simply placing Pass Line and Pass Line Odds bets, we can use a Martingale or a Cancellation betting system to increase our bet on each loss, and decrease our betting amount on a win. Since we have two dierent bets in play a single bet created on the come out roll, a second odds bet if possible the simple Martingale system doesnt work well. In some casinos, the behind the line odds bet can be double the pass line bet, or even 10 times the pass line bet, giving us some complex betting strategies. For example, we could apply the Martingale system only to the odds bet, leaving the pass line bet at the table minimum. Well set this complexity aside for the moment, build a simple player rst.

31.2 CrapsPlayer Superclass Design

CrapsPlayer is a subclass of Player and places bets in Craps. This an abstract class, with no actual body for the Player.placeBets() method. However, this class does implement the basic win() and lose() methods used by all subclasses. Since this is a subclass of the basic Roulette player, we inherit several useful features. Most of the features of Player are repeated here for reference purposes only.

31.2.1 Fields
stake The players current stake. Initialized to the players starting budget. roundsToGo The number of rounds left to play. Initialized by the overall simulation control to the maximum number of rounds to play. In Roulette, this is spins. In Craps, this is the number of throws of the dice, which may be a large number of quick games or a small number of long-running games. In Craps, this is the number of cards played, which may be large number of hands or small number of multi-card hands. table The CrapsTable used to place individual Bets.

31.2.2 Constructors
__init__(self, table) Parameter table (CrapsTable) The table Constructs the Player with a specic CrapsTable for placing Bets.

31.2.3 Methods
playing(self ) Returns true while the player is still active. A player with a stake of zero will be inactive. Because of the indenite duration of a craps game, a player will only become inactive after their roundsToGo is zero and they have no more active bets. This method, then, must check the CrapsTable to see when all the bets are fully resolved. Additionally, the players betting rules should stop placing new bets when the roundsToGo is zero. placeBets(self )

192

Chapter 31. CrapsPlayer Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Updates the CrapsTable with the various Bet s. When designing the CrapsTable, we decided that we needed to deduct the price of the bet from the stake when the bet is created. See the Roulette Table Roulette Table Overview for more information on the timing of this deduction, and the Craps Bet Bet Overview for more information on the price of a bet. win(self, bet ) Parameter bet (Bet) that was a winner Notication from the CrapsGame that the Bet was a winner. The amount of money won is available via theBet winAmount(). lose(self, bet ) Parameter bet (Bet) that was a loser Notication from the CrapsGame that the Bet was a loser.

31.3 CrapsPlayerPass Subclass

CrapsPlayerPass is a CrapsPlayer who places a Pass Line bet in Craps.

31.3.1 Methods
placeBets(self ) If no Pass Line bet is present, this will update the Table with a bet on the Pass Line at the base bet amount. Otherwise, this method does not place an additional bet.

31.4 Craps Martingale Subclass

CrapsMartingale is a CrapsPlayer who places bets in Craps. This player doubles their Pass Line Odds bet on every loss and resets their Pass Line Odds bet to a base amount on each win.

31.4.1 Fields
lossCount The number of losses. This is the number of times to double the pass line odds bet. betMultiple The the bet multiplier, based on the number of losses. This starts at 1, and is reset to 1 on each win. It is doubled in each loss. This is always set so that betM ultiple = 2lossCount .

31.4.2 Methods
placeBets(self )

31.3. CrapsPlayerPass Subclass

193

Building Skills in Object-Oriented Design, Release 2.1.1-Python

If no Pass Line bet is present, this will update the Table with a bet on the Pass Line at the base bet amount. If no Pass Line Odds bet is present, this will update the Table with an Pass Line Odds bet. The amount is the base amount times the betMultiple. Otherwise, this method does not place an additional bet. win(self, bet ) Parameter bet (Bet) that was a winner Uses the superclass win() method to update the stake with an amount won. This method then resets lossCount to zero, and resets betMultiple to 1. lose(self, bet ) Parameter bet (Bet) that was a loser Increments lossCount by 1 and doubles betMultiple.

31.5 Craps Player Deliverables

There are six deliverables for this exercise. The CrapsPlayer abstract superclass. Since this class doesnt have a body for the placeBets() method, it cant be unit tested directly. A CrapsPlayerPass class that is a proper subclass of CrapsPlayer, but simply places bets on Pass Line until the stake is exhausted. A unit test class for CrapsPlayerPass. This test should synthesize a xed list of Outcomes, Throw s, and calls a CrapsPlayerPass instance with various sequences of craps, naturals and points to assure that the pass line bet is made appropriately. The CrapsMartingale subclass of CrapsPlayer. A unit test class for CrapsMartingale. This test should synthesize a xed list of Outcomes, Throw s, and calls a CrapsMartingale instance with various sequences of craps, naturals and points to assure that the bet doubles appropriately on each loss, and is reset on each win. The unit test class for the CrapsGame class should still work with the new CrapsPlayerPass. Using a non-random generator for Dice, this should be able to conrm correct operation of the CrapsGame for a number of bets.

194

Chapter 31. CrapsPlayer Class

CHAPTER

THIRTYTWO

DESIGN CLEANUP AND REFACTORING

We have taken an intentionally casual approach to the names chosen for our various classes and the relationships among those classes. At this point, we have a considerable amount of functionality, but it doesnt reect our overall purpose, instead it reects the history of its evolution. This chapter will review the design from Craps and more cleanly separate it from the design for Roulette. We expect two benets from the rework in this chapter. First, we expect the design to become simpler in the sense that Craps is separated from Roulette, and this will give us room to insert Blackjack into the structure with less disruption in the future. Second, and more important, the class names will more precisely reect the purpose of the class, making it easier to understand the system, which means it will be easier to debug, maintain and adapt.

32.1 Design Review

We can now use our application to generate some more usable results. We would like the Simulator class to be able to use our Craps game, dice, table and players in the same way that we use our Roulette game, wheel, table and players. The idea would be to give the Simulators constructor Craps-related objects instead of Roulette-related objects and have everything else work normally. Since we have generally made Craps a subclass of Roulette, we are reasonably condent that this should work. Our Simulators constructor requires a Game and a Player. Since CrapsGame is a subclass of Game and CrapsPlayer is a subclass of Player, we can construct an instance of Simulator. Looking at this, however, we nd a serious problem with the names of our classes and their relationships. When we designed Roulette, we used generic names like Table, Game and Player unwisely. Further, theres no reason for Craps to be dependent on Roulette. We would like them to be siblings, and both children of some abstract game simulation.

195

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Soapbox On Refactoring We feel very strongly that our :design by refactoring helps beginning designers produce a more functional design more quickly than the alternative approach, which is to attempt to dene the game abstraction and player abstraction rst and then specialize the various games. When dening an abstract superclass, some designers will build a quick and dirty design for some of the subclasses, and use this to establish the features that belong in the superclass. We nd that a more successful superclass design comes from have more than one working subclasses and a clear understanding of the kinds of extensions that are likely to emerge from the problem domain. While our approach of refactoring working code seems expensive, the total eort is often smaller. The largest impediment weve seen seems to stem from the project-management mythology that once something passes unit tests it is done for ever and can be checked o as completed. We feel that it is very important to recognized that passing a unit test is only one of many milestones. Passing an integration test, and passing the sanity test are better indicators of done-ness. The sanity test is the designers ability to explain the class structure to someone new to the project. We feel that class and package names must make obvious sense in the current project context. We nd that any explanation of a class name that involves the words historically, or originally means that there are more serious design deciencies that need to be repaired. We now know enough to factor out the common features of Game and CrapsGame to create three new classes from these two. However, to nd the common features of these two classes, well see that we have to unify Dice and Wheel, as well as Table and CrapsTable and Player and CrapsPlayer. Looking into Dice and Wheel, we see that well have to tackle Bin and Throw rst. Unifying Bin and Throw is covered in Design Heavy. If the necessary design for RandomEvent is already in place, skip to Unifying Dice and Wheel . After that, we can refactor the other classes.

32.1.1 Unifying Bin and Throw

We need to create a common superclass for Bin and Throw, so that we can then create some commonality between Dice and Wheel. This unication will make the Vector of results and the next() method identical. The rst step, then, is to identify what are the common features of Bin and Throw. The relatively simple Bin and the more complex Throw can be unied in one of two ways. 1. Use Throw as the superclass. A Roulette Bin doesnt need a specic list of losing Outcomes. Indeed, we dont even need a subclass, since a Roulette Bin can just ignore features of a Craps Throw. 2. Create a superclass based on Bin, make Bin a subclass and change Throw to add features to the new superclass. The rst design approach is something we call the Swiss Army Knife design pattern: create a structure that has every possible feature, and then ignore the features you dont need. This creates a distasteful disconnect between the use of Bin and the declaration of Bin: we only use a the Set of winning Outcomes, but the object has the losing Set that isnt used by anything else in the Roulette game. We also note that a key feature of OO languages is inheritance, which adds features to a superclass. The Swiss Army Knife design approach, however, works by subtracting features. Since the languages were using dont really support class denition by subtraction, we try to ignore the features, creating a distance between the OO language and our design intent. Our rst decision, then, is to refactor Throw and Bin to make them instances of a common superclass, which well call RandomEvent. See the Craps Throw Throw Overview for our initial thoughts on this, echoed in the Soapbox on Refectoring sidebar. The responsibilities for RandomEvent are essentially the same as Bin. We can then make Bin a subclass that doesnt add any new features, and Throw a subclass that adds a number of features, including the value of 196 Chapter 32. Design Cleanup and Refactoring

Building Skills in Object-Oriented Design, Release 2.1.1-Python

the two dice and the Set of losing Outcome s. Note that we have made Throw and Bin siblings of a common superclass. See Soapbox on Architecture for more information on our preference for this kind of design.

32.1.2 Unifying Dice and Wheel

When we take a step back from Dice and Wheel, we see that they are nearly identical. They dier in the construction of the Bins or Throws, but little else. Looking forward, the deck of cards used for Blackjack is completely dierent. Dice and a Roulette wheel use :phrase:selection with replacement : an event is picked at random from a pool, and is eligible to be picked again any number of timers. Cards, on the other hand, are selection without replacement : the cards form a sequence of events of a dened length that is randomized by a shue. If we have a 5-deck shoe, we can only see ve kings of spades during the game, and we only have 260 cards. However, we can roll an indenite number of 7s on the dice. We note that there is a supercial similarity between the rather complex BinBuilder methods and the simpler method in ThrowBuilder. However, there is no compelling reason for polymorphism between these two classes. We dont have to factor these into a common class hierarchy. Our second design decision, then, is to create a RandomEventFactory out of Dice and Wheel . This refactoring will make the Vector of results and the next() method part of the superclass. Each subclass provides the initialization method that constructs the Vector of RandomEvents. When we move on to tackle cards, well have to create a subclass that uses a dierent denition of next() and adds shuffle(). This will allow a deck of cards to do selection without replacement.

32.1.3 Refactoring Table and CrapsTable

We see few dierences between Table and CrapsTable. When we designed CrapsTable we had to add a relationship between CrapsTable and CrapsGame so that a table could ask the game to validate individual Bets based on the state of the game. If we elevate the CrapsTable to be the superclass, we eliminate a need to have separate classes for Craps and Roulette. We are dangerously close to embracing a Swiss Army Knife design. The distinction may appear to be merely a matter of degree: one or two features can be pushed up to the superclass and then ignored in a subclass. However, in this case, both Craps and Roulette will use the Game to validate bets: the feature will not be ignored. It happens that the Roulette Game will permit all bets, but we have made that the Games responsibility, not the Tables. Indeed, viewed this way, the Roulette version of Table implicitly took a responsibility away from the Roulette Game because the Table failed to collaborate with the Game for any game-state specic rules. At the time, we overlooked this nuance because we knew that Roulette was stateless and we were comfortable making that assumption part of the design. Our third design decision is to merge Table and CrapsTable into a new Table class and use this for both games. This will simplify the various Game classes by using a single class of Table for both games.

32.1.4 Refactoring Player and CrapsPlayer

Before we can nally refactor Game, we need to be sure that we have sorted out a proper relationship between our various players. In this case, we have a large hierarchy, which will we hope to make far larger as we explore dierent betting alternatives. Indeed, the central feature of this simulation is to expand the hierarchy of players as needed to explore betting strategies. Therefore, time spent organizing the Player hierarchy is time well spent. Wed like to have the following hierarchy. Player.

32.1. Design Review

197

Building Skills in Object-Oriented Design, Release 2.1.1-Python

RoulettePlayer. * RouletteMartingale. * RouletteRandom. * RouletteSevenReds. * Roulette1326. * RouletteCancellation. * RouletteFibonacci. CrapsPlayer. * CrapsMartingale. Looking forward to Blackjack, see see that there is much richer player interaction, because there are player decisions that are not related to betting. This class hierarchy seems to enable that kind of expansion. We note that there are two dimensions to this class hierarchy. One dimension is the game (Craps or Roulette), the other dimension is a betting system (Matingale, 1326, Cancellation, Fibonacci). For Blackjack, there is also a playing system in addition to a betting system. Sometimes this multi-dimensionsal aspect of a class hierarchy indicates that we could be using multiple inheritance. In the case of Python, we have multiple inheritance in the language, and we can pursue this directly. Java, however, demands the Strategy design pattern to add a exible betting strategy object to the basic interface for playing the game. In Roulette, where we are placing a single bet, there is almost no distinction between game interface and the betting system. However, in Craps, we made a distinction in our Martingale player by separating their Pass Line bet (where the payout doesnt match the actual odds very well) from their Pass Line Odds bet (where the payout does match the odds). This means that our Martingale Craps player really has two betting strategies objects: a at bet strategy for Pass Line and a Martingale Bet strategy for the Pass Line Odds. If we separate the player and the betting system, we could easily mix and match betting systems, playing systems and game rules. In the case of Craps, where we can have many working bets (Pass Line, Come Point Bets, Hardways Bets, plus Propostions), each player would have a mixture of betting strategies used for their unique mixture of working bets. This leads to an interesting issue in the composition of such a complex object. For the current exercise, however, we wont formally separate the player from the various betting strategies. Rather than fully separate the players game interface and betting system interface, we can simply adjust the class hierarchy and the class names to those shown above. We need to make the superclass, Player independent of any game. We can do this by extracting anything Roulette-specic from the original Player class and renaming our Roulette-focused Passenger57 to be RoulettePlayer, and x all the Roulette player subclasses to inherit from RoulettePlayer. We will encounter one design diiculty when doing this. That is the dependency from the various Player1326State classes on a eld of Player1326. Currently, we will simply be renaming Player1326 to Roulette1326. However, as we go forward, we will see how this small issue will become a larger problem. In Python, we can easily overlook this, as described in Python and Interface Design.

198

Chapter 32. Design Cleanup and Refactoring

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Python and Interface Design Because of the run-time binding done in Python, there is no potential problem in having the Player1326State classes depend on a eld of Player1326 . Since Java checks the validity of these references at compile time, we are forced to provide proper declarations for the outcome eld in Player1326 . In the event of change, this declaration may no longer be valid, alterting us to a dependency. In Python, however, the attributes of an object are kept in a dictionary, which is created dynamically during execution, making it diicult to assure that one class properly includes the attributes that will be required by a collaborating class. We dont discover problems in Python until the application crashes at run time because of a missing attribute. Some Python programmers will bypass the simple access methods and check an objects internal __dict__ for the presence of the expected outcome attribute. We dont condone this: the application should crash because of design changes, not work around them. While the dynamic nature of Python attributes has some benecial eects, we nd it can be easily abused. We consider references to the __dict__ object to raise serious quality concerns, and discourage its use by new OO designers.

32.1.5 Refactoring Game and CrapsGame

Once we have a common RandomEventFactory, a common Table, and a common Player, we can separate Game from RouletteGame and CrapsGame to create three new classes. The abstract superclass, Game. This will contain a RandomEventFactory, a Table and have the necessary interface to reset the game and execute one cycle of play. This class is based on the existing Game, with the Roulette-specic cycle() replaced with an abstract method denition. The concrete subclass, RouletteGame. This has the cycle() method appropriate to Roulette that was extracted from the original Game class. The concrete subclass, CrapsGame. This has the cycle() method appropriate to Craps. This is a small change to the parent of the CrapsGame class. While this appears to be a tremendous amount of rework, it reects lessons learned incrementally through the previous chapters of exercises. This refactoring is based on considerations that would have been challenging, perhaps impossible, to explain from the outset. Since we have working unit tests for each class, this refactoring is easily validated by rerunning the existing suite of tests.

32.2 RandomEventFactory Design

RandomEventFactory is a superclass for Dice, Wheel, Cards, and other casino random-event generators.

32.2.1 Fields
rng The random number generator, a subclass of random.Random. Generates the next random number, used to select a RandomEvent from the bins collection. current The most recently returned RandomEvent.

32.2. RandomEventFactory Design

199

Building Skills in Object-Oriented Design, Release 2.1.1-Python

32.2.2 Constructors
__init__(self, rng) Saves the given Random Number Generator. Calls the initialize() method to create the vector of results.

32.2.3 Methods
initialize(self ) Create a collection of RandomEvent objects with the pool of possible results. Each subclass must provide a unique implementation for this. next(self ) Return the next RandomEvent. Each subclass must provide a unique implementation for this.

32.3 Wheel Class Design

Wheel is a subclass of RandomEventFactory that contains the 38 individual Bins on a Roulette wheel. As a RandomEventFactory, it contains a random number generator and can select a Bin at random, simulating a spin of the Roulette wheel.

32.3.1 Constructors
__init__(self, rng) Creates a new wheel. Create a sequence of the Wheel.events with with 38 empty Bins. Use the superclass to save the given random number generator instance and invoke initialize().

32.3.2 Methods
addOutcome(self, bin, outcome) Adds the given Outcome to the Bin with the given number. initialize(self ) Creates the events vector with the pool of possible events. This will create an instance of BinBuilder, bb, and delegate the construction to the buildBins() method of the bb object.

32.3.3 Dice Class

Dice is a subclass of RandomEventFactory that contains the 36 individual throws of two dice. As a RandomEventFactory, it contains a random number generator and can select a Throw at random, simulating a throw of the Craps dice.

200

Chapter 32. Design Cleanup and Refactoring

Building Skills in Object-Oriented Design, Release 2.1.1-Python

32.3.4 Constructors
__init__(self, rng) Create an empty set of Dice.events. Use the superclass to save the given random number generator instance and invoke initialize().

32.3.5 Methods
addOutcome(self, faces, outcome) Adds the given Outcome to the Throw with the given NumberPair. This allows us to create a collection of several one-roll Outcomes. For example, a throw of 3 includes four one-roll Outcomes: Field, 3, any Craps, and Horn. initialize(self ) Creates the 8 one-roll Outcome instances (2, 3, 7, 11, 12, Field, Horn, Any Craps). It then creates the 36 Throw s, each of which has the appropriate combination of Outcomes.

32.4 Table Class

Table contains all the Bets created by the Player. A table has an association with a Game, which is responsible for validating individual bets. A table also has betting limits, and the sum of all of a players bets must be within this limits.

32.4.1 Fields
minimum This is the table lower limit. The sum of a Players bets must be greater than or equal to this limit. maximum This is the table upper limit. The sum of a Players bets must be less than or equal to this limit. bets This is a LinkedList of the Bets currently active. These will result in either wins or losses to the Player. game The Game used to determine if a given bet is allowed in a particular game state.

32.4.2 Constructors
__init__(self ) Creates an empty LinkedList of bets.

32.4.3 Methods
setGame(self, game) Saves the given Game to be used to validate bets. isValid(self, bet ) 32.4. Table Class 201

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Validates this bet. The rst test checks the Game to see if the bet is valid. allValid(self, bet ) Validates the sum of all bets within the table limits. Returns false if the minimum is not met or the maximum is exceeded. placeBet(self, bet ) Adds this bet to the list of working bets. If the sum of all bets is greater than the table limit, then an exception should be thrown (Java) or raised (Python). This is a rare circumstance, and indicates a bug in the Player more than anything else. __iter__(self ) Returns an Iterator over the list of bets. This gives us the freedom to change the representation from LinkedList to any other Collection with no impact to other parts of the application. In Python, similarly, we can return an iterator over the available list of Bet instances. The more traditional Python approach is to return the list itself, rather than an iterator over the list. With the introduction of the generators in Python 2.3, however, it is slightly more exible to return an iterator rather than the collection itself. The iterator could be built, for example, using the yield statement instead of the iter() function. __str__(self ) Reports on all of the currently placed bets.

32.4.4 Player class

Player places bets in a Game. This an abstract class, with no actual body for the placeBets() method. However, this class does implement the basic win() and lose() methods used by all subclasses. Roulette Player Hierarchy. The classes in the Roulette Player hierarchy need to have their superclass adjusted to conform to the newly-dened superclass. The former Passenger57 is renamed to RoulettePlayer. All of the various Roulette players become subclasses of RoulettePlayer. In addition to renaming Player1326 to Roulette1326, we will also have to change the references in the various classes of the Player1326State class hierarchy. We suggest leaving the class names alone, but merely changing the references within those ve classes from Player1326 to Roulette1326. Craps Player Hierarchy. The classes in the Craps Player hierarchy need to have their superclass adjusted to conform to the newly-dened superclass. We can rename CrapsPlayerMartigale to CrapsMartigale, and make it a subclass of CrapsPlayer. Other than names, there should be no changes to these classes.

32.4.5 Fields
stake The players current stake. Initialized to the players starting budget. roundsToGo The number of rounds left to play. Initialized by the overall simulation control to the maximum number of rounds to play. In Roulette, this is spins. In Craps, this is the number of throws of the dice, which may be a large number of quick games or a small number of long-running games. In Craps, this is the number of cards played, which may be large number of hands or small number of multi-card hands. table The Table used to place individual Bets.

202

Chapter 32. Design Cleanup and Refactoring

Building Skills in Object-Oriented Design, Release 2.1.1-Python

32.4.6 Constructors
__init__(self, table) Constructs the Player with a specic Table for placing Bets.

32.4.7 Methods
playing(self ) Returns true while the player is still active. There are two reasons why a player may be active. Generally, the player has a stake greater than the table minimum and has a roundsToGo greater than zero. Alternatively, the player has bets on the table; this will happen in craps when the game continues past the number of rounds budgeted. placeBets(self ) Updates the Table with the various Bets. When designing the Table, we decided that we needed to deduct the amount of a bet from the stake when the bet is created. See the Table Roulette Table Overview for more information. win(self, bet ) Notication from the Game that the Bet was a winner. The amount of money won is available via bet.winAmount(). lose(self, bet ) Notication from the Game that the Bet was a loser.

32.5 Game Class

Game manages the sequence of actions that denes casino games, including Roulette, Craps and Blackjack. Individual subclasses implement the detailed playing cycles of the games. This superclass has methods for notifying the Player to place bets, getting a RandomEvent and resolving the Bet s actually present on the Table.

32.5.1 Fields
eventFactory Contains a Wheel or Dice or other subclass of RandomEventFactory that returns a randomly selected RandomEvent with specic Outcome s that win or lose. table Contains a CrapsTable or RouletteTable which holds all the Bets placed by the Player. player Holds the Player who places bets on the Table.

32.5.2 Constructors
We based this constructor on an design that allows any of these objects to be replaced. This is the Strategy (or Dependency Injection) design pattern. Each of these objects is a replaceable strategy, and can be changed by the client that uses this game.

32.5. Game Class

203

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Additionally, we specically do not include the Player instance in the constructor. The Game exists independently of any particular Player, and we defer binding the Player and Game until we are gathering statistical samples. __init__(self, eventFactory, table) Constructs a new Game, using a given RandomEventFactory and Table.

32.5.3 Methods
cycle(self, player ) This will execute a single cycle of play with a given Player. For Roulette is is a single spin of the wheel. For Craps, it is a single throw of the dice, which is only one part of a complete game. This method will call player.placeBets() to get bets. It will call eventFactory.next() to get the next Set of Outcomes. It will then call table.bets() to get an Iterator over the Bets. Stepping through this Iterator returns the individual Bet objects. The bets are resolved, calling the thePlayer win(), otherwise call the thePlayer lose(). reset(self ) As a useful default for all games, this will tell the table to clear all bets. A subclass can override this to reset the game state, also.

32.6 RouletteGame Class

RouletteGame is a subclass of Game that manages the sequence of actions that denes the game of Roulette.

32.6.1 Methods
cycle(self, player ) This will execute a single cycle of the Roulette with a given Player. It will call player.placeBets() to get bets. It will call wheel.next() to get the next winning Bin . It will then call table.bets() to get an Iterator over the Bets. Stepping through this Iterator returns the individual Bet objects. If the winning Bin contains the Outcome, call the thePlayer win(), otherwise call the thePlayer lose().

32.7 CrapsGame Class

CrapsGame is a subclass of Game that manages the sequence of actions that denes the game of Craps. Note that a single cycle of play is one throw of the dice, not a complete craps game. The state of the game may or may not change.

32.7.1 Methods
cycle(self, player ) This will execute a single cycle of play with a given Player. 1. It will call player.placeBets() to get bets. It will validate the bets, both individually, based on the game state, and collectively to see that the table limits are met. 204 Chapter 32. Design Cleanup and Refactoring

Building Skills in Object-Oriented Design, Release 2.1.1-Python

2. It will call dice.next() to get the next winning Throw. 3. It will use the throw.updateGame() to advance the game state. 4. It will then call table.bets() to get an Iterator; stepping through this Iterator returns the individual Bet objects. It will use the Throws resolveOneRoll() method to check one-roll propositions. If the method returns true, the Bet is resolved and should be deleted. It will use the Throws resolveHardways() method to check the hardways bets. If the method returns true, the Bet is resolved and should be deleted. pointOutcome(self ) Returns the Outcome based on the current point. This is used to create Pass Line Odds or Dont Pass Odds bets. This delegates the real work to the current CrapsGameState object. moveToThrow(self, bet, throw ) Moves a Come Line or Dont Come Line bet to a new Outcome based on the current throw. This delegates the move to the current CrapsGameState object. This method should just as a precaution assert that the value of theThrow is 4, 5, 6, 8, 9 or 10. These point values indicate that a Line bet can be moved. For other values of theThrow, this method should raise an exception, since theres no reason for attempting to move a line bet on anything but a point throw. reset(self ) This will reset the game by setting the state to a new instance of GamePointOff. It will also tell the table to clear all bets.

32.8 Refactoring Deliverables

There are six deliverables for this exercise. If necessary, create RandomEvent, and revisions to Throw and Bin. See Design Heavy . Create RandomEventFactory, and associated changes to Wheel and Dice. The existing unit tests will conrm that this change has no adverse eect. Refactor Table and CrapsTable to make a single class of these two. The unit tests for the original CrapsTable should be merged with the unit tests for the original Table. Refactor Player and CrapsPlayer to create a better class hierarchy with CrapsPlayer and RoulettePlayer both sibling subclasses of Player . The unit tests should conrm that this change has no adverse eect. Refactor Game and CrapsGame to create three classes: Game, RouletteGame and CrapsGame. The unit tests should conrm that this change has no adverse eect. Create a new main program class that uses the existing Simulator with the CrapsGame and CrapsPlayer classes.

32.8. Refactoring Deliverables

205

Building Skills in Object-Oriented Design, Release 2.1.1-Python

206

Chapter 32. Design Cleanup and Refactoring

CHAPTER

THIRTYTHREE

SIMPLE CRAPS PLAYERS

This chapter denes a variety of player strategies. Most of this is based on strategies already dened in Roulette, making the explanations considerably simpler. Rather than cover each individual design in separate chapters, well rely on the experience gained so far, and cover four variant Craps players in this chapter. Well mention a fth, but leave that as a more advanced exercise.

33.1 Simple Craps Players Overview

When we looked at our Player hierarchy, we noted that we could easily apply a number of strategies to the Pass Line Odds bet. We use the Martingale strategy for our CrapsMartingale player. We could also use the 1-3-2-6 system, the Cancellation system, or the Fibonacci system for those odds bets. In each of these cases, we are applying the betting strategy to one of the two bets the player will use. An additional design note for this section is the choice of the two basic bets: the Pass Line and the Pass Line Odds bet. It is interesting to compare the results of these bets with the results of the Dont Pass Line and the Dont Pass Odds Bet. In particular, the Dont Pass Odds Bets involve betting large sums of money for small returns; this will be compounded by any of these betting systems which accelerate the amount of the bet to cover losses. This change should be a simple matter of changing the two base bets used by all these variant players. All of these players have a base bet (either Pass Line or Dont Pass) and an odds bet (either Pass Line Odds or Dont Pass Odds). If we create a superclass, called SimpleCraps , we can assure that all these simple betting variations will work for Pass Line as well as Dont Pass Line bets. The responsibility of this superclass is to dene a consistent set of elds and constructor for all of the subclasses. Craps 1-3-2-6. This player uses the 1-3-2-6 system for managing their odds bet. From the Roulette 1-3-2-6 player (see Player 1-3-2-6 Overview ) we can see that this player will need to use the Player1326State class hierarchy. The craps player will use one of these objects track the state changes of their odds. The base bet will not change. However, the current denitions of the Player1326State class hierarchy specically reference Roulette1326. Presenting us with an interesting design problem. How do we repair our design so that Player1326State can work with Roulette1326 and Craps1326? The only dependency is the eld outcome, which both Roulette1326 and Craps1326 must provide to Player1326State objects. We have three choices: extract the eld and make it part of an interface, refactor the eld up to the superclass, change the dention of Player1326State to make it more self-contained. Common Interface. The relationship between a subclass of Player and Player1326State can be formalized through an interface. We could dene something like Bet1326_Able, and use this for Roulette1326 and Craps1326. In this case, this appears to be an example of the Very Large Hammer design pattern. The problem seems too small for this language feature.

207

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Common Superclass. Refactoring the single instance variable up to the superclass makes a relatively minor change. However, it places a feature in the superclass which all but a few subclasses must ignore. This is another example of Swiss Army Knife design, where we will be subtracting a feature from a superclass. Delegate. If we change the Player1326State class to keep its own copy of the desired Outcome we cleanly remove any dependence on Player. The Player is still responsible for keeping track of the Outcome s, and has subcontracted or delegated this responsibility to an instance of Player1326State. The down side of this is that we must provide this Outcome to each state constructor as the state changes. The solution we embrace is changing the denition of Player1326State to include the Outcome. This delegates responsibility to the state, where it seems to belong. This will change all of the constructors, and all of the state change methods, but will cleanly separate the Player1326State class hierarchy from the Player class hierarchy. Python Duck-Typing In Python, the relationship between a Player1326State object and the Craps1326 is completely casual. We dont have to sweat the details of where precisely the Outcome is declared. In Java, however, we must be very careful to sort out the relationship between Craps1326, and Roulette1326. Pythons exibility is called duck typing: if it walks like a duck and quacks like a duck, it is a duck. In this case, any class with an outcome attribute is a candidate owner for a Player1326State object. Craps Cancellation. When can examine the Roulette Cancellation Cancellation Player Overview and see that this player will need to use a List of individual betting amounts. Each win for an odds bet will cancel from this List, and each loss of an odds bet will append to this List. As with the Craps Martingale player, we will be managing a base Pass Line bet, as well as an odds bet that uses the Cancellation strategy. The Cancellation algorithm can be easily transplanted from the original Roulette version to this new Craps version. Craps Fibonacci. When can examine the Roulette Fibonacci Fibonacci Player Overview and see that this player will need to compute new betting amounts based on wins and losses.

33.2 CrapsPlayer Design

Well extend CrapsPlayer to create a CrapsSimplePlayer that can place both Pass Line and Pass Line Odds bets, as well as Dont Pass Line and Dont Pass Odds bets. This will allow us to drop the CrapsPlayerPass class, and revise the existing CrapsMartingale. We have to rework the original Roulette-focused Player1326State hierarchy, and the Roulette1326 class to use the new version of the state objects. Once this rework is complete, we can add our Craps1326 and CrapsCancellation players. For additional exposure, the more advanced student can rework the Roulette Fibonacci player to create a CrapsFibonacci player.

33.3 CrapsSimplePlayer superclass

CrapsSimplePlayer is a subclass of CrapsPlayer and places two bets in Craps. The simple player has a base bet and an odds bet. The base bet is one of the bets available on the come out roll (either Pass Line 208 Chapter 33. Simple Craps Players

Building Skills in Object-Oriented Design, Release 2.1.1-Python

or Dont Pass Line), the odds bet is the corresponding odds bet (Pass Line Odds or Dont Pass Odds). This class implements the basic procedure for placing the line bet and the behind the line odds bet. However, the exact amount of the behind the line odds bet is left as an abstract method. This allows subclasses to use any of a variety of betting strategies, including Martingale, 1-3-2-6, Cancellation and Fibonacci.

33.3.1 Fields
lineOutcome Outcome for either Pass Line or Dont Pass Line. A right bettor will use a Pass Line bet; a wrong bettor will use the Dont Pass Line. oddsOutcome Outcome for the matching odds bet. This is either the Pass Line Odds or Dont Pass Line Odds bet. A right bettor will use a Pass Line Odds bet; a wrong bettor will use the Dont Pass Line Odds.

33.3.2 Constructors
__init__(self, table, line, odds) Parameters table (CrapsTable) The table on which bets are palced line (Outcome) The line bet outcome odds (Outcome) The odds bet outcome Constructs the CrapsSimplePlayer with a specic Table for placing Bets. Additionally a line bet (Pass Line or Dont Pass Line) and odds bet (Pass Line Odds or Dont Pass Odds) are provided to this constructor. This allows us to make either Pass Line or Dont Pass Line players.

33.3.3 Methods
placeBets(self ) Updates the Table with the various Bets. There are two basic betting rules. 1. If there is no line bet, create the line Bet from the line Outcome. 2. If there is no odds bet, create the behind the line odds Bet from the odds Outcome. Be sure to check the price of the Bet before placing it. Particularly, Dont Pass Odds bets may have a price that exceeds the players stake. This means that the Bet object must be constructed, then the price must be tested against the stake to see if the player can even aord it. If the stake is greater than or equal to the price, subtract the price and place the bet. Otherwise, simply ignore it.

33.4 Craps Martingale Player

CrapsMartingale is a subclass of CrapsSimplePlayer who places bets in Craps. This player doubles their Pass Line Odds bet on every loss and resets their Pass Line Odds bet to a base amount on each win.

33.4. Craps Martingale Player

209

Building Skills in Object-Oriented Design, Release 2.1.1-Python

33.4.1 Fields
lossCount The number of losses. This is the number of times to double the pass line odds bet. betMultiple The the bet multiplier, based on the number of losses. This starts at 1, and is reset to 1 on each win. It is doubled in each loss. This is always betM ultiple = 2lossCount .

33.4.2 Methods
placeBets(self ) Extension to the superclass placeBets() method. This version sets the amount based on the value of CrapsMartingale.betMultiple. win(self, bet ) Parameter bet (Bet) The bet that was a winner Uses the superclass win() method to update the stake with an amount won. This method then resets lossCount to zero, and resets betMultiple to 1. lose(self, bet ) Parameter bet (Bet) The bet that was a loser Increments lossCount by 1 and doubles betMultiple.

33.5 Player1326 State

Player1326State is the superclass for all of the states in the 1-3-2-6 betting system.

33.5.1 Fields
outcome The Outcome on which a Player will bet.

33.5.2 Constructors
__init__(self, outcome) Parameter outcome (Outcome) The outcome on which to bet The constructor for this class saves Outcome on which a Player will bet.

33.5.3 Methods
Much of the original design for this state hierarchy should remain in place. See Player 1-3-2-6 Class for more information on the original design. nextLost(self )

210

Chapter 33. Simple Craps Players

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Constructs the new Player1326State instance to be used when the bet was a loser. This method is the same for each subclass: it creates a new instance of Player1326NoWins. This method is dened in the superclass to assure that it is available for each subclass. This will use the outcome to be sure the new state has the Outcome on which the owning Player will be betting.

33.6 Craps1326 Player

Craps1326 is a subclass of CrapsSimplePlayer who places bets in Craps. This player changes their Pass Line Odds bet on every loss and resets their Pass Line Odds bet to a base amount on each win. The sequence of bet multipliers is given by the current Player1326State object.

33.6.1 Fields
state This is the current state of the 1-3-2-6 betting system. It will be an instance of one of the four states: No Wins, One Win, Two Wins or Three Wins.

33.6.2 Constructors
__init__(self, table, line, odds) Parameters table (CrapsTable) The table on which bets are palced line (Outcome) The line bet outcome odds (Outcome) The odds bet outcome Uses the superclass to initialize the Craps1326 instance with a specic Table for placing Bets, and set the line bet (Pass Line or Dont Pass Line) and odds bet (Pass Line Odds or Dont Pass Odds). Then the initial state of Player1326NoWins is constructed using the odds bet.

33.6.3 Methods
placeBets(self ) Updates the Table with a bet created by the current state. This method delegates the bet creation to state objects currentBet() method. win(self, bet ) Parameter bet (Bet) The bet that was a winner Uses the superclass method to update the stake with an amount won. Uses the current state to determine what the next state will be by calling states objects nextWon() method and saving the new state in state lose(self, bet ) Parameter bet (Bet) The bet that was a loser

33.6. Craps1326 Player

211

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Uses the current state to determine what the next state will be. This method delegates the next state decision to state objects nextLost() method, saving the result in state.

33.7 CrapsCancellation Player

CrapsCancellation is a subclass of CrapsSimplePlayer who places bets in Craps. This player changes their Pass Line Odds bet on every win and loss using a budget to which losses are appended and winings are cancelled.

33.7.1 Fields
sequence This List keeps the bet amounts; wins are removed from this list and losses are appended to this list. THe current bet is the rst value plus the last value.

33.7.2 Constructors
__init__(self, table, line odds) Parameters table (CrapsTable) The table on which bets are palced line (Outcome) The line bet outcome odds (Outcome) The odds bet outcome Invokes the superclass constructor to initialize this instance of CrapsCancellation. Then calls resetSequence() to create the betting budget.

33.7.3 Methods
There are few real changes to the original implementation of CancellationPlayer. See Cancellation Player Class for more information. placeBets(self ) Creates a bet from the sum of the rst and last values of sequence and the preferred outcome. This uses the essential line bet and odds bet algorithm dened above. If no line bet, this is created. If theres s line bet and no odds bet, then the odds bet is created. If both bets are created, there is no more betting to do.

33.8 Simple Craps Players Deliverables

There are eight deliverables for this exercise. The CrapsSimplePlayer abstract superclass. Since this class doesnt have a body for the oddsBet() method, it cant be unit tested directly.

212

Chapter 33. Simple Craps Players

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A revised CrapsMartingale class, that is a proper subclass of CrapsSimplePlayer . The existing unit test for CrapsMartingale should continue to work correctly after these changes. A revised Player1326State class hierarchy. Each subclass will use the outcome eld instead of getting this information from a Player1326 instance. The unit tests will have to be revised slightly to reect the changed constructors for this class. A revised Roulette1326 class, which reects the changed constructors for this Player1326State. The unit tests should indicate that this change has no adverse eect. The Craps1326 subclass of CrapsSimplePlayer. This will use the revised Player1326State. A unit test class for Craps1326. This test should synthesize a xed list of Outcomes, Throws, and calls a Craps1326 instance with various sequences of craps, naturals and points to assure that the bet changes appropriately. The CrapsCancellation subclass of CrapsSimplePlayer. A unit test class for CrapsCancellation. This test should synthesize a xed list of Outcomes, Throws, and calls a CrapsCancellation instance with various sequences of craps, naturals and points to assure that the bet changes appropriately.

33.8. Simple Craps Players Deliverables

213

Building Skills in Object-Oriented Design, Release 2.1.1-Python

214

Chapter 33. Simple Craps Players

CHAPTER

THIRTYFOUR

ROLL-COUNTING PLAYER CLASS

Craps allows a large number of opportunities to bet. The bets have dierent resolutions, leading to quite a complex game. In Roulette, each bets odds are easier to calculate becuase all bets are resolved at once. Placing multiple bets in Roulette is a simple modication to the payout odds and the probability of a win. A common Craps strategy is to add bets as a kind of insurance against losing the line bet. This means that well have numerous working bets: the mandatory line bet, the behind the line odds bet, plus an additional bets. For example, buying the 6 or 8 is a bet that has a payout that matches the actual odds. Well tackle a particularly complex betting strategy. In this case, a player that judges that a game has gone too long without a successful resolution. This is a common fallacy in probability theory. A seven is not due. The odds of throwing a seven are always 1/6.

34.1 Roll-Counting Overview

There is a distinction between one-roll odds and cumulative odds. The one roll odds of rolling a 7 are 1/6. This means that a Pass Line bet will win one time in six on the come out roll. The cumulative odds of rolling a 7 on a number of rolls depends on not rolling a seven (a 5/6 chance) for some number of rolls, followed by rolling a 7. The odds are given in the following table. Throws 1 2 3 4 5 6 Rule 1/6 5/6 1/6 (5/6)2 1/6 (5/6)3 1/6 (5/6)4 1/6 (5/6)5 1/6 Odds of 7 17% 31% 42% 52% 60% 67%

This cumulative chance of rolling a 7 means that the odds of the game ending with a loss because of throwing a 7 grow as the game progresses. We can compute the following sequence of odds of losing for larger numbers of rolls of the dice after the point is established: 17%, 28%, 35%, 40%, 43%, 46%, 47%, 48%, 49%. The idea is that the longer a game runs, the more likely you are to lose your initial Pass Line bet. Consequently, some players count the thows in the game, and eectively cancel their bet by betting against themselves on the Seven proposition. Note that the Seven proposition is a 1/6 probability that pays 5 for 1, (eectively 4:1). While the basic probability analysis of this bet is not encouraging, it does have an interesting design problem: the player now has multiple states. They have Pass Line bet, they can use a Martingale strategy for their Pass

215

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Line Odds bet, they are counting throws, and they are using a Martingale strategy for a Seven proposition starting with the seventh throw of the game. We also note that this analysis doesnt work for wrong bettors using the Dont Pass Line bet. Their concern is the opposite: a short game may cause them to lose their Dont Pass bet, but a long game makes it more likely that they would win. One simulation for this is to place Dont Pass Odds bets after several throws, where the odds bet appears more likely to win. Decomposition. This leads us to consider the player as a composite object with a number of states and strategies. It also leads us to design a class just to handle Martingale betting. Note that when we were looking at the design for the various players in Design Cleanup and Refactoring, we glanced at the possibility of separating the individual betting strategies from the players, and opted not to. However, we did force each strategy to depend on a narrowly-dened interface of oddsBet(), won() and lost(). We can exploit this narrow interface in teasing apart the various strategies and rebuilding each variation of Player with a distinct betting strategy object. The separation of Player from BettingStrategy involves taking the betting-specic information out of each player class, and replacing the various methods and elds with one or more BettingStrategy objects. In the case of Roulette players, this is relatively simple. In the case of Craps players, we note that we have two bets, one with a trivial-case betting strategy where the bet never changes. If we add this NoChange strategy, we can redene all Craps players bets using BettingStrategy objects. The responsibilities of a BettingStrategy are to maintain a preferred Outcome, maintain a bet amount, and change that amount in response to wins and losses. The existing win() and lose() methods are a signicant portion of these reponsibilities. The oddsBet() method of the various CrapsSimplePlayer embodies other parts of this, however, the name is inappropriate and it has a poorly thought-out dependency on Player. The responsibilities of a Player are to keep one or more betting strategies, so as to place bets in a Game. All of the Roulette players will construct a single BettingStrategy object with their preferred Outcome. The various CrapsSimplePlayer classes will have two BettingStrategys: one for the line bet and one for the odds bet. The only dierence among the simple strategies is the actual BettingStrategy object, simplifying the Player class hierarchy to a single Roulette player and two kinds of Craps players: the stub player who makes only one bet and the other players who make more than one bet and use a betting strategy for their odds bet. Implementing SevenCounter. Once we have this design in place, our SevenCounter player can then be composed of a Pass Line bet, a Pass Line Odds bet, and a Seven proposition bet that will only be used after seven rolls have passed in a single game. The line bet uses the NoChange strategy. The other two bets can use any of the strategies we have built: Martingale, 1-3-2-6 or Cancellation. Currently, there is no notication to the CrapsPlayer of unresolved bets. The player is only told of winners and losers. The opportunity to place bets indicates that the dice are being rolled. Additionally, the ability to place a line bet indicates that a game is beginning. We can use these two conditions to count the throws in during a game, eectively counting unresolved bets.

216

Chapter 34. Roll-Counting Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

34.2 BettingStrategy Design

BettingStrategy is the abstract superclass for all betting strategies. It contains a single Outcome, tracks wins and losses of Bet s built on this Outcome, and computes a bet amount based on a specic betting strategy.

34.2.1 Fields
outcome This is the Outcome that will be watched for wins and losses, as well as used to create new Bets.

34.2.2 Constructors
__init__(self, outcome) Parameter outcome (Outcome) The outcome on which this strategy will create bets Initializes this betting strategy with the given Outcome.

34.2.3 Methods
createBet(self ) Returns a new Bet using the outcome Outcome and any other internal state of this object. win(self, bet ) Parameter bet (Bet) The bet which was a winner Notication from the Player that the Bet was a winner. The Player has responsibility for handling money, this class has responsibility for tracking bet changes. lose(self, bet ) Parameter bet (Bet) The bet which was a loser Notication from the Player that the Bet was a loser. __str__(self ) Returns a string with the name of the class and appropriate current state information. For the superclass, it simply returns the name of the class. Subclasses will override this to provide subclass-specic information.

34.3 NoChangeBetting Class

NoChangeBetting is a subclass of BettingStrategy that uses a single, xed amount for the bet. This is useful for unit testing, for modeling simple-minded players, and for line bets in Craps.

34.3.1 Fields
betAmount This is the amount that will be bet each time. A useful default value is 1.

34.2. BettingStrategy Design

217

Building Skills in Object-Oriented Design, Release 2.1.1-Python

34.3.2 Constructors
__init__(self, outcome) Parameter outcome (Outcome) The outcome on which this strategy will create bets Uses the superclass initializer with the given Outcome.

34.3.3 Methods
createBet(self ) Returns a new Bet using the outcome Outcome and betAmount. win(self, bet ) Parameter bet (Bet) The bet which was a winner Since the bet doesnt change, this does nothing. lose(self, bet ) Parameter bet (Bet) The bet which was a loser Since the bet doesnt change, this does nothing. __str__(self ) Returns a string with the name of the class, outcome and betAmount.

34.4 MartingaleBetting Class

MartingaleBetting is a subclass of BettingStrategy that doubles the bet on each loss, hoping to recover the entire loss on a single win.

34.4.1 Fields
lossCount The number of losses. This is the number of times to double the pass line odds bet. betMultiple The the bet multiplier, based on the number of losses. This starts at 1, and is reset to 1 on each win. It is doubled in each loss. This is always betM ultiple = 2lossCount .

34.4.2 Constructors
__init__(self, outcome) Parameter outcome (Outcome) The outcome on which this strategy will create bets Uses the superclass initializer with the given Outcome. Sets the initial lossCount and betMultiplier.

218

Chapter 34. Roll-Counting Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

34.4.3 Methods
createBet(self ) Returns a new Bet using the outcome Outcome and the betMultiple. win(self, bet ) Parameter bet (Bet) The bet which was a winner Resets lossCount to zero, and resets betMultiple to 1. lose(self, bet ) Parameter bet (Bet) The bet which was a loser Increments lossCount by 1 and doubles betMultiple. __str__(self ) Returns a string with the name of the class, outcome, the current betAmount and betMultiple.

34.5 Bet1326Betting Class

Bet1326Betting is a subclass of BettingStrategy that advances the bet amount through a sequence of multipliers on each win, and resets the sequence on each loss. The hope is to magnify the gain on a sequence of wins.

34.5.1 Fields
state This is the current state of the 1-3-2-6 betting system. It will be an instance of one of the four subclasses of Player1326State: No Wins, One Win, Two Wins or Three Wins.

34.5.2 Constructors
__init__(self, outcome) Parameter outcome (Outcome) The outcome on which this strategy will create bets Initializes this betting strategy with the given Outcome. Player1326NoWins using outcome. Creates an initial instance of

34.5.3 Methods
createBet(self ) Returns a new Bet using the currentBet() method from the state object. win(self, bet ) Parameter bet (Bet) The bet which was a winner Determines the next state when the bet is a winner. Uses states nextWon() method and saves the new state in state. lose(self, bet )

34.5. Bet1326Betting Class

219

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Parameter bet (Bet) The bet which was a loser Determines the next state when the bet is a loser. Uses states nextLost(), method saving the result in myState. __str__(self ) Returns a string with the name of the class, outcome and state.

34.6 CrapsOneBetPlayer class

CrapsOneBetPlayer is a subclass of CrapsPlayer and places one bet in Craps. The single bet is one of the bets available on the come out roll (either Pass Line or Dont Pass Line). This class implements the basic procedure for placing the line bet, using an instance of BettingStrategy to adjust that bet based on wins and losses.

34.6.1 Fields
lineStrategy An instance of BettingStrategy that applies to the line bet. Generally, this is an instance of NoChangeBetting because we want to make the minimum line bet and the maximum odds bet behind the line.

34.6.2 Constructors
__init__(self, table, lineStrategy) Constructs the CrapsOneBetPlayer with a specic Table for placing Bet s. This will save the given BettingStrategy in lineStrategy.

Python Creation of A Player

passLine= table.dice.get( "Pass Line" ) bet= new MartingaleBetting( passLine ) passLineMartin= new CrapsOneBetPlayer( bet );

1. Get the basic Pass Line Outcome from the Dice. 2. Creates a Martingale betting strategy focused on the basic Pass Line outcome. 3. Creates a one-bet player, who will employ the Martingale betting strategy focused on the basic Pass Line outcome.

34.6.3 Methods
placeBets(self ) Updates the Table with the various Bet s. There is one basic betting rule. 1. If there is no line bet, create the line Bet from the lineStrategy.

220

Chapter 34. Roll-Counting Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Be sure to check the price of the Bet before placing it. Particularly, Dont Pass Odds bets may have a price that exceeds the players stake. This means that the Bet object must be constructed, then the price must be tested against the stake to see if the player can even aord it. If the stake is greater than or equal to the price, subtract the price and place the bet. Otherwise, simply ignore it. win(self, bet ) Parameter bet (Bet) The bet which was a winner Notication from the Game that the Bet was a winner. The amount of money won is available via theBet winAmount(). If the bets Outcome matches the lineStrategy s Outcome, notify the strategy, by calling the lineStrategy s win() method. lose(self, bet ) Parameter bet (Bet) The bet which was a loser Notication from the Game that the Bet was a loser. If the bets Outcome matches the lineStrategy s Outcome, notify the strategy, by calling the lineStrategy s lose() method.

34.7 CrapsTwoBetPlayer class

CrapsTwoBetPlayer is a subclass of CrapsOneBetPlayer and places one or two bets in Craps. The base bet is one of the bets available on the come out roll (either Pass Line or Dont Pass Line). In addition to that, an odds bet (either Pass Line Odds or Dont Pass Odds) can also be placed. This class implements the basic procedure for placing the line and odds bets, using two instances of BettingStrategy to adjust the bets based on wins and losses. Typically, the line bet uses an instance of NoChangeBetting. The odds bets, however, are where we want to put more money in play.

34.7.1 Fields
oddsStrategy An instance of BettingStrategy that applies to the line bet.

34.7.2 Constructors 34.7.3 Methods

placeBets(self ) Updates the Table with the various Bet s. There are two basic betting rules. 1. If there is no line bet, create the line Bet from the lineStrategy. 2. If there is no odds bet, create the odds Bet from the oddsStrategy. win(self, bet ) Parameter bet (Bet) The bet which was a winner Notication from the Game that the Bet was a winner. The superclass handles the money won and the line bet notication. This subclass adds a comparison between the bets Outcome and the

34.7. CrapsTwoBetPlayer class

221

Building Skills in Object-Oriented Design, Release 2.1.1-Python

oddsStrategys Outcome; if they match, it will notify the strategy, by calling the oddsStrategy s win() method. lose(self, bet ) Parameter bet (Bet) The bet which was a loser Notication from the Game that the Bet was a loser. The superclass handles the line bet notication. If the bets Outcome matches the oddsStrategys Outcome, notify the strategy, by calling the oddsStrategys lose() method.

34.8 CrapsSevenCountPlayer class

CrapsSevenCountPlayer is a subclass of CrapsTwoBetPlayer and places up to three bets in Craps. The base bet is a Pass Line bet. In addition to that, a Pass Line Odds bet can also be placed. If the game runs to more than seven throws, then the 7 proposition bet (at 4:1) is placed, using the Martingale strategy. The Pass Line bet uses an instance of NoChangeBetting. The Pass Line Odds bet uses an instance of Bet1326Betting.

34.8.1 Fields
sevenStrategy The BettingStrategy for the seven bet. Some argue that this should be a no-change strategy. The bet is rare, and if eect the player bets against themself with this. One could also argue that it should be a Martingale because each throw after the seventh are less and less likely to win. throwCount The number of throws in this game. This is set to zero when we place a line bet, and incremented each time we are allowed to place bets.

34.8.2 Constructors
__init__(self, table) This will create a NoChangeBetting strategy based on the Pass Line Outcome. It will also create a MartingaleBetting strategy based on the Pass Line Odds Outcome. These will be given to the superclass constructor to save the game, the line bet and the odds bet. Then this constructor creates a Bet1326Betting strategy for the Seven Proposition Outcome.

34.8.3 Methods
placeBets(self ) Updates the Table with the various Bet s. There are three basic betting rules. 1. If there is no line bet, create the line Bet from the lineStrategy. Set the throwCount to zero. 2. If there is no odds bet, create the odds Bet from the oddsStrategy. 3. If the game is over seven throws and there is no seven proposition bet, create the proposition Bet from the sevenStrategy. Each opportunity to place bets will also increment the throwCount by one.

222

Chapter 34. Roll-Counting Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

win(self, bet ) Parameter bet (Bet) The bet which was a winner Notication from the Game that the Bet was a winner. The superclass handles the money won and the line and odds bet notication. lose(self, bet ) Parameter bet (Bet) The bet which was a loser Notication from the Game that the Bet was a loser. The superclass handles the line and odds bet notication.

34.9 Roll-Counting Deliverables

There are two groups of deliverables for this exercise. The rst batch of deliverables are the new Betting Strategy class hierarchy and unit tests. The second batch of deliverables are the two revised Craps Player classes, the nal Roll Counter Player, and the respective unit tests. Also, note that these new classes make the previous CrapsSimplePlayer, CrapsMartingale, Craps1326 and CrapsCancellation classes obsolete. There are two choices for how to deal with this change: remove and reimplement. The old calsses can be removed, and the Simulator reworked to use the new versions. The alternative is to reimplement the original classes as Facade over the new classes. Betting Strategy class hierarchy. There are four classes, each with an associated unit test in this group of deliverables. The BettingStrategy superclass. This class is abstract; there is no unit test. The NoChangeBetting class. A unit test for the NoChangeBetting class. This will simply conrm that the win() and lose() methods do not change the bet amount. The MartingaleBetting class. A unit test for the MartingaleBetting class. This will conrm that the win() method resets the bet amount and lose() method doubles the bet amount. The Bet1326Betting class. A unit test for the Bet1326Betting class. This will conrm that the win() method steps through the various states, and the lose() method resets the state. CrapsPlayer class hierarchy. There are three classes, each with an associated unit test in this group of deliverables. The CrapsOneBetPlayer class. A unit test for the CrapsOneBetPlayer class. One test can provide a No Change strategy for a Pass Line bet to verify that the player correctly places Line bets. Another test can provide a Martingale strategy for a Pass Line bet to verify that the player correctly changes bets on wins and losses. The CrapsTwoBetPlayer class. A unit test for the CrapsTwoBetPlayer class. One test can provide a No Change strategy for a Pass Line bet and a Martingale strategy for a Pass Line Odds bet to verify that the player correctly places Line bets and correctly changes bets on wins and losses. The CrapsSevenCountPlayer class.

34.9. Roll-Counting Deliverables

223

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A unit test for the CrapsSevenCountPlayer class. This will require a lengthy test procedure to assure that the player correctly places a Seven proposition bet when the game is over seven throws long.

224

Chapter 34. Roll-Counting Player Class

CHAPTER

THIRTYFIVE

CONCLUSION
The game of Craps has given us an opportunity to extend and modify an application with a considerable number of classes and objects. It is large, but not overly complex, and produces interesting results. Further, as a maintenance and enhancement exercise, it gave us an opportunity to work from a base of software, extending and rening the quality of the design. We omitted exercises which would integrate this package with the Simulator and collect statistics. This step, while necessary, doesnt include many interesting design decisions. The nal deliverable should be a working application that parses command-line parameters, creates the required objects, and creates an instance of Simulator to collect data. Refactoring. We note that many design decisions required us to explore and refactor a great deal of the applications design. In writing this part, we found it very diicult to stick to our purpose of building up the design using realistic steps and realistic problem solving. A book with a description of an already-completed structure does not help new designers learn the process of design, and does not help people to identify design problems and correct them. Because of that philosophy, the complete refactoring of the design in the Design Cleanup and Refactoring chapter was an important activity because it is the kind of thing that distinguishes a good design from a haphazard design. Our observation is that this kind of design rework happens late in the life of a project, and project managers are uncomfortable evaluating the cost and benet of the change. Further, programmers are unable to express the cost of accumulating technical debt by making a series of less-than-optimal decisions. Simpler is Better. Perhaps, the most important lesson that we have is the constant search for something we call The Big Simple. We see the history of science as a search for simpler explanations of natural phenomena. The canonical example of this is the distinction between the geocentric model and the heliocentric model of the solar system. Both can be made to work: astronomers carefully built extremely elaborate models of the geocentric heavens and produced accurate predictions of planetary positions. However, the model of planetary motion around the sun describes real phenomena more accurately and has the added benet of being much simpler than competing models. To continue this rant, we nd that software designers and their managers do not feel the compulsion and do not budget the time to identify the grand simplication that is possible. In some cases, the result of simplifying the design on one axis will create more classes. Designers lack a handy metric for understandability; managers are able to count individual classes, no matter how transparently simple. Designers often face diiculties in defending a design with many simple classes; some people feel that a few complex classes is simpler because it has fewer classes. As our trump card, we reference the metrics for complexity. McCabes Cyclomatic Complexity penalizes ifstatements. By reducing the number of if-statements to just those required to create an object of the proper class, we reduce the complexity. The Halstead metrics penalize programs with lots of internal operators and operands when compared with the number of operands in the interface. Halstead measures simple, transparent classes as less complex. Neither measure penalizes overall size in lines of code, but rather they

225

Building Skills in Object-Oriented Design, Release 2.1.1-Python

penalize decision-making and hidden, internal state. A badly designed, complex class has hidden internal states, often buried in nested if-statements. We emphasize small, simple classes with no hidden features. Our concrete examples of this simplication process are contained in three large design exercises. In Throw Builder Class, we showed a kind of rework necessary to both generalize and isolate the special processing for Craps. In Design Cleanup and Refactoring, we reworked classes to create an easy-to-explain architecture with layers and partitions of responsibility. Finally, in Roll-Counting Player Class, we uncovered a clean separation between game rules and betting strategies.

226

Chapter 35. Conclusion

Part IV

Blackjack

227

Building Skills in Object-Oriented Design, Release 2.1.1-Python

This part describes the more complex game of Blackjack. Both the player and the dealer are trying to build a hand that totals 21 points without going over. The hand closest to 21 wins. This game has a number of states and a number of complex state-change rules. It has very few dierent kinds of bets, but moderately complex rules for game play. However, it does have the most sophisticated playing strategy, since the player has a number of choices to make. The chapters of this part presents the details on the game, an overview of the solution, and a series of six relatively complex exercises to build a complete simulation of the game. In the case of Blackjack, we have to create a design that allows for considerable variation in the rules of the game as well as variation in the players betting strategies.

229

Building Skills in Object-Oriented Design, Release 2.1.1-Python

230

CHAPTER

THIRTYSIX

BLACKJACK DETAILS
In the rst section we will present elements of the game of Blackjack. Blackjack uses cards and has fairly complex rules for counting the number of points in a hard of cards. Blackjack oers relatively few bets, most of which are available based on the state of the game. Well cover these bets and the conditions under which they are allowed in the second sectionj. Finally, we will describe some common betting and playing strategies that we will simulate. In this case, we have playing strategies that are unique to Blackjack, combined with betting strategies initially dened in Roulette and reworked in Craps.

36.1 Blackjack Game

Blackjack centers around hands composed of cards drawn from one or more standard 52-card decks. The standard deck has 13 ranks in 4 suits; the suit information has no bearing on game play. The player and the house are both dealt hands, starting with two cards. The house has one card exposed (the up card ) and one card concealed (the hole card ), leaving the player with incomplete information about the state of the game. The players objective is to make a hand that has more points than the dealer, but less than or equal to 21 points. The player is responsible for placing bets when they are oered, and taking additional cards to complete their hand. The dealer will draw additional cards according to a simple rule: when the dealers hand is 16 or less, they will draw cards (or hit ), when it is 17 or more, they will not draw additional cards (or stand pat ). An interesting complication is the point values of the cards. The number cards (2-10) have the expected point values. The face cards (Jack, Queen and King) all have a value of 10 points. The Ace can count as one point or eleven points. Because of this, an Ace and a 10 or face card totals 21. This two-card winner is called blackjack. Also, when the points include an ace counting as 11, the total is called soft ; when the ace counts as 1, the total is called hard. For example, A-5 is called a soft 16 because it could be considered a hard 6. A-10-5 is a hard 16. The betting surface is marked with two places for bets: a single bet, placed before any cards are dealt, and an insurance bet, oered only when the dealers up card is an ace. There are a few additional bets, and a few player choices. Well step through some variations on the sequence of play to see the interactions a player has during a game. Note that a casino table seats a number of players; like Craps and Roulette, the player opposes the house, and the presence or absence of other players has no bearing on the game. Note: Rule Variations There are seemingly endless variations in the exact playing rules used by dierent casinos. Well focus on a relatively common version of the rules. With this as a basis, a number of variations can be explored. Typical Scenario. The player places an initial bet. Since the bet is blind, it is like an ante in poker. The player and dealer are each dealt a pair of cards. Both of the players are face up, the dealer has one 231

Building Skills in Object-Oriented Design, Release 2.1.1-Python

card up and one card down. If the dealers card is an ace, the player is oered insurance. The details will be described in a separate sceanario, below. Initially, the player has a number of choices. If the two cards are the same rank, the player can elect to split into two hands. This is a separate scenario, below. The player can double their bet and take just one more card. In some casinos this opportunity may be limited to certain totals on the cards, for instance, only 10 and 11; or it may be limited to a certain number of cards, for instance, two cards. The more typical scenario is for the player to take additional cards (a hit ) until either their hand totals more than 21 (they bust ), or their hand totals exactly 21, or they elect to stand. If the players hand is over 21, their bet is resolved immediately as a loss. Resolving these bets early is an important part of the houses edge in Blackjack. If the players hand is 21 or less, however, it will be compared to the dealers hand for resolution. The dealer then reveals the hole card and begins taking cards according to their xed rule. When their total is 16 or less, they take an additional card; if their total is 17 or more, they stand pat. This rule is summarized as hit on 16, stand on 17. In some casinos a dealer will hit a soft 17 (A-6), which improves the houses edge slightly. If the dealer busts, the player wins. If the dealer did not bust, then the hands are compared: if the players total is more than the dealer, the player wins; if the totals are equal, the bet is a push; otherwise the dealers total is more than the play and the player loses. If the players hand is an ace and a 10-point card (10, Jack, Queen or King), the hand is blackjack and the the ante is paid o at 3:2. Otherwise, winning hands that are not blackjack are paid o at 1:1. Dealer Shows An Ace. If the dealers up card is an ace, the player is oered an insurance bet. This is an additional proposition that pays 2:1 if the dealers hand is exactly 21 (a 4/13 probability). The amount of the bet is half the original ante. If this insurance bet wins, it will, in eect, cancel the loss of the ante. After oering insurance to the player, the dealer will check their hole card and resolve the insurance bets. If the hole card is 10-point card, the dealer has blackjack, the card is revealed, and insurance bets are paid. If the hole card is not a 10-point card, the insurance bets are lost, but the card is not revealed. In the unusual case that the dealer shows an ace and the player shows blackjack (21 in two cards), the player will be oered even money instead of the insurance bet. If the player accepts the even money oer, their hand is resolved at 1:1 immediately, without examining the dealers hole card or playing out the hand. If the player declines even money, they can still bet or decline insurance. Checking the odds carefully, there is a 4/13 (30.7%) chance of the dealer having 21, but insurance is paid as if the odds were 1/3 (33.3%). Since the player knows they have 21, there is a 4/13 probability of a push plus winning the insurance bet (both player and dealer have 21) and a 9/13 probability of winning at 3:2, but losing the insurance bet (eectively a push). Split Hands. When dealt two cards of the same rank, the player can split the cards to create two hands. This requires an additional bet on the new hand. The dealer will deal an additional card to each new hand, and the hands are played independently. Generally, the typical scenario described above applies to each of these hands. The general rule of thumb is to always split aces and eights. The ideal situation is to split aces, and get dealt a 10-point card on each ace. Both hands pay 3:2. A more common situation is to have a low card (from 2 to 7) paired up with the ace, leading to soft 13 through soft 18. Depending on the dealers up card, these are opportunities to double down, possibly increasing the bet to 4 times the original amount. Some casinos restrict doubling down on the split hands. In rare cases, one or more of the new cards will match the original pair, possibly allowing further splits. Some casinos restrict this, only allowing a single split. Other casinos prevent resplitting only in the case of aces.

232

Chapter 36. Blackjack Details

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Note that the players election to split hands is given after any oer and resolution of insurance bets.

36.2 Available Bets and Choices

Unlike Roulette and Craps, Blackjack has only a few available bets. Generally, the following choices all involve accepting an oer by creating an additional bet. Ante. This bet is mandatory to play. It must be within the table limits. Insurance. This bet is oered only when the the dealer shows an ace. The amount must be half the ante. Note that the even money oer is an option for resolution of the ante instead of an insurance bet. It is sometimes described as a separate kind of bet, but this doesnt seem accurate. Split. This can be thought of as a bet that is oered only when the the players hand has two cards are of equal rank. Or this can be thought of as a playing option, akin to hitting or standing. The amount of the bet must match the original ante. Double. This can be thought of as a bet that is oered instead of a taking an ordinary hit. Some casions only oer this when the the players hand has two cards. Or this can be thought of as a playing option, akin to hitting or standing. The amount of the bet must match the original ante. Blackjack also oers the player some choices that dont involve creating additional bets. In the casino these are shown through gestures that can be seen clearly by dealers and other casino sta. Even Money. This resolution of insurance is oered only when the the dealer shows an ace and the player shows 21 in two cards. It is oered instead of an insurance bet. If accepted, the hand is resolved. If declined, the insurance oer can then be accepted or declined. Hit. The player is oered the opporutunity to take another card when their total is less than 21. If they decline the hit, they are standing pat. In casinos that allow a double down on any number of cards, the player has three choices: hit, double down or stand. Play begins with a sequence of oers which can be accepted or declined: insurance, even money resolution of insurance, and splitting a hand. After these oers, the player must select between the three remaining choices (hit, double or stand) for each of their hands with a total less than 21. In Roulette, there are no additional oers for the player to accept or decline. In Craps, we ignored the only oer made to a player. See Some Betting Strategies for details on this rule. Adding this interaction to Craps would require dening an additional method for CrapsPlayer to accept or decline an oer. We would also have the CrapsGame interrogate the Table for the presence of come point odds bets, make the oer to the player, and then activate or deactivate the bet for the next throw only. This level of interaction was a nuance we elected to ignore at that time.

36.3 Betting Strategies

Because of the complexity of Blackjack, the strategies for play focus on the cards themselves, not on the bets. Some players use a single xed bet amount. Some players will attempt to count cards, in an eort to determine the approximate distribution of cards in the deck, and vary their play or bets accordingly. The casinos actively discourage counting in a number of ways. The most common way is to shue 5 decks of cards together, and only deal the rst 156 or so of the available 260 cards. Additionally, they will ask people to leave who are obviously counting. The players responses to the various oers are what denes the playing strategy in Blackjack. The information available to the player is their hand (or hands), and the dealers up card. Therefore, all of the strategies for play decompose to a matrix showing the players total vs. dealers up card and a recommendation for which oers to accept or decline. 36.2. Available Bets and Choices 233

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Most players will decline the insurance oer, except when they hold a 21. In that rare case the even money oer should be declined, since the expected value analysis of the result shows a slightly better payout by competing against the dealer. The decision matrix has two parts: accepting or rejecting the split oer, and choosing among hit, stand or double down. You can buy cards in casino gift-shops that summarize a playing strategy in a single, colorful matrix with a letter code for split, hit, double and stand. Note that each decision to hit results in a new card, changing the situation used for decision-making. This makes the strategy an interesting, stateful algorithm. A player could easily add the betting strategies weve already dened to their Blackjack play strategies. A player could, for example, use the Martingale system to double their bets on each hand which is a loss, and reset their betting total on each hand which is a win. Indeed, our current design permits this, since we disentangled the betting strategies from the individual games in Roll-Counting Player Class.

234

Chapter 36. Blackjack Details

CHAPTER

THIRTYSEVEN

BLACKJACK SOLUTION OVERVIEW

We will present a survey of the new classes gleaned from the general problem statement in Problem Statement as well as the problem details in Blackjack Details. This survey is drawn from a quick overview of the key nouns in these sections. We will not review those nouns already examined for Craps and Roulette.

37.1 Preliminary Survey of Classes

In reading the background information and the problem statement, we noticed a number of nouns that seemed to be new to the game of Blackjack. Card Deck Point Value Hand Number Card Face Card Oer Insurance Split Double Hit Stand Player Game The following table summarizes some of the new classes and responsibilities that we can identify from the problem statement. This is not the complete list of classes we need to build. As we work through the exercises, well discover additional classes and rework some of these classes more than once. We also have a legacy of classes available from the Roulette and Craps solutions. We would like to build on this infrastructure as much as possible.

235

Building Skills in Object-Oriented Design, Release 2.1.1-Python

37.2 Preliminary Class Structure

Card Responsibilities Three apparent subclasses: NumberCard, FaceCard and Ace` A standard playing card with a rank and a suit. Also has a point value from 1 to 11. Aces have point values that depend on the Hand. Collaborators Collected in a Deck; collected into Hands for each Player; collected into a Hand for the dealer; added to by Game. Deck Responsibilities A complete set of 52 standard Cards. Collaborators Used by the Game to contain Cards. Hand Responsibilities A collection of Cards with one or two point values: a hard value (an ace counts as 1) and a soft value (an ace counts as 11). The house will reveal one Card to the player. Collaborators A Player may have 1 or more Hands; a Hand has 2 or more Cards. The Game adds Cards to the Hand. The Game checks the number of cards, the point totals and the ranks of the cards to oer dierent bets. The Game compares the point totals to resolve bets. Player Responsibilities Places the initial ante Bets, updates the stake with amounts won and lost. Accepts or declines oered additional bets, including insurance, and split. Accepts or declines oered resolution, including even money. Chooses among hit, double and stand options. Collaborators Uses Table, and one or more Hands. Examines the dealers Hand. Used by game to respond to betting oers. Used by Game to record wins and losses. Game Responsibilities Runs the game: oers bets to Player, deals the Cards from the Deck to Hands, updates the state of the game, collects losing bets, pays winning bets. Splits Hands. Responds to player choices of hit, double and stand. This encapsulates the basic sequence of play into a single class. Collaborators Uses Deck, Table, Outcome, Player.

37.3 A Walkthrough
The unique, new feature of Blackjack is the more sophisticated collaboration between the game and the player. This interaction involves a number of oers for various bets, and bet resolution. Additionally, it includes oers to double, hit or stand. Well examine parts of a typical sequence of play to assure ourselves that we have all of the necessary collaborations and responsibilities.

236

Chapter 37. Blackjack Solution Overview

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A good way to structure this task is to do a CRC walkthrough. For more information on this technique see A Walkthrough of Roulette. Well present the overall sequence of play, and leave it to the student to manage the CRC walkthrough.

Typical Blackjack Game

1. Place Bets. The Game will ask the player to place a bet. If the player doesnt place a bet, the session is over. 2. Create Hands. The Game will deal two cards to the Players initial Hand. The Game will create an initial hand of two cards for the dealer. One of the cards is the up card, and is visible to the player. 3. Insurance? The Game gets the Dealers Hands up card. If it is an Ace, then insurance processing is perforemed. (a) Oer Even Money. The Game examines the Players hand for two cards totalling a soft 21, blackjack. If so, the Game oers the Even Money resolution to the Player. If the player accepts, the entire game is resolved at this point. The ante is paid at even money; there is no insurance bet. (b) Oer Insurance. The Game oers insurance to the Player, who can accept by creating a bet. For players with blackjack, this is a second oer after even money is declined. If the player declines, there are no further insurance considerations. (c) Examine Hole Card. The Game examines the Dealers Hands hole card. If is is a 10-point value, the insurance bet is resolved as a winner, the ante is resolved as a loser, and for this player, the game is over. Otherwise the insurance is resolved as a loser, the hole card is not revealed, and play will continue. Note that in a casino with multiple players, it is possible for a player declining insurance to continue to play with the dealers hole card revealed. For casinos that oer early surrender this is the time to surrender. 4. Split? The Game examines the Players Hand to see if the two cards are of equal rank. If so, it oers a split. The player accepts by creating an additional Bet. The original hand is removed; The Game splits the two original Cards then deals two additional Cards to create two new Hands. Some casinos prevent further splitting, others allow continued splitting of the resulting hands. 5. Play Out Player Hands. The following are done to play out each of the Players Hands. (a) Bust? Double? Hit? Stand? While the given Hand is under 21 points, the Game must extend three kinds of oers to the Player. If the Player accepts a Hit, the hand gets another card and this process repeats. If the Player accepts Double Down, the player must create an additional bet, and the hand gets one more card and play is done. If the Player Stands Pat, the play is done. If the hand is 21 points or over, play is done. (b) Resolve Bust. The Game must examine each Hand; if it is over 21, the Hand is resolved as a loser. 6. Play Out Dealer Hand. The Game then examines the Dealer Hand and deals Cards on a point value of 16 or less, and stops dealing Cards cards on point value of 17 or more. (a) Dealer Bust? The Game then examines the Dealer Hand to see if it is over 21. If so, the players bets are resolved as winners. Player Hands with two cards totalling 21 ( blackjack ) are paid 3:2, all other hands are paid 1:1.

37.3. A Walkthrough

237

Building Skills in Object-Oriented Design, Release 2.1.1-Python

7. Compare Hands. For each hand still valid, the Game compares the Players Hand point value against the Dealers Hand point value. Higher point value wins. In the case of a tie, it is a push and the bet is returned. When the Player wins, a winning hand with two cards totalling 21 (blackjack) is paid 3:2, any other winning hand is paid 1:1.

37.4 Blackjack Solution Questions and Answers

Will we really need both Deck and the multiple deck Shoe? Wouldnt it be simpler to combine this functionality into a single class? There are two separate responsibilities here. The deck owns the basic responsibility to build the 52 cards. The shoe, on the other hand, owns the responsibility to deal cards to hands. We want to be able to simulate games with 1 to 8 decks. A single deck game can simply deal directly from the deck. In a multi-deck game, all of the decks are shued together and loaded into a shoe for dealing. The dierence between one deck and a ve-deck shoe is that the shoe can produce 20 kings in a row. While rare, our simulation does need to cover situations like this. Also, we may want to build a slightly dierent shoe that simulates the continuous shuing machine that some casinos use. In this case, each hand is reshued back into the shoe, preventing any attempt at card counting. We dont want to disturb the basic, common deck when introducing this additional feature. Wont all those player interactions break our design? Thats unlikely. All of that player interaction is in addition to the placeBets() interface. Since weve separated the core features of all players from the game-specic features, we can add a subclass to player that will be handle the Blackjack interaction. This new player subclass will have a number of additional methods to handle insurance, even money, split and the regular play questions of hit, double and stand. In parallel, weve separated the core features of all games from the unique features for a specic game. We can now add a subclass for Blackjack which adds a number of methods to oer insurance, even money, split and the regular play questions of hit, double and stand to the Blackjack player. I cant nd an Outcome in Blackjack. Is it the Ante? If so, the odds vary based on the players Hand, but that doesnt seem to be a RandomEvent. Good point. Well examine this in detail in the exercises. Clearly, the bets are placed on the Ante and Insurance as the two core Outcome s in Blackjack. The Insurance outcome (really a dealer has blackjack outcome) is xed at 2:1. The ante payo depends on a complex condition of the hand: for a soft 21, or blackjack, it pays 3:2; otherwise it pays 1:1. This will lead to a new subclass of Outcome that collaborates with the hand to determine the payout to use. The even money is oered before ordinary insurance to a player with blackjack. It, however, pays even money on the ante, and doesnt create a new bet; in this respect it could be thought of as a change in the outcome on which the ante bet is created. Accepting the even money oer is a little bit like moving the ante to a even money for dealer blackjack outcome, which has 1:1 odds instead of 3:2 odds. Further, this special outcome is resolved before the dealer peeks at their hole card. Perhaps this is a special best resolution procedure, not a proper instance of Outcome.

238

Chapter 37. Blackjack Solution Overview

CHAPTER

THIRTYEIGHT

CARD, DECK AND SHOE CLASSES

This chapter introduces a number of simple classes. When exploring Roulette, we introduced classes very slowly. In this chapter, we are introducing the class hierarchy for cards, the deck of cards and the shoe, used by the dealer to create hands. The next chapter will introduce the hand, and the problems of scoring the value of the hand.

38.1 Card, Deck and Shoe Overview

Card. The standard playing card has two attributes: a rank (Ace, 2 through 10, Jack, Queen, or King) and a suit (Clubs, Diamonds, Hearts or Spades). The set of all 52 combinations comprises a full deck. Additionally, in Blackjack, a card has a point value, which is a number from 1 to 11, and is based on the rank. In the case of an Ace, it is also based on the hand in which the card is evaluated. This collaboration with a hand complicates the responsibilities for a single card. However, the basic set of responsibilities of the Card class include keeping the rank, suit and point value of a single standard playing card. The issue of suit requires some care. In the game of Blackjack, suits dont matter. Indeed, for the purposes of simulation, we could simply discard the notion of suit. However, for the purposes of making a reasonably complete model of real-world objects, it makes sense to implement the suit of a card, even if we do nothing with it. Wed like to provide symbolic variable names for the suits. In Java well use static nal variables to dene the suits. In Python well use class variables. This will give us four named constants for each of the suits, something that tends to make the software easier to read. Similarly, well provide named constants for the three face cards and the ace. Points. We have three dierent rules for establishing the point value of a card. This is a big hint that we have three subclasses. One subclass includes the Aces, which are either 1 or 11 points. Another subclass includes the number cards from 2 to 10, where the rank is the point value. Finally, the third subclass includes the face cards, where the point value is xed at 10. These three subclasses implement slightly dierent versions of a method to return the point value of the card. For more discussion on this, see Card-Deck-Shoe Questions and Answers FAQ for more discussion. The problem of determining the value of aces is something we will have to defer until after we create the hand class. First, well implement the basic card, then well develop the hand, and decide how the hand computes its number of points from the cards. We do note that the terminology used gives us some hint as to how to structure our design. Players refer to a soft total and a hard total. An A-6 hand is called a soft 17, and a hard 7. A soft hand has some exibility in how it is played. If you hit a soft 17, and get a face card (for example, a Jack), you now have an A-J-6, totalling hard 17. Considering this, we can provide two point-value functions for each rank, a hard value and a soft value. For all but aces, the two values are the same. For aces, the two values are dierent. This would allow a hand to compute a hard total and a soft

239

Building Skills in Object-Oriented Design, Release 2.1.1-Python

total for the hand as a whole. While this design does have some potential problems in dealing with multiple aces in a single hand, well let it stand until we have more design in place. Deck. A deck of cards has responsibility for shuing and dealing the various cards. Additionally, it should constuct the complete set of 52 cards. We note that shuing uses a random number generator, but a deck isnt the same kind of random event factory that a Wheel or pair of Dice is. In the case of Wheel and Dice, the random events were based on random selection with replacement : an individual event can be regenerated any number of times. In the case of a deck, however, once a card has been dealt, it will not show up until the deck is shued. In Roulette and Craps, the odds depended on a RandomEvent: either the Bin or Throw. In Blackjack, the antes win amount depends on the players entire hand. This means that being dealt an individual card isnt the same kind of thing that a throw of the dice is; rather it suggests that the dealers shoe is the random event factory, and the entire hand is the random event created by dealing cards. Continuing this line of thought, an Outcomes win amount could depend on a RandomEvent, if we consider the entire hand to be the random event. Considering an entire hand to be a single random event is skating on pretty thin ice, so we wont force-t a Deck into the random event factory part of our framework. Instead, we will let Deck stand alone. Well design a simple initialization for a deck that constructs the 52 cards. Beyond that, the notions of shuing and dealing can be assigned to the shoe. Since a Deck is a container, we have to examine the available collection classes to determine which concrete class we need. Interestingly, we only need two features of Collection: the add() method and the iterator(). These methods are implemented for all of the variations of Set and List. The only collection we can disregard is HashSet as that will use more storage than necessary. Shoe. The dealers shoe is where Cards are shued and dealt to create individual hands. A shoe will be built from some number of Decks. More properly, the shoe will contain the Cards from those Decks. The Deck objects arent really used for much more than constructing batches of individual Card objects. The Shoe responsibilities include proper initialization using a given number of decks, periodic shuing and proper dealing. In a casino, the shuing involves a ritual of spreading the cards on the table and stirring them around thouroughly, then stacking them back into the shoe. The top-most card is extracted, shown to the players, discarded, and a marker card is cut into the shoe at least two decks from the end, leaving about 100 cards unplayable after the marker. While most of the ritual does not require careful modeling, the presence of undealt cards at the end of the shoe is important as a way to defeat card-counting strategies. Since a Shoe is a container, we have to examine the available collection classes to determine which concrete class we need. Interestingly, we only need two features of Collection: the addAll() method to put another deck into the shoe and the iterator(). These methods are implemented for all of the variations of List. We will have to disregard the various Set implementations because they impose their own unique orders on the elements, dierent from our shued order. The simplest shuing algorithm iterates through all of the Cards in the Shoe s collection, and exchanges that Card with the card in a randomly-selection position. In order to move Card s around freely within the structure, an ArrayList or Vector could be used. See Card-Deck-Shoe Questions and Answers for more discussion on shuing.

38.2 Card-Deck-Shoe Questions and Answers

Why are there three subclasses of Card? Isnt it simpler to have one class and use an if-statement to sort out the point values. Primarily, there are three classes because they have dierent behaviors. Merging them into a single class and sorting out the behaviors with an if-statement is often a problem. 240 Chapter 38. Card, Deck and Shoe Classes

Building Skills in Object-Oriented Design, Release 2.1.1-Python

First, and most important, if-statements add complexity. The question wouldnt it be simpler to use an if-statement is a kind of oxymoron. Second, and almost as important, if-statements dilute responsibility assignments. Combining all three subclasses into one puts three slightly dierent responsibilities into one place, making it more diicult to debug problems. Further, we could wind up repeating or other reusing the if-statement in inappropriate ways. If we create separate subclasses, the clear separation of responsibility becomes a matter of denition, not a matter of following a complex thread of programming logic. Third, if-statements limit growth, adaptation and change. If we have a modication to the rules, for example, making 1-eyed Jacks wild, we would prefer to simply introduce another subclass. We nd that chasing down one or more related if-statements to assure ourselves that we are correctly handling the new subtlety rapidly gets out of hand. Is that the best shuing algorithm Wont it sometimes move a card twice? Wont it sometimes put a card back into the original spot? Yes, it may move some cards twice and it may leave a card in position. This is part of random behavior. This algorithm touches every card, swapping it with a randomly selected card. We are assured that every card was put into a random position. Sometimes a card will have been moved more than once, but the minimum criteria is that every card has been moved. While a shuing algorithm that models the real world is tempting, this adds complexity for no actual improvement in the randomization. A popular technique in the real world is to cut the deck in half and then rie the cards into a single pile. If done with the kind of perfection that software provides (cutting the deck exactly in half and exactly alternating the cards) this shue leads to a perfectly predictable cycle of orders. What makes this shue work in the real world is the random inaccuracies in cutting and riing. We dont see any value in modeling these physical phenomenon. A similar analysis holds for the kind of shue done in the casino. In essence, they do a shallow copy if the original List object, and then rebuild the shoes List by picking cards at random from the copy. This produces a result that is statistically indistinguishable from our algorithm, which uses an element-by-element swap. One fruitless side-track is using the seemingly-random hash code values of the Card objects. This only puts the cards into a single xed, but arbitrary order. An apparently interesting alternative is to generate a random index for each Card and then sort by this index or assemble a SortedSet. We note that sorting is O(n log n), where our algorithm is O(n), running much faster than any sort.

38.3 Card Superclass

class Card() Card denes a basic playing card. It has a rank, a suit, a hard point value and a soft point value. The point value methods are dened for the number cards from 2 to 10. Two subclasses handle face cards, where the point values are both 10, and aces, where the soft point value is 1, and the hard point value is 11. This class also denes symbolic names for the suits (Clubs, Diamonds, Hearts and Spades) and face cards (Jack, Queen and King). Here are Unicode characters for the suits.

38.3. Card Superclass

241

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Unicode U+02660 U+02663 U+02665 U+02666

Name BLACK BLACK BLACK BLACK

SPADE SUIT CLUB SUIT HEART SUIT DIAMOND SUIT

Symbol

Theres one tiny problem with the Unicode characters: the suits dont sort into the order preferred by Bridge players. Since its irrelevant for this class of problems, well ignore this. For the truly fussy, we would need to assure that the suits have a symbol character and a preferred sort order.

38.3.1 Fields
In Python, symbolic names are often declared within the class, not within the initialization method function.
class Card( object ): Clubs, Diamonds, Hearts, Spades = u'\u2663', u'\u2666', u'\u2665', u'\u2660' Jack, Queen, King = 11, 12, 13 ... oneEyedJack= Card( Card.Jack, Card.Hearts )

rank The rank of the card. This is a number from 1 (Ace) to 13 (King). suit The suit of the card. This is a character code for Clubs, Diamonds, Hearts or Spades.

38.3.2 Constructors
__init___(self, rank, suit ) Parameters rank (integer) rank of this card. suit (character) Character code for this card. Initializes the attributes of this Card.

38.3.3 Methods
getRank(self ) Returns the rank of this card. softValue(self ) Returns the soft value of this card. The superclass simply returns the rank. Subclasses can override this. Face cards will return 10, Aces will return 11. hardValue(self ) Returns the hard value of this card. The superclass simply returns the rank. Subclasses can override this. Face cards will return 10, Aces will return 1.

242

Chapter 38. Card, Deck and Shoe Classes

Building Skills in Object-Oriented Design, Release 2.1.1-Python

__str__(self ) Ideally, wed like this function to return the rank and suit of this card. However, in Python 2, theres an issue with the denition of the str() function. In Python 2, this function must return an ASCII-encoded string. If were using Unicode suit names, this function isnt helpful, and we have to resort to an explicit Unicode formatting string. unicode(self ) This will return a proper unicode string. We have to use print someCard.unicode() instead of print someCard.

38.4 FaceCard Class

FaceCard is a Card with a point value of 10. This denes jack, queens and kings.

38.4.1 Methods
softValue(self ) Returns the soft value of this card, 10. hardValue(self ) Returns the hard value of this card, 10. __str__(self ) Returns a short String displaying the rank and suit of this card. The ranks should be translated to single letters: 11 to 'J', 12 to 'Q' and 13 to 'K'.

38.5 AceCard Class

AceCard is a Card with a soft point value of 11 and a hard point value of 1. This dences Aces.

38.5.1 Methods
softValue(self ) Returns the soft value of this card, 11. hardValue(self ) Returns the hard value of this card, 1. __str__(self ) Returns a short String displaying the rank and suit of this card. The rank is always 'A'.

38.6 Deck class

Deck denes the standard deck of 52 cards. It both constructs the deck and acts as a container for one instance of a deck. 38.4. FaceCard Class 243

Building Skills in Object-Oriented Design, Release 2.1.1-Python

38.6.1 Fields
cards The collection of individual cards. The specic type of collection could be any of the Set or List implementation classes.

38.6.2 Constructors
__init__(self ) Creates the Collection, cards, and then creates the 52 cards. A simple nest pair of loops to iterate through the suits and ranks will work nicely for this.

Creating A Full Deck

For all four suits: Create the AceCard of this suit and a rank of 1; add to the cards For ranks 2 to 10: Create a Card of this suit and rank; add to the cards For ranks 11 to 13: Create a FaceCard of this suit and rank; add to the cards

38.6.3 Methods
getCards(self ) Returns the collection of cards in cards.

38.7 Shoe class

Shoe denes the dealers shoe, which contains from 1 to 8 decks of cards. For one deck shoes, one card is reserved as undealable. For multiple deck shoes from 1 to 3 decks can be left undealt. The exact number is selected at random within 6 cards of the expected number of decks.

38.7.1 Fields
deal An Interator that is used to pick the next card from the shoe. stopDeal The approximate number of decks to be left undealt in the shoe.

244

Chapter 38. Card, Deck and Shoe Classes

Building Skills in Object-Oriented Design, Release 2.1.1-Python

38.7.2 Constructors
__init__(self, decks, stopDeal, rng=None) If rng is None, create a generator from random.Random. Parameters decks (integer) A number of decks to create stopDeal (integer) An approximate number of decks left undealt in the shoe rng (random.Random) A random number generator to use. Initializes the Shoe by creating the required number of decks and building the cards List. This saves the stopDeal value, which is the number of decks left in the shoe. Typically, this is two, and approximately 104 cards are left in the shoe. This also saves the random number generator used to shue. To facilitate testing, this initializes the dealing iterator to the unshued List of cards. This will produce cards in a xed order.

38.7.3 Methods
shuffle(self ) Shues the shoe by swapping every element in the Shoe.cards List with a random element. Creates an Iterator, Shoe.deal that can be used to deal cards. If the stopDeal is non-zero, do the following to exclude several decks from the deal. 1. Create a random number, v, such that 6 v 6. 2. Step through the deal iterator stop 52 + v times, removing approximately stop cards from being dealt as part of ordinary play. __iter__(self ) Return the Shoe.deal iterator so that the game can get cards from the Shoe.

38.8 Card-Deck-Shoe Deliverables

There are ten deliverables for this exercise. The three classes of the Card class hierarchy, including FaceCard and AceCard. A class which performs a unit tests of the Card class hierarchy. The unit test should create several instances of Card, FaceCard and AceCard. The Deck class. A class which performs a unit test of the Deck class. This simply creates a Deck object and conrms the total number of cards. A thorough test would also check some individual Card objects in the cards collection. The Shoe class. A class which performs a unit test of the Shoe class. This simply creates a Shoe object and conrms that it deals cards. In order to test the shuffle() method, you will need to construct the Shoe with a random number generator that has a xed seed and produces cards in a known sequence.

38.8. Card-Deck-Shoe Deliverables

245

Building Skills in Object-Oriented Design, Release 2.1.1-Python

246

Chapter 38. Card, Deck and Shoe Classes

CHAPTER

THIRTYNINE

HAND AND OUTCOME CLASSES

This chapter introduces the hand, and the problems of scoring the value of the hand. It also introduces a subclass of Outcome that can handle the more complex evaluation rules for Blackjack.

39.1 Hand Overview

The hand of cards is both a container for cards, but is also one dimension of the current state of the players playing strategy. A player may have multiple hands, and the hands are resolved independently. For each hand, the responsibilities include collecting the cards, producing a hard and soft point total and determining the appropriate payout odds. Collecting cards is trivial. Each hand is a simple bag or multiset of cards. A hand cant be any kind of Set because there can easily be duplicate cards in a multiple deck game. In an 8-deck game, there are 8 separate instances of A . Determining the payout odds for a Hand is somewhat more complex.

39.1.1 Payout Odds

To be compatible with other games, well be creating Bet objects which are associated with Outcomes. In Roulette, there were a profusion of Outcomes, each relatively simple with xed odds. In Craps, there were fewer Outcomes, some of which had odds that depended on a throw of the dice. The players Hand must be associated with an Outcome so that we can match the Outcome of a Bet and the Outcome of a Hand to determine the payout. In the case of Roulette, each Bin was simply a set of winning Outcomes. In the case of Craps, both the dice and the game had a mixtures of winning, losing and unresolved Outcomes. The Bets were simply associated with Outcomes and the Wheel, Dice or Game gave us sets of winning (and losing) Outcomes. In Blackjack, there are relatively few outcomes. And its not clear how each Outcome associates with a Hand. Survey of Outcomes. To gure out how to associate Hand and Outcome, well start by enumerating all the individual Outcomes. 1. Insurance at 2:1. This is a winner when the dealers hand is blackjack; and the Ante bet will be a loser. It is only oered when the up card is an Ace. This is a loser when the dealers hand is not blackjack, and the Ante bet is unresolved. 2. Even Money at 1:1. This is oered in the rare case of the players hand is blackjack and the dealers up card being an Ace. If accepted, it can be looked at as a switch of the Ante bet to an Even Money outcome, which is then resolved as a winner. 247

Building Skills in Object-Oriented Design, Release 2.1.1-Python

3. Ante paying 1:1. This payout occurs when the players hand is less than or equal to 21 and the dealers hand goes over 21. This payout also occurs when the players hand is less than or equal to 21 and greater than the dealers hand. This outcome is a loser as soon as the players hand goes over 21. It is also a loser when the players hand is less than or equal to 21 and less than the dealers hand. 4. Ante paying 3:2. This payout occurs when the players hand is blackjack. The odds depend on the players hand. 5. Ante as a push, paying 0:1. This payout occurs when the players hand is less than or equal to 21 and equal to the dealers hand. The odds depend on both player and dealers hand. In Craps, we identied the Field and Horn Outcomes which had winning amounts that depended on the specic Throw. See Throw Builder Overview , for more information. Additionally, we compared Bets against the Outcomes as part of the transition from one GameState to another. See Resolving Bets, for more information. In order for this to work correctly, some bets were changed frm a generic Outcome (a Line Bet) to an Outcome with dierent odds. What Class is Hand? It appears that the hand, as a whole, is not simply associated with an Outcome. It appears that a Hand must produce an Outcome based on its content, the dealers content, and possibly the state of the game. This is a small change from the way Dice and Bin work. Those classes were associated with an Outcome. A Blackjack Hand, however, must do a bit of processing to determine which Outcome it represents. A two-card hand totalling soft 21 produces a blackjack Outcome that pays 3:2. All other hands produce an Outcome that pays 1:1 and could be resolved as a win, a loss or are a push. Also, changes to the state of the game depend on the values of both hands, as well as the visible up card in the dealers hand. This makes the state of the hand part of the evolving state of the game, unlike the simple RandomEvents we saw in Roulette and Craps. Choices. We have a few ways we can deal with Hand. We can make Hand a subclass of RandomEvent, even though its clearly more complex than other events. We can make Hand a unique kind of class, unrelated to other games. Hand is an Event? While a Hand appears to be a subclass of RandomEvent, it jars our sensibilities. A Hand is built up from a number of Cards. One could rationalize calling a Hand an event by claiming that the Shoe is the random event factor. The act of shuing is when the event is created. The complex event is then revealed one card at a time. It seems that we need to dene a Hand class that shares a common interface with a RandomEvent, but extends the basic concept because a hand has an evolving state. Hand is dierent. While we can object to calling a Hand a single event, its diicult to locate a compelling reason for making a Hand into something radically dierent from the other subclasses of RandomEvent: Bin and Dice. Hand Features. Our rst design decision, then is to dene Hand as a kind of RandomEvent. Well need to create several Outcomes: Insurance, Even Money, Ante and Blackjack. The Hand will produce an appropriate Outcome based on the hands structure, the game state, and the dealers hand. Generally, each Hand will produce a simple Ante outcome as a winner or loser. Sometimes a Hand will produce a Blackjack outcome. Sometimes the Player and Blackjack Game will collaborate to add an Insurance or Even Money outcome to the Hand.

248

Chapter 39. Hand and Outcome Classes

Building Skills in Object-Oriented Design, Release 2.1.1-Python

39.1.2 Hard and Soft Totals

Our second design problem is to calculate the point value of the hand. Because of aces, hands can have two dierent point totals. If there are no aces, the total is hard. When there is an Ace, there are two totals. The hard total uses Aces as 1. The soft total uses Aces as 11. We note that only one ace will participate in this hard total vs. soft total decision-making. If two aces contribute soft values, the hand is at least 22 points. Therefore, we need to note the presence of at least one ace to use the soft value of 11, all other cards will contribute their hard values to the hands total value. A Hand has a two states. Hard. When the hand has no aces, there is one total. Also, when a hand has one or more aces, but the soft total is over 21, then the hand only has a hard total. Soft. When a hand has at least one Ace and the soft total is 21 or less, there is a hard and a soft total. A hand with no cards could be said to be in the hard total state. As each card is added, the state may change. It may change to soft when an Ace is added and the soft total will be 21 or less. It may changes to hard when the soft total is over 21. State Change Rules. We can see that our Hand state change is split among the various subclass of Card. The Card (which will be inhereted by FaceCard) we need to check the Hand hard total. If the hard total is over 21, the state change method needs to switch the Hand to the hard total state. Otherwise, it leaves the state alone. The AceCard, however, must use a slightly dierent algorithm. If the hard total is over 21, the state change method needs to switch the Hand to the hard total state. Otherwise, the hard total is 21 or under, and the state needs to switch the Hand to the soft total state. Since there are only two distinct total algorithms, we can actually put both of them right into each Hand object and use simple assignment statements to choose which algorithm is in force at any given state of the Hand.

39.1.3 Blackjack Outcomes

As a nal design decision, we need to consider creating any subclass esof Outcome to handle the variable odds for the Ante bet. We dont need a subclass for the Insurance or Even Money, because the base Outcome does everything we need. The notable complication here is that there are three dierent odds. If the players hand beats the dealers hand and is blackjack, the odds are 3:2. If the players hand beats the dealers hand, but is not blackjack, the odds are 1:1. If the players hand equals the dealers hand, the result is a push, eectively 0:1 odds. It doesnt seem like a new class of Outcomes are necessary. We simply have some alternative outcomes that can be produced by a Hand. The player can only place one of four basic bets. The Ante bet that starts play. This is assmed to be 1:1 unless game conditions change this to 3:2 or a push. This is the essential Outcome for the players primary Bet. The Insurance and Event Money bets. One of these may be active after the rst cards are dealt. For the insurance Outcome to be active, the dealer must be showing an Ace and the players hand is not 21. For even money Outcome to be active, the dealer must be showing an Ace and the players hand is 21.

39.1. Hand Overview

249

Building Skills in Object-Oriented Design, Release 2.1.1-Python

These are resolved immediately if the dealer does not have 21, these bets are lost. If the dealer has 21, these bets win, but the Ante is a loss. The Double Down bet. This is generally oered at any time. It can be looked at as an an additional amount added to the ante bet and a modication to game play. Objects. It seems simplest to create a few Outcome instances: Ante, Insurance, Even Money and Double Down. The Table will have to merge the double-down bet amount into the Ante bet amount.

39.2 Hand Total Class Design

HandTotal computes a total of the cards in a hand. There are distinct subclasses for the two algorithms.

39.2.1 Constructor
__init__(self, hand ) Parameter hand (Hand) The hand for which to compute a total Creates a new HandTotal object associated with a given hand.

39.2.2 Methods
total(self, card=None) Computes a total of all the cards in the associated hand. If card is not None, omit the the indicated card from the total. This method is abstract, it should return NotImplemented. Each subclass will provide an implementation. Parameter card (Card) A card to exclude from the total

39.3 Hand Hard Total Class Design

class HandHardTotal() Computes a hard total of the cards in a hand. total(self, card=None) Computes the hard total of all the cards in the associated hand. If card is not None, omit the the indicated card from the total. Parameter card (Card) A card to exclude from the total

39.4 Hand Soft Total Class Design

class HandSoftTotal() Computes a soft total of the cards in a hand.

250

Chapter 39. Hand and Outcome Classes

Building Skills in Object-Oriented Design, Release 2.1.1-Python

total(self, card=None) Computes the soft total of all the cards in the associated hand. If card is not None, omit the the indicated card from the total. Parameter card (Card) A card to exclude from the total

39.5 Card Class Updates

Each subclass of Card needs to provide a method that sets the hands total algorithm. setAltTotal(hand ) Parameter hand (Hand) The hand for which to set a total algorithm There are two dierent implementations for this method in Card, and AceCard. Note that each implementation diers only in what is done with the soft total is 21 or less. With a little care, this single dierence can be factored out into the AceCard subclass.

39.5.1 Card Class

To determine what alternative total object is appropriate for this hand, do the following. 1. Get the soft total of all cards except this one. 2. Add the soft points of this card. This happens to be the same as the hard points, but a good program doesnt repeat this piece of information; it uses the Card.softValue() method. 3. If this is over 21, then set the Hand.altTotal to to Hand.hard. If the soft total is 21 or less, then the Hand.altTotal is left untouched.

39.5.2 AceCard Class

To determine what alternative total object is appropriate for this hand, do the following. 1. Get the soft total of all cards except this one. 2. Add the soft points of this card. This happens to be 11 for an Ace, but a good program doesnt repeat this piece of information; it uses the Card.softValue() method. 3. If this is over 21, then set the Hand.altTotal to to Hand.hard. Otherwise, the soft total is 21 or less, then set the Hand.altTotal to Hand.soft.

39.6 Hand Class Design

class Hand() Hand contains a collection of individual Cards, and determines an appropriate total point value for the hand.

39.5. Card Class Updates

251

Building Skills in Object-Oriented Design, Release 2.1.1-Python

39.6.1 Fields
cards Holds the collection of individiual Cards of this hand. hard A instance of HandHardTotal. This means that hand.hard.total() will produce the hard total of the cards in this hand. This method is used as one of the point totals for the hand. soft A instance of HandSoftTotal. This means that hand.soft.total() will produce the soft total of the cards in this hand. altTotal This is a reference to either Hand.hardTotal or Hand.softTotal. This will produce a soft total when one is a appropriate; when a soft total isnt appropriate, it will produce a hard total. This is set by each Card. This method is used as one of the point totals for the hand.

39.6.2 Constructors
__init__(self, card=None) Parameter card (Card) A card to add Creates an empty hand. The Hard.cards variable is initialized to an empty sequene. The Hand.hard and Hand.soft objects are initialized to HandHardTotal and HandSoftTotal objects. Also, Hand.altTotal is set to Hand.hard. If card is provided, then use the add() method to add this card to the hand..

39.6.3 Methods
add(self, card ) Parameter card (Card) A card to hadd Add this card to the Hand.cards list. Evaluate Card.setAltTotal() to update the hard vs. soft total calculation for this hand. value(self ) Computes the alternate total of this hand using the Hand.altTotal object. If there are any aces, and the soft total is 21 or less, this will be the soft total. If there are no aces, or the soft total is over 21, this will be the hard total. size(self ) Returns the number of cards in the hand, the size of the List. blackjack(self ) Returns true if this hand has a size of two and a value of 21.

252

Chapter 39. Hand and Outcome Classes

Building Skills in Object-Oriented Design, Release 2.1.1-Python

busted(self ) Returns true if this hand a value over 21. __iter__(self ) Returns an iterator over the cards of the List. __str__(self ) Displays the content of the hand as a String with all of the card names.

39.7 Hand Deliverables

There are six deliverables for this exercise. The HandTotal class hierarchy: HandTotal, HandHardTotal, HandSoftTotal. A unit test for each of these classes. The Hand class. A class which performs a unit tests of the Hand class. The unit test should create several instances of Card, FaceCard and AceCard, and add these to instances of Hand, to create various point totals. The Card and AceCard modications required to set the appropriate values in a Hand A set of unit tests for assembling a hand and changing the total object in use to correctly compute hard or soft totals for the hand.

39.7. Hand Deliverables

253

Building Skills in Object-Oriented Design, Release 2.1.1-Python

254

Chapter 39. Hand and Outcome Classes

CHAPTER

FORTY

BLACKJACK TABLE CLASS

The bets in Blackjack are associated with a hand. This will lead us to create a subclass of table to handle this complexity. In order to manage the relationships between hand and bet, well rework hand, also.

40.1 Blackjack Table Overview

When we look at the game of Blackjack, we note that a players Hand can be split. In some casinos, resplits are allowed, leading to the possibility of 3 or more Hands. Each individual Hand has a separate ante Bet and seperate resolution. This is dierent from the way bets are currently managed for Roulette and Craps. We have several alternatives. Assign responsibility to the Table to keep track of bets by Player tied to the various Hands. This would make the Hand a potential key into a Map that associated a Hand with a Bet. However, Hands change state, making them poor choices for keys to a Map. We could put a reference to a Hand into the Bet. In this way, as each Bet is resolved, the relevant Hand can be evaluated. We could put a reference to the Ante Bet in the Hand. In this way, as each Hand is resolved, the relevant Bet can be paid or lost. Well design Hand to contain the associated ante Bet. This is least disruptive to the Bet which is a simple thing used widely in other games. Additional Bets. While most Bets are associated with a specic Hand, the insurance Bet is always resolved before an additional hand can be created. There doesnt seem to be an essential association between the initial Hand and the insurance Bet. We can treat insurance as a Bet that follows the model established for Craps and Roulette. Currently, Bets are placed on the Table. If we create a subclass named BlackjackTable that uses a Hand when creating a Bet, we can have this method do both tasks: it can attach the Bet to the Hand, and it can save the Bet on the Table.

40.2 BlackjackTable Class

class BlackjackTable() BlackjackTable is a Table that handles the additional association between Bets and specic Hands in Blackjack.

255

Building Skills in Object-Oriented Design, Release 2.1.1-Python

40.2.1 Constructors
__init__(self ) Uses the superclass constructor to create an empty Table.

40.2.2 Methods
placeBet(self, bet, hand ) Parameters bet (Bet) A bet for this hand; an ante bet is required. An insurance bet, even money bet or double down bet is optional. hand (Hand) A hand on which the player is creating a Bet. Updates the given hand to reference the given bet. Then uses the superclass placeBet() to add this bet to the list of working bets. __str__(self )

40.3 Hand Rework

Hand contains a collection of individual Cards, and determines an appropriate total point value for the hand. We need to add a eld and some appropriate methods for associating a Bet with a Hand.

40.3.1 Fields
ante Holds a reference to the ante Bet for this hand. When this hand is resolved, the associated bet is paid and removed from the table.

40.3.2 Methods
setBet(self, ante) Parameter ante (Bet) The initial bet required to play Sets the ante Bet that will be resolved when this hand is nished. getBet(self ) Returns the ante Bet for this hand.

40.4 Blackjack Table Deliverables

There are four deliverables for this exercise. The revised Hand class. A class which performs a unit tests of the Hand class. The unit test should create several instances of Card, FaceCard and AceCard, and add these to instances of Hand, to create various point totals. Additionally, the unit test should create a Bet and associate it with the Hand. 256 Chapter 40. Blackjack Table Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

The BlackjackTable class. A class which performs a unit tests of the BlackjackTable class. The unit test should create several instances of Hand and Bet to create multiple Hands, each with unique Bets.

40.4. Blackjack Table Deliverables

257

Building Skills in Object-Oriented Design, Release 2.1.1-Python

258

Chapter 40. Blackjack Table Class

CHAPTER

FORTYONE

BLACKJACK GAME CLASS

After reviewing the procedure provided in A Walkthrough, well dene the basic game class for Blackjack. This will require a stub class for a Blackjack Player. Well also revisit the fundamental relationship between Game, Hand and Player. Well invert our viewpoint from the Player containing a number of Hands to the Hands sharing a common Player.

41.1 Blackjack Game Overview

The sequence of operations in the game of Blackjack is quite complex. We can describe the game in either of two modes: as a sequential procedure or as a series of state changes. A sequential description means that the state is identied by the step that is next in the sequence. The state change description is what we used for Craps, see Craps Game. Each state denition included a set of methods that represented conditions that could change state; and each Throw object invoked one of those methods, in eect, announcing a condition that caused a state change. Additionally, we need to look at the various collaborations of the Game. We also need to address the question of handling the dealers rules. Maintaining State. The sequential description of state, where the current state is dened by the step that is next, is the default description of state in most programming languages. While it seems obvious beyond repeating, it is important to note that each statement in a method changes the state of the application; either by changing state of the object that contains the method, or invoking methods of other, collaborating objects. In the case of an active class, this description of state as next statement is adequate. In the case of a passive class, this description of state doesnt work out well because passive classes have their state changed by collaborating objects. For passive objects, instance variables and state objects are a useful way to track state changes. In the case of Roulette, the cycle of betting and the game procedure were a simple sequence of actions. In the case of Craps, however, the game was only loosely tied to each the cycle of betting and throwing the dice, making the game state a passive part of the cycle of play. In the case of Blackjack, the cycle of betting and game procedure are more like Roulette. Most of a game of Blackjack is simply sequential in nature: the initial oers of even money, insurance and splitting the hands are optional steps that happen in a dened order. When lling the players Hands, there are some minor sub-states and state changes. Finally, when all of the players Hands are bust or standing pat, the dealer lls their hand. Finally, the hands are resolved with no more player intervention. Most of the game appears to be a sequence of oers from the Game to the BlackjackPlayer; these are oers to place bets, or accept cards, or a combination of the two, for each of the players Hands.

259

Building Skills in Object-Oriented Design, Release 2.1.1-Python

41.2 Blackjack Collaboration

In Craps and Roulette, the Player was the primary collaborator with the Game. In Blackjack, however, focus shifts from the Player to the Hand. This changes the responsibilities of a BlackjackPlayer: the Hand can delegate certain oers to the BlackjackPlayer for a response. The BlackjackPlayer can become a plug-in strategy to the Hand, providing responses to oers of insurance, even money, splitting, doubling-down, hitting and standing pat. The BlackjackPlayers response will change the state of the Hand. Some state changes involve getting a card, and others involve placing a bet, and some involve a combination of the two. Well use the procedure denition in A Walkthrough. Following this procedure, we see the following methods that a Hand and a BlackjackPlayer will need to respond to the various oers from the BlackjackGame. The rst portion of the game involves the BlackjackPlayer, the second portion invovles one or more Hand s. The collaboration is so intensive, we have created a kind of swimlane table, showing the operations each object must perform. This will allow us to expand Hand and BlackjackTable as well as dene the interface for BlackjackPlayer. Table 41.1: Blackjack Overall Collaboration BlackjackGame calls players placeBet Hand BlackjackPlayer creates empty Hand, creates initial Ante Bet BlackjackTable accepts the ante bet, associate with the hand

gets the initial hand; deal 2 cards to hand deal 2 cards to dealers hand gets up card from dealers hand; if this card requires insurance, do the insurance procedure iterate through all hands; is the given hand splittable? if the hand is splittable, oer a split bet

add cards add cards return up card

returns the initial hand

if splitting, move card and deal cards; loop back, looking for split oers iterate through all hands; if the hand is less than 21, do the ll-hand procedure while the dealers point value is 16 or less, deal another card

return true if two cards of the same rank get the players response; return it to the game take a card out; add a card

returns a list iterator

to split: create a split bet, and an empty hand; return the new hand

accept the split bet for the new hand

returns a list iterator return point value of the hand; add a card return point value of the hand return point value of the hand

if the dealer busts, iterate through all hands resolving the ante as a winner if the dealer does not bust, iterate through all hands comparing against the dealers points, determining win, loss or push

returns a list iterator

resolve bet as winner and remove resolve bet and remove

There are a few common variation in this procedure for play. Well set them aside for now, but will visit them in Variant Game Rules. 260 Chapter 41. Blackjack Game Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Insurance. The insurance procedure involves additional interaction between Game and the the Players initial Hand. The following is done only if the dealer is showing an ace. Table 41.2: Blackjack Insurance Collaboration BlackjackGame if players hand is blackjack: oer even money Hand return true if 2 cards, soft 21 BlackjackPlayer to accept, return true BlackjackTable

if player accepted even money oer: change bet, resolve; end of game oer insurance to accept, create new bet; return true return point value

update bet; resolve and remove bet accept insurance bet resolve and remove bet

if player accepted insurance oer: check dealers hand; if blackjack, insurance wins, ante loses, game over; otherwise insurance loses

Filling the Hands. The procedure for lling each Hand involves additional interaction between Game and the the Players initial Hand. An Iterator used for perform the following procedure for each individual player Hand. Table 41.3: Blackjack Fill-Hand Collaboration BlackjackGame Hand BlackjackPlayer BlackjackTable resolve and remove bet update bet resolve the ante as a loss

if players hand is blackjack: resolve ante bet

return true if 2 cards, soft 21

while points less than 21, oer play options of double or hit; rejecting both oers is a stand. if over 21, the hand is a bust

return point value; pass oers to player return point value

to double, increase the bet for this hand and return true; to hit, return true

There is some variation in this procedure for lling Hands. The most common variation only allows a double-down when the Hand has two cards. Hand-specic Decisions. Some of the oers are directly to the BlackjackPlayer, while others require informing the BlackjackPlayer which of the players Hands is being referenced. How do we identify a specic hand? One choice is to have the BlackjackGame make the oer to the Hand. The Hand can pass the oer to the BlackjackPlayer; the Hand includes a reference to itself. An alternative is to have the BlackjackGame make the oer directly to the BlackjackPlayer, including a reference to the relevant Hand. While the dierence is minor, it seems slightly more sensible for the BlackjackGame to make oers directly to the BlackjackPlayer, including a reference to the relevant Hand.

41.2. Blackjack Collaboration

261

Building Skills in Object-Oriented Design, Release 2.1.1-Python

41.3 Dealer Rules

In a sense, the dealer is a special player. They have a xed set of rules for hitting and standing. They are not actually oered an insurance bet, nor can they split or double down. However, the dealer does participate in the hand-lling phase of the game, deciding to hit or stand pat. The dealers rules are quite simple. Should the Dealer be a special subclass of BlackjackPlayer; one that implements only the dealers rules? Or should the Dealer be a feature of the Game. In this case, the Game would maintain the dealers Hand and execute the card-lling algorithm. Using an subclass of BlackjackPlayer is an example of Very Large Hammer design pattern. We only want a few features of the BlackjackPlayer class. Refactoring. To avoid over-engineering these, we could refactor player into two components. An object that handles hand-lling, and an object that handles betting strategies. The dealer would only use the hand-lling component of a player. Mutability. To avoid over-engineering this, we can look at features that are likely to change. The dealer hand-lling rules seem well-established throughout the industry. Further, a change to the hand-lling rules of the dealer would change the nature of the game enough that we would be hard-pressed to call in Blackjack. A dierent hand-lling rule would constitute a new kind of game. Were condent, then, that the dealers hand can be a feature of the BlackjackGame class.

41.4 BlackjackPlayer Class

class BlackjackPlayer() BlackjackPlayer is a subclass of Player that responds to the various queries and interactions with the game of Blackjack.

41.4.1 Fields
hand Some kind of List which contains the initial Hand and any split hands that may be created.

41.4.2 Constructors
__init__(self, table) Parameter table (BlackjackTable) The table on which bets are placed Uses the superclass to construct a basic Player. Uses the newGame() to create an empty List fot the hands.

41.4.3 Methods
newGame(self )

262

Chapter 41. Blackjack Game Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Creates a new, empty list in which to keep Hands. placeBets(self ) Creates an empty Hand and adds it to the List of Hands. Creates a new ante Bet. Updates the Table with this Bet on the initial Hand. getFirstHand(self ) Returns the initial Hand. This is used by the pre-split parts of the Blackjack game, where the player only has a single Hand. __iter__(self ) Returns an iterator over the List of Hands this player is currently holding. evenMoney(self, hand ) Parameter hand (Hand) the hand which is oered even money Returns true if this Player accepts the even money oer. The superclass always rejects this oer. insurance(self, hand ) Parameter hand (Hand) the hand which is oered insurance Returns true if this Player accepts the insurance oer. In addition to returning true, the Player must also create the Insurance Bet and place it on the BlackjackTable. The superclass always rejects this oer. split(self, hand ) Parameter hand (Hand) the hand which is oered an opportunity to split If the hand has two cards of the same rank, it can be split. Dierent players will have dierent rules for determine if the hand should be split ot not. If the players rules determine that it wil accepting the split oer for the given Hand, hand, then the player will 1. Create a new Ante bet for this hand. 2. Create a new one-card Hand from the given hand and return that new hand. If the players rules determine that it will not accept the split oer, then None or null is returned. If the hand is split, adding cards to each of the resulting hands is the responsibility of the Game. Each hand will be played out independently. doubleDown(self, hand ) Parameter hand (Hand) the hand which is oered an opportunity to double down Returns true if this Player accepts the double oer for this Hand. The Player must also update the Bet associated with this Hand. This superclass always rejects this oer. hit(self, hand ) Parameter hand (Hand) the hand which is oered an opportunity to hit Returns true if this Player accepts the hit oer for this Hand. The superclass accepts this oer if the hand is 16 or less, and rejects this oer if the hand is 17 more more. This mimics the dealers rules. Failing to hit and failing to double down means the player is standing pat. __str__(self )

41.4. BlackjackPlayer Class

263

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Displays the current state of the player, and the various hands.

41.5 Card Rework

Card must provide the Game some information required to oer insurance bets. Well need to add an offerInsurance() method on the class Card. The Card superclass must respond with False. This means that the FaceCard subclass will also respond with False. The AceCard subclass, however, must respond with True to this method.

41.6 Hand Rework

Hand should retain some additional hand-specic information. Since some game allow resplitting of split hands, its helpful to record whether or not a player has declined or accepted the oer of a split.

41.6.1 Fields
player Holds a reference to the Player who owns this hand. Each of the various oers from the Game are delegated to the Player. splitDeclined Set to true if split was declined for a splittable hand. Also set to true if the hand is not splittable. The split procedure will be done when all hands return true for split declined.

41.6.2 Methods
splittable(self ) Returns true if this hand has a size of two and both Cards have the same rank. Also sets Hand.splitDeclined to true if the hand is not splittable. getUpCard(self ) Returns the rst Card from the list of cards, the up card.

41.7 BlackjackGame Class

class BlackjackGame() BlackjackGame is a subclass of Game that manages the sequence of actions that dene the game of Blackjack. Note that a single cycle of play is one complete Blackjack game from the initial ante to the nal resolution of all bets. Shuing is implied before the rst game and performed as needed.

264

Chapter 41. Blackjack Game Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

41.7.1 Fields
shoe This is the dealers Shoe with the available pool of cards. dealer This is the dealers Hand.

41.7.2 Constructors
__init__(self, shoe, table) Parameters shoe (Shoe) The dealers shoe, populated with the proper number of decks table (BlackjackTable) The table on which bets are placed Constructs a new BlackjackGame, using a given Shoe for dealing Cards and a BlackjackTable for recording Bets that are associated with specic Hands.

41.7.3 Methods
cycle(self ) A single game of Blackjack. This steps through the following sequence of operations. 1. Call BlackjackPlayer.newGame() to reset the player. Call BlackjackPlayer.getFirstHand() to get the initial, empty Hand. Call Hand.add() to deal two cards into the players initial hand. 2. Reset the dealers hand and deal two cards. 3. Call BlackjackGame.hand.getUpCard() to get the dealers up card. If this card returns true for the Card.offerInsurance(), then use the insurance() method. Only an instance fo the subclass AceCard will return true for offerInstance(). All other Card classes will return false. 4. Iterate through all Hands, assuring that no hand it splittable, or split has been declined for all hands. If a hand is splittable and split has not previously been declined, call the Hands split() method. If the split() method returns a new hand, deal an additional Card to the original hand and the new split hand. 5. Iterate through all Hands calling the fillHand() method to check for blackjack, deal cards and check for a bust. This loop will nish with the hand either busted or standing pat. 6. While the dealers hand value is 16 or less, deal another card. This loop will nish with the dealer either busted or standing pat. 7. If the dealers hand value is bust, resolve all ante bets as winners. The OutcomeAnte should be able to do this evaluation for a given Hand compared against the dealers bust. 8. Iterate through all hands with unresolved bets, and compare the hand total against the dealers total. The OutcomeAnte should be able to handle comparing the players hand and dealers total to determine the correct odds. insurance(self )

41.7. BlackjackGame Class

265

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Oers even money or insurance for a single game of blackjack. This steps through the following sequence of operations. 1. Get the players BlackjackPlayer.getFirstHand(). Is it blackjack? If the player holds blackjack, then call BlackjackPlayer.evenMoney(). If the even money oer is accepted, then move the ante bet to even money at 1:1. Resolve the bet as a winner. The bet will be removed, and the game will be over. 2. Call BlackjackPlayer.insurance(). If insurance declined, this method is done. 3. If insurance was accepted by the player, then check the dealers hand. Is it blackjack? If the dealer hold blackjack, the insurance bet is resolved as a winner, and the ante is a loser; the bets are removed and the game will be over. If the dealer does not have blackjack, the insurance bet is resolved as a loser, and the ante remains. If insurance was declined by the player, nothing is done. fillHand(self, hand ) Parameter hand (Hand) the hand which is being lled Fills one of the players hands in a single game of Blackjack. This steps through the following sequence of operations. 1. While points are less than 21, call BlackjackPlayer.doubleDown() to oer doubling down. If accepted, deal one card, lling is done. If double down is declined, call BlackjackPlayer.hit() to oer a hit. If accepted, deal one card. If both double down and hit are declined, lling is done, the player is standing pat. 2. If the points are over 21, the hand is bust, and is immediately resolved as a loser. The game is over. __str__(self ) Displays the current state of the game, including the player, and the various hands.

41.8 Blackjack Game Deliverables

There are eight deliverables for this exercise. The stub BlackjackPlayer class. A class which performs a unit test of the BlackjackPlayer class. Since this player will mimic the dealer, hitting a 16 and standing on a 17, the unit test can provide a variety of Hand s and conrm which oers are accepted and rejected. The revised Hand class. A class which performs a unit tests of the Hand class. The unit test should create several instances of Card, FaceCard and AceCard, and add these to instances of Hand, to create various point totals. Since this version of Hand interacts with a BlackjackPlayer, additional oers of split, double, and hit can be made to the Hand. The revised Card class. Revised unit tests to exercise the Card.offerInsurance() method. The revised BlackjackGame class. 266 Chapter 41. Blackjack Game Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A class which performs a unit tests of the BlackjackGame class. The unit test will have to create a Shoe that produces cards in a known sequence, as well as BlackjackPlayer. The cycle() method, as described in the design, is too complex for unit testing, and needs to be decomposed into a number of simpler procedures.

41.8. Blackjack Game Deliverables

267

Building Skills in Object-Oriented Design, Release 2.1.1-Python

268

Chapter 41. Blackjack Game Class

CHAPTER

FORTYTWO

SIMPLE BLACKJACK PLAYER CLASS

Our objective is to provide variant player strategies. This chapter will upgrade our stub player class to give it a complete, working strategy. This simple player can serve as the superclass for more sophisticated strategies.

42.1 Blackjack Player Overview

In addition to the players own hand, the player also has the dealers up card available for determining their response to the various oers. The player has two slightly dierent goals: not bust and have a point total larger than the dealers. While there is some overlap between these goals, these lead to two strategies based on the dealers up card. When the dealer has a relatively low card (2 through 6), the dealer has an increased probability of going bust, so the players strategy is to avoid going bust. When the dealer has a relatively high card (7 through 10), the dealer will probably have a good hand, and the player has to risk going bust when looking for a hand better than the dealers. A Simple Strategy. Well provide a few rules for a simple player strategy. This strategy is not particularly good. Any book on Blackjack, and a number of web sites, will have a superior strategy. A better strategy will also be considerably more complex. Well implement this one rst, and leave it to the student to research more sophisticated strategies. 1. Reject insurance and even money oers. 2. Accept split for aces and eights. Reject split on other pairs. 3. Hit any hand with 9 or less. The remaining rules are presented in the following table. Table 42.1: Blackjack Player Strategy Player Shows 10 or 11 hard 12 to 16 soft 12 to 16 17 to 21 2-6 hit stand hit stand 7-10, Ace double down hit hit stand

These rules will boil down to short sequences of if-statements in the split(), hit() and doubleDown() methods. In some contexts, complex if-statements are deplorable. Specically, complex if-statements are often a standin for proper allocation of responsibility. In this class, however, the complex if-statements implement a kind of index or lookup scheme. We have, for this exercise, 8 alternatives which depend on a two-dimensional index. One dimension contains four conditions that describe the players hand. The other dimension involves two conditions that describe the dealers hand. When we look at the various collections, we see that we can

269

Building Skills in Object-Oriented Design, Release 2.1.1-Python

index by primitive types or object instances. In this case, we are indexing by conditions; we would have to map each condition to either a numeric code or a distinct object in order to eliminate the if-statements. When we look at the conditions that describe the players hand, these are clearly state-like objects. Each card can be examined and a state transition can be made based on the the current state and the card. After accepting a card, we would check the total and locate the appropriate state object. We can then use this state object to index into a collection. When we look at the conditions that describe the dealers hand, there are only two state-like objects. The dealers op card can be examined, and we can locate the appropriate state object. We can use this state object to index into a collection. The nal strategy could be modeled as a collection with a two-part index. This can be nested collection objects, or a Map that uses a 2-valued tuple as an index.

42.2 SimpleBlackjackPlayer Design

class SimpleBlackjackPlayer() SimpleBlackjackPlayer is a subclass of BlackjackPlayer that responds to the various queries and inteactions with the game of Blackjack. This player implements a very simple strategy, shown above in the Blackjack Player Strategy table.

42.2.1 Methods
evenMoney(self, hand ) Parameter hand (Hand) the hand which is being oered even money Returns true if this Player accepts the even money oer. This player always rejects this oer. insurance(self, hand ) Parameter hand (Hand) the hand which is being oered insurance Returns true if this Player accepts the insurance oer. This player always rejects this oer. split(self, hand ) Parameter hand (Hand) the hand which is being oered the opportunity to split Returns a new, empty Hand if this Player accepts the split oer for this Hand. The Player must create a new Hand, create an Ante Bet and place the on the new Hand on the BlackjackTable. If the oer is declined, both set Hand.splitDeclined to true and return null. This player splits when the hands cards ranks are aces or eights, and declines the split for all other ranks. doubleDown(self, hand ) Parameter hand (Hand) the hand which is being oered the opportunity to double down Returns true if this Player accepts the double oer for this Hand. The Player must also update the Bet associated with this Hand. This player tries to accept the oer when the hand points are 10 or 11, and the dealers up card is 7 to 10 or ace. Otherwise the oer is rejected.

270

Chapter 42. Simple Blackjack Player Class

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Note that some games will restrict the conditions for a double down oer. For example, some games only allow double down on the rst two cards. Other games may not allow double down on hands that are the result of a split. hit(self, hand ) Parameter hand (Hand) the hand which is being oered the opportunity to hit Returns true if this Player accepts the hit oer for this Hand. If the dealer up card is from 2 to 6, there are four choices for the player. When the hand is 11 or less, hit. When the hand is a hard 12 to 16, stand. When the hand is a soft 12 to soft 16 (hard 2 to hard 6), hit. When the hand is 17 or more, stand. If the dealer up card is from 7 to 10 or an ace, there are four choices for the player. When the hand is 11 or less, double down. When the hand is a hard 12 to 16, hit. When the hand is a soft 12 to soft 16 (hard 2 to hard 6), hit. When the hand is 17 or more, stand. Otherwise, if the point total is 9 or less, accept the hit oer.

42.3 Blackjack Player Deliverables

There are two deliverables for this exercise. The SimpleBlackjackPlayer class. A class which performs a unit test of the SimpleBlackjackPlayer class. The unit test can provide a variety of Hands and conrm which oers are accepted and rejected.

42.3. Blackjack Player Deliverables

271

Building Skills in Object-Oriented Design, Release 2.1.1-Python

272

Chapter 42. Simple Blackjack Player Class

CHAPTER

FORTYTHREE

VARIANT GAME RULES

There are a number of variations in the sequence of game play oered by dierent casinos. In addition to having variations on the players strategy, we also need to have variations on the casino rules.

43.1 Variant Game Overview

There are wide variations in the ways people conduct Blackjack games. Well list a few of the variations we have heard of. Additional Win Rule: Charlie. Informal games may allow a ve-card Charlie or six-card Charlie win. If the players hand stretches to ve (or six) cards, they are paid at 1:1. Additional Oer: Surrender. This variation allows the player to lose only half their bet. The oer is made after insurance and before splitting. Surrender against ace and 10 is a good strategy, if this oer is part of the game. No Resplit. This variation limits the player to a single split. In the rare event of another matching card, resplitting is not allowed. No Double After Split. This variation prevents the player from a double-down after a split. Ordinarily, splitting Aces is followed by double-downs because 4/13 cards will lead to blackjack. Dealer Hits Soft 17. This variation forces the dealer to hit soft 17 instead of standing. This tends in increase the house edge slightly. Blackjack Pays 6:5. This variation is often used in single-deck games. It limits the value of cardcounting, since the house edge is stacked more heavily against the player. The restrictions on splitting and doubling down are small changes to the oers made by a variation on BlackjackGame. This would require creating a subclass of BlackjackGame to implement the alternative rules. The variations in the dealers rules (hitting a soft 17) is also a small change best implemented by creating a subclass of BlackjackGame. Reducing the number of decks is an easy change to our application. Since our main application constructs the Shoe before constructing the Game, it can construct a single-deck Shoe. Payout Odds. Handling the variations in the payout odds is a bit more complex. 1. The Player creates each Hand, associated with a Bet. The Bet is associated with the simple Ante outcome. 2. At the end of the Game, the Hand does a comparison between itself and the Dealers Hand to determine the odds for the Ante outcome.

273

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Its a 1:1 Outcome if the player does not have blackjack. Its a 3:2 bet if the player does have blackjack. Allocating this responsiblity to the Hand was a bad design decision. We should have allocated responsibility to the Game. We will need to add a method to the Game which compares a players Hand with the dealers Hand, sets the Outcome correctly, and resolves the bet.

43.2 BlackjackGame Rework

BlackjackGame is a subclass of Game that manages the sequence of actions that dene the game of Blackjack. Note that a single cycle of play is one complete Blackjack game from the initial ante to the nal resolution of all bets. Shuing is implied before the rst game and performed as needed.

43.2.1 Methods
adjustOdds(self, hand ) Parameter hand (Hand) The hand which has the ante bet odds corrected This method is used at the end of the game to resolve the players Ante bet. If the player went bust, their hand was already resolved. If the dealer went bust, all remaining bets are resolved as winners. The remaining situation (both player and dealer have non-bust hands) requires this BlackjackGame.adjustOdds() method. This method will set the hands outcomes odds to 3:2 if the player holds Blackjack.

43.3 OneDeckGame Class

OneDeckGame is a subclass of BlackjackGame that manages the sequence of actions for a one-deck game of Blackjack with a 6:5 blackjack OutcomeAnteLowOdds. Typically, this is built with a one-deck instance of Shoe. The OneDeckGame.adjustOdds() method overrides the superclass. This method will set the hands outcomes odds to 6:5 if the player holds Blackjack.

43.4 Variant Game Deliverables

There are three deliverables for this exercise. The revised BlackjackGame class. All of the original unit tests should apply to the refactored function that sets the outcome odds. The OneDeckGame class. A class to perform a unit test of the OneDeckGame class.

274

Chapter 43. Variant Game Rules

CHAPTER

FORTYFOUR

CONCLUSION
The game of Blackjack has given us an opportunity to further extend and modify an application with a considerable number of classes and objects. These exercises gave us an opportunity to work from a base of software, extending and rening our design. We omitted concluding exercises which would integrate this package with the Simulator and collect statistics. This step, while necessary, doesnt include many interesting design decisions. The nal deliverable should be a working application that parses command-line parameters, creates the required objects, and creates an instance of Simulator to collect data. We have specically omitted delving into the glorious details of specic player strategies. We avoided these details for two reasons. Intellectual Property. We didnt create any strategies. Rather than get permission to quote existing blackjack strategies, we leave it to the interested student to either buy any of the available books on Blackjack or download a strategy description from the Internet. Irrelevant Complexity. We nd that all of the detailed programming required to implement a particular strategy doesnt have too much to do with object-oriented design. Rather, it is an exercise in construction of detailed if-statements. It appears to be the wrong focus for learning OO design. Next Steps. There are a number of logical next steps for the programmers looking to build skills in object-oriented design. Well split these along several broad fronts. Additional Technology. There are several technology directions that can be pursued for further design experience. This application is a command-line application, which makes it easy to build and test. Another area for building skills in design is the implementation of programs that make extensive use of a database. A graphical user interface (GUI) application will add the technology for graphical display, and responding to user input events. A web application will add the technology for HTTP web transactions, HTML, CSS, and potentially a database. Application Areas. We selected simulation because its part of the historical foundation for objectoriented programming. We selected casino games because they have a reasonable level of complexity. Clearly, numerous other application areas can be selected as the basis for problems. The number and variety of human endevors that can be automated is quite large. Moving beyond simulation or doing simulation on something more complex than a casino table game is a good next step.

275

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Additional Depth in Design Patterns. Its possible to use additional and dierent design patterns to extend and rene the application that you have built. Any book or web site on OO design patterns will provide numerous examples of patterns. These can be used for add exibility to these casino game simulators.

276

Chapter 44. Conclusion

Part V

Fit and Finish

277

Building Skills in Object-Oriented Design, Release 2.1.1-Python

A nished application includes more than just a working program. There are two additional questions. How do you know it works? That is, do you have any tests that demonstrate that it works correctly? We address this by creating unit tests. How is it designed? We address this by creating documentation within the programs source les. This documentation can be extracted to create a tidy, complete reference for all the classes and functions within our application.

279

Building Skills in Object-Oriented Design, Release 2.1.1-Python

280

CHAPTER

FORTYFIVE

PYTHON UNITTEST TESTING

Python unittest encourages us to build a test module separate from the application class were testing. The test module will have one more more TestCase classes, each of which has one or more test methods. Workow. One approach to unit testing is to build the tests rst, then write a class which at least doesnt crash, but may not pass all the tests. Once we have this in place, we can now debug the tests until everything looks right. This is called test-driven development. Wed start with unittest les in the test directory of our project. In this case, wed have testCard.py and testDeck.py. We also need skeleton les in our src directory: cards.py and deck.py. Generally, the process for creating a class with the unit tests has the following outline. 1. Write a skeleton for the target class. One that doesnt really do anything, but at least compiles and has the right method names. 2. Write the test class. This will create instances of the target class, exercise those instances, and make assertions about the state of those instances. 3. Run the test, knowing that will fail. Sometimes our target class is so simple that we get it right the rst time. Other times, the target class is more complex and there are parts we didnt nish when we wrote the skeleton version. 4. While tests are still failing. (a) Refactor. We nish our target class. Sometimes we simply ll in the missing parts. Other times, we have more serious design work to do. (b) Run the test suite. 5. At this point, the target class passes the suite of tests. However, it may still fail to meet other quality criteria. For example, it may have a convoluted structure, or it may be ineicient, or it may lack appropriate documentation. In any case, were not really done with development. While our target class fails to meet our quality standards. (a) Refactor. We correct the quality problems in our target class. (b) Run the test suite. If we have refactored properly, the tests still pass. If we have introduced a problem, tests will fail. Using unittest. A test runner will locate all the TestCase instances, execute them and summarize the pass/fail status of each individual test method. Generally, well organize our project directory to have a test directory and a src directory. Here is an example of running a test from the command line.

281

Building Skills in Object-Oriented Design, Release 2.1.1-Python

python test/testDeck.py .. ---------------------------------------------------------------------Ran 2 tests in 0.001s OK

45.1 Dependencies
Lets assume weve built two classes in some chapter; pretend that were building Card and Deck . One class denes a standard playing card and the other class deals individual card instances. We need unit tests for each class. Generally, unit tests are taken to mean that a class is tested in isolation. In our case, a unit test for Card is completely isolated because it has no dependencies. However, our Deck class depends on Card, leading us to make a choice. Either we have to create a Mock Card that can be used to test Deck in complete isolation, or our Deck test will depend on both Deck and Card. The choice depends on the relative complexity of Card, and whether or not Deck and Card will evolve independently. Some folks demand that all testing be done in complete isolation with a lot of mock classes. Other folks are less strict on this, recognizing that Deck and Card are very tightly coupled and Card is very simple. The Mock Card is almost as complex as Card.

45.2 Example
testCard.py
import unittest import card class TestCard( unittest.TestCase ): def setUp( self ): self.aceClubs= card.Card(1,card.Clubs) self.twoClubs= card.Card(2,card.Clubs) self.tenClubs= card.Card(10,card.Clubs) self.kingClubs= card.Card(13,card.Clubs) self.aceDiamonds= card.Card(1,card.Diamonds) def testString( self ): self.assertEquals( " AC", str(self.aceClubs) ) self.assertEquals( " 2C", str(self.twoClubs) ) self.assertEquals( "10C", str(self.tenClubs) ) def testOrder( self ): self.assertTrue( self.tenClubs < self.kingClubs ) self.assertFalse( self.tenClubs >= self.kingClubs ) self.assertTrue( self.kingClubs < self.aceClubs ) self.assertTrue( self.aceClubs == self.aceDiamonds ) if __name__ == "__main__": unittest.main()

282

Chapter 45. Python unittest Testing

Building Skills in Object-Oriented Design, Release 2.1.1-Python

1. Generally, we create a number of object instances in the setup method. In this case, we created ve distinct Card instances. These object constructors imply several things in our card module. (a) There will be a set of manifest constants for the suits: Clubs, Diamonds, Hearts and Spades. (b) The constructor (Card.__init__()) will accept a rank and a suit constant. Note that we didnt write tests to create all suits or all ranks. We can add these later. In some cases, where we are working in large teams, we may need to produce tests which exhaustively enumerate all possibe alternatives. For the purposes of learning OO design, we only need to sketch out our class by dening the tests it must pass. 2. In testString(), we exercise the __str__() method of the Card class to be sure that it formats cards correctly. These tests tell us what the formatting algorithm will look like. 3. In testOrder(), we exercise the __cmp__() method of the Card class to be sure that it compares card ranks correctly. Note that we have explicitly claimed that the equality test only checks the rank and ignores the suit; this is typical for Blackjack, but wont work well for Bridge or Solitaire. Note that we didnt exhaustively test all possible comparisons among the four cards we dened. We only need to execute the various paths within the __cmp__() method. When we initially create the test class, we may not have written the Card class yet. What we have to do is develop enough tests to get started, and add tests as we start writing our Card class. 4. This is the standard main program for unittest modules. Our initial Card class needs to have just enough of an API to allow the tests to run. Heres our skeleton Card class.

card.py
Diamonds= "D" Clubs= "C" Spades= "S" Hearts= "H" class Card( object ): def __init__( self, rank, suit ): pass def __str__( self ): return "" def __cmp__( self ): return 0

This class will minimally participate in the testing. It wont pass many tests, but it serves as a basis for developing the class implementation.

45.2. Example

283

Building Skills in Object-Oriented Design, Release 2.1.1-Python

284

Chapter 45. Python unittest Testing

CHAPTER

FORTYSIX

PYTHON DOCTEST TESTING

Python doctest module requires us to put our test cases and expected results into the docstring comments on a class, method or function. The test case information becomes a formal part of the API documentation. When a docstring includes doctest comments, the string serves dual duty as formal test and a working example. Workow. To use doctest is to build the class, exercise it in the Python interpreter, then put snippets of the interactive log into our docstrings. Generally, we follow this outline. 1. Write and debug the class, including docstring comments. 2. Exercise the class in an interactive Python interpreter. You can use IDLE or any other command line. 3. Copy the snippets out of the interactive log. Paste them into the docstring comments. 4. Run doctest to be sure that youve copied and pasted correctly. Using Doctest. There are two ways to use doctest. Start with Python 2.6, you can run doctest from the command line like this.
python -m doctest -v roulette.py

You can also add a test function to a module which runs doctest on the module. This test function should have a name which begins with _ to make it a name thats private to the module and not part fo the modules interface.
def _test(): import doctest doctest.testmod() if __name__ == "__main__": _test()

This _test() function is the main function of the module, so that when you run the module, it performs its internal doctests.
python roulette.py

Test-Driven Development. Note that the strict test-driven development (TDD) approach to unit testing is to build the tests rst, then write a class which at least doesnt crash, but may not pass all the tests. Once we have this in place, we can now debug the tests until everything looks right. This is called test-driven development, because the test cases come rst, driving the rest of the work.

285

Building Skills in Object-Oriented Design, Release 2.1.1-Python

The test-driven approach doesnt t well with doctest. Its very diicult to develop the test output without having the class available to exercise interactively. Also, we dont want to put comprehensive test scripts into our class docsting; they can easily get too long to be useful. Also, its diicult to fake the interactive log when designing the class. Its usually simpler to have a working class and then add test cases. Finally, we need to pick essential features for demonstration in the doctest comments. We cant easily write comprehensive tests that cover all features or all line of code in doctest comments. Limitations. Because of a few small gaps in Python 2s support for Unicode, its diicult to use doctest for testing the Card class. Specically, Python 2 source is assumed to be ASCII. The Unicode results of formatting a card cant easily be put into the Python source. Even if the explicit source encoding comment is used, the doctest module cant cope well with tests that involve Unicode. It can be challenging to do stand-alone unit tests with Mock classes to replace real classes. This can make for a rather long doctest string.

46.1 Example
Lets assume weve built two classes in some chapter; pretend that were building Card and Deck. One class denes a standard playing card and the other class deals individual card instances. Well dene some minimal doctests. The rst step is to develop our baseline class. Heres a module that seems like it works. This version uses ASCII suit names instead of Unicode suit names.

card.py - Initial
Clubs="C" Diamonds="D" Hearts="H" Spades="S" class Card( object ): """The card superclass, appropriate for number cards. .. todo:: Doctest goes here """ def __init__( self, rank, suit ): self.rank= rank self.suit= suit self.points= rank def hardValue( self ): return self.points def softValue( self ): return self.points def __eq__( self, other ): return self.rank == other.rank

286

Chapter 46. Python doctest Testing

Building Skills in Object-Oriented Design, Release 2.1.1-Python

def __lt__( self, other ): return self.rank < other.rank def __le__( self, other ): return self.rank <= other.rank def __gt__( self, other ): return self.rank > other.rank def __ge__( self, other ): return self.rank >= other.rank def __str__( self ): return self.unicode() def unicode (self ): return u"%d%s" % (self.rank, self.suit)

class AceCard( Card ): """The Ace class. .. todo:: Doctest goes here """ def __init__( self, suit ): self.rank= 14 self.suit= suit def hardValue( self ): return 1 def softValue( self ): return 11 def unicode (self ): return u"A%s" % (self.suit,)

Exercise the Class. Once we have the class, we need to exercise it using interactive Python. Heres what we saw.
MacBook-5:python slott$ python Python 2.6.3 (r263:75184, Oct 2 2009, 07:56:03) [GCC 4.0.1 (Apple Inc. build 5493)] on darwin Type "help", "copyright", "credits" or "license" for more information. >>> from card import * >>> c2d = Card( 2, Diamonds ) >>> str(c2d) '2D' >>> cas = AceCard( Spades ) >>> str(cas) 'AS' >>> cas.softValue() 11

During our session, small parts of the script show the preferred use cases for our class. We copy these from the interactive session and paste them into our class docstrings. Update the Docstrings. After we have some output that shows correct behavior of our class, we can put

46.1. Example

287

Building Skills in Object-Oriented Design, Release 2.1.1-Python

that output into the class docstrings. Heres our updated card.py module with doctest comments.

card.py - Revised
Clubs="C" Diamonds="D" Hearts="H" Spades="S" class Card( object ): """Create a new Ordinary (non-Ace, non-Face) Card. >>> c2d = Card( 2, Diamonds ) >>> str(c2d) '2D' """ etc. class AceCard( Card ): """The Ace class. >>> cas = AceCard( Spades ) >>> str(cas) 'AS' >>> cas.softValue() 11 """

46.2 Add the Test Framework

There are two doctest frameworks. Doctest can be used for stand-alone testing. It can also be incorporated into unittest testing. The general approach to stand-alone testing is to add the following to a module. When you run the module, it executes all doctest strings found throughout the module. Tests will pass silently. Any output is a failure of some kind.
def _test(): import doctest doctest.testmod() if __name__ == "__main__": _test()

46.3 Mixed unittest and doctest

A more complete approach is to use unittest for most testing and doctest to emphasize specic use cases in the docstrings. When doing this, youll have a test module that looks something like the following.

288

Chapter 46. Python doctest Testing

Building Skills in Object-Oriented Design, Release 2.1.1-Python

testCard.py
import unittest import doctest import cards suite = unittest.TestSuite() suite.addTest(doctest.DocTestSuite(cards)) runner = unittest.TextTestRunner() runner.run(suite)

46.3. Mixed unittest and doctest

289

Building Skills in Object-Oriented Design, Release 2.1.1-Python

290

Chapter 46. Python doctest Testing

CHAPTER

FORTYSEVEN

PYTHON DOCUMENTATION
Application Program Interface (API) documentation is absolutely essential. The easiest and most reliable way to produce this documentation is by using a tool that examines the source itself and develops the document directly from the programs. By using cleverly formatted Python docstrings, we can augment that analysis with easy-to-understand descriptions. In the case of Python, there are several tools for extracting documentation from the source. Two populare ones are Epydoc http://epydoc.sourceforge.net/. Sphinx http://sphinx.pocoo.org/. Workow. Generally, the workow has the following outline. 1. Develop the skeleton class. 2. Develop unit tests for most features. 3. Rework the class until it passes the unit tests. This may involve adding or modifying tests as we understand the class better. This also involves writing docstrings for modules, classes and methods. 4. Revisit the modules, classes and methods, nishing each description using the ReStructured Text (RST) markup language. We wont cover all of this language, just enough to provide a sense of how the tool is used to make clean, professional API documentation. 5. Run the documentation tool (epydoc or sphinx) to create the documentation. Fix any errors in your docstring markup. Also, rework the documentation as necessary to be sure that youve capture important information about the class, the methods, the design decisions. Using the Epydoc. When you install Epydoc, it creates a script that you can use. To run Epydoc, youll use a command similar to the following.
epydoc -v --docformat restructuredtext --output apidoc card.py

The -v option provides detailed debugging information for incorrectly formatted docstrings. The --docformat option species that the various docstrings use RST markup. The --output option denes the directory to which the documentation is written. The arguments (in this case, card.py) is a list of package directories or module les to be documented. Note that RST is not the default markup language for Epydoc. You must either specify this on the command line or in the module source. To identify the formatting in each module you must provide the format as the value of a __docformat__ module global variable.
#!/usr/bin/env python """

291

Building Skills in Object-Oriented Design, Release 2.1.1-Python

Some docstring for a module. """ docformat = "restructuredtext en"

Using Sphinx. When you install Sphinx, it creates a script that you can use. Generally, youll run the $ sphinx-quickstart script to create an empty documentation directory.
$ sphinx-quickstart

This interactive script will ask a number of questions, allowing you to pick sensible directories and le names. Youll need to include two extensions to enable Sphinx to create your API documentation. autodoc will create module documents based on the RST markup in the docstrings. autosummary will create summaries of packages and modules. Youll then add content to your index.rst le. This le is the root for the documentation. It can be very simple, or or it can do include other les to break a large document into manageable pieces. The most important commands are the following. To run Sphinx, youll use a command similar to the following.
$ sphinx-build -b html sourcedir builddir

ReStructured Text. Basic RST rules allow you to have a nice-looking document, with numbered and bulleted lists. You can include code examples and doctest results, also. With the inline markup, you can specify italic, bold or code samples. Additionally, RST oers eld markup to tie your documentation to specic pieces of Python syntax. This includes method parameters, return values, class instance variables, module variables, etc.

47.1 Basic RST Markup

We write our module, class and method docstrings using the RST markup language. RST is one of many structured text markup languages, where simple indentation and formatting rules are used to interpret your documetation. These formatting rules help the tools deduces the structure of your document (sections, list, etc.) and the presentation (bold, italic, or font changes.) These rules are generally pretty simple rules that plain text documents have been using since documents were rst stored on computers. Fundamentally, the RST formatting rules are implemented by a module named docutils. For additional documentation on RST, see the Sphinx web site (http://sphinx.pocoo.org) and the Docutils web site (http://docutils.sourceforge.net/). In addition to document structure, deduced from the layout, RST also uses explicit directives and interpreted text (or inline) markup. Document Structure. Generally, you must write in paragraphs. A blank line is the end of one paragraph and the beginning of another. Also, dierent indentation will signal paragaph changes. The indent rule is handy for making bullet lists or numbered lists. A line that is indented and begins with - is a bulleted list. A line that starts with digits and periods is a numbered list. Lists can nest within each other. If you need to have sections and subsections (youll almost always do this in the docstring for a module or package), you underline the title with lines of ===, --- or ~~~. 292 Chapter 47. Python Documentation

Building Skills in Object-Oriented Design, Release 2.1.1-Python

To include code code samples and doctest, youll use either the simple :: marker. Or the explicit .. code-block:: directive. Generally, the rst paragraph must be a pithy summary of the material to follow. This paragraph will often be used in overviews and index pages. Example testCard Module. For example, our module document might look like this.
#!/usr/bin/env python """The testCard module includes a number of unit tests for the Card and Deck classes in the card module. Overview ======== This module tests the following classes: - Card - Deck Card Tests ---------.. autoclass:: TestCards

Deck Tests ---------.. autoclass:: TestDeck

Usage ===== This module uses the standard text runner, so it can be executed from the command line as follows:: python testcard.py """

1. We started with an overview paragraph. It summarizes the module in a single sentence. 2. We have an underlined section heading. In this document, the highest level of the outline is underlined with =s. The second level is underlined with -. RST is very exible with the underlining rules. 3. We have a bullet list. Simply indent, and begin the paragraph with -. Lists can be nested, the RST processor will work this out based on the indentation of the various sublists. 4. We used the .. autoclass:: directive to gather docstring documentation from the class docstring and include it in the overall module docstring. 5. We included a literal block of code after a paragraph which ends with ::. The following material is indented to show what should be formatted specially. Interpreted Text, or Inline Markup. In addition to basic document structure rules, RST includes a way to show that a span of characters should be interpreted specially. This is sometimes called Inline Markup because it is within the structural markup. *words* will render the words in italic. This can also be done with :emphasis:`words`. **words** will render the words in bold. This can also be done with :strong:`words`.

47.1. Basic RST Markup

293

Building Skills in Object-Oriented Design, Release 2.1.1-Python

``words`` will render the words in a fixed font to indicate that it is a code sample. Literal programming text can also use :literal:`words`. Code samples can also be done with :samp:`word {replaceable}`. In this case, the material enclosed in {} is displayed in italics: word replaceable. Additionally, there are several cross-reference inline markup formats available for code elements. Heres an example of a paragraph using some inline markup to make some characters italic, other bold, and still others a xed-width nd that makes them looke like code.
This method computes the square root of a number, *n*. It returns a value, *r*, such that :samp:`r**2 == n`. **Note**. Square root of a negative number will raise a ``TypeError`` exception.

47.2 RST Field Markup

RST Field markup is primarily used to tie your documentation directly to Python language structures. Field markup is a separate paragraph of the form :tag arg: body. We wont cover all of the tags, just enough to produce basic documentation. Field markup also provides several standard kinds of standard formal paragraphs within your document. Well look at these separately. Functions and Method Functions. The following tags are used to dene detailed features of a method function. :param p: description A description of the parameter, p, for a function or method. :return: description A description of the return value from the function or method. :keyword k: description A description of the keyword parameter, k, for a function or method. Heres an example of a method function docstring that uses @param tags.
def __init__( self, rank, suit ): """Build a card with a given rank and suit. :param rank: numeric rank, ace is 1, 2-13 for 2 though King. :param suit: String suit, use one of the module constants. """ self.rank= 14 if rank == 1 else rank self.suit= suit

Classes and Modules. The following tags are used to dene specic variables in a module or class. :ivar v: description A description of the an instance variable of a class. Generally, this will be in the class level docstring, and will refer to one of the classs self. variables. :var v: description A global variable of a module. Generally, this will be in the module docstring, and will refer to one of the global variables created by the module. Heres an example of a class denition docstring that uses :ivar: tags.
class Card( object ): """A single playing card, suitable for Blackjack or Poker. While a suit is retained, it doesn't figure into the ordering of cards, as it would in Bridge.

294

Chapter 47. Python Documentation

Building Skills in Object-Oriented Design, Release 2.1.1-Python

**Reminder**. The internal rank of an Ace is 14. however, expects a value of 1. :ivar rank: the numeric rank of the card. :ivar suit: the string suit of the card. """

The constructor,

2-13, with ace as 14.

Directives. RST has a number of useful directives available for structuring the documentation. Sphinx uses all of the basic RST directives and adds a large number of additional directives. A directive is an RST markup command starts with a special .. line, and may include a number of lines indented within the directive. One of the common ones is the .. code-block:: directive. The :: is an essential part of the directive syntax. It looks like this.
Example. .. code-block:: python def fact(n): if n == 0: return 1 return n*fact(n-1) More Text.

The directive starts with the .. line and continues through the indented section. The directive content ends with the end of the indentation. Epydoc uses the :tag: syntax; it doesnt dene new RST directives. Because of this, there are two styles for some of the more advanced markup: an Epydoc tag style and a Sphinx directive style. Standard Paragraphs. There are several kinds of standard paragraphs that are part of any well-written document set. These include things like Related Topics. Epydoc usess the :seealso: tag to generate a Related Topics paragaph with references to other documents. Sphinx uses the .. seealso:: directive line to generate additional references. Admonitions (Notes, Warnings, etc). Epydoc uses tags like :note:, :attention:, :bug:, and :warning: tags to generate standard types of admonition paragraphs. Sphinx uses directives like .. note::, .. attention::, and .. warning:: directives to generate standard types of admonition paragraphs. Other adminitions include caution, danger, error, hint, important, tip. Status. You can track the development and deployment status of your programs. Tags like :version:, :todo:, :deprecated:, :since:, :status:, and change: are used by both Epydoc and Sphinx. Bibliographic Information. You can provide standard publication information. Tags like :author:, :organization:, :copyright:, :license: and :contact: can be used.

47.2. RST Field Markup

295

Building Skills in Object-Oriented Design, Release 2.1.1-Python

47.3 Class Example

Heres an example of a module with detailed class documentation.
#!/usr/bin/env python """The cards module defines the Card class and constants for the four suits. :author: S. Lott :license: http://creativecommons.org/licenses/by-nc-nd/3.0/us/ .. data:: Clubs Use this to build Cards in the clubs suit .. data:: Diamonds Use this to build Cards in the diamonds suit .. data:: Hearts Use this to build Cards in the hearts suit :.. data:: Spades: Use this to build Cards in the spades suit """ Clubs="C" Diamonds="D" Hearts="H" Spades="S" class Card( object ): """A single playing card, suitable for Blackjack or Poker. While a suit is retained, it doesn't figure into the ordering of cards, as it would in Bridge. .. note:: Aces and Facecards. Ace and Facecards are separate subclasses. attribute:: rank The numeric rank of the card. 2-13, rank of 14 when used in Poker.

ace has an effective

attribute:: suit The string suit of the card. This should be from the named constants (Clubs, Diamonds, Hearts, Spades).

""" def __init__( self, rank, suit ): """Build a card with a given rank and suit. :param rank: numeric rank, 2-10. Aces and FaceCards are separate. :type rank: integer in the range 2 to 10 inclusive. :param suit: String suit, use one of the module constants. """ self.rank= rank self.suit= suit self.points= rank def hardValue( self ):

296

Chapter 47. Python Documentation

Building Skills in Object-Oriented Design, Release 2.1.1-Python

"""For blackjack, the hard value of this card. :returns: int """ return self.points def softValue( self ): """For blackjack, the soft value of this card. :returns: int """ return self.points def __eq__( self, other ): """Compare cards, ignoring suit. >>> Card( 2, Diamonds ) == Card( 2, Spades ) True >>> Card( 2, Diamonds ) == Card( 10, Spades ) False """ return self.rank == other.rank def __lt__( self, other ): """Compare cards, ignoring suit. >>> Card( 2, Diamonds ) < Card( 3, Spades ) True >>> Card( 10, Diamonds ) < Card( 10, Spades ) False """ return self.rank < other.rank def __le__( self, other ): return self.rank <= other.rank def __gt__( self, other ): return self.rank > other.rank def __ge__( self, other ): return self.rank >= other.rank def __str__( self ): return "%d%s" % (self.rank, self.suit)

47.3. Class Example

297

Building Skills in Object-Oriented Design, Release 2.1.1-Python

298

Chapter 47. Python Documentation

Part VI

Back Matter

299

CHAPTER

FORTYEIGHT

BIBLIOGRAPHY
48.1 Use Cases 48.2 Computer Science 48.3 Design Patterns 48.4 Statistics 48.5 Python 48.6 Java 48.7 Casino Games

301

Building Skills in Object-Oriented Design, Release 2.1.1-Python

302

Chapter 48. Bibliography

CHAPTER

FORTYNINE

TOOLSET
The following toolset was used for production of this book. Python 2.6.3. Sphinx 0.63. Docutils 0.5. Komodo Edit 5.2.2. pyPDF 1.12. MacTeX-2008.

303

Building Skills in Object-Oriented Design, Release 2.1.1-Python

304

Chapter 49. Toolset

CHAPTER

FIFTY

INDICES AND TABLES

Index Search Page

305

Building Skills in Object-Oriented Design, Release 2.1.1-Python

306

Chapter 50. Indices and tables

BIBLIOGRAPHY

[Jacobson92] Ivar Jacobson, Magnus Christerson, Patrik Jonsson, Gunnar vergaard. Object-Oriented Software Engineering. A Use Case Driven Approach. 1992. Addison-Wesley. 0201544350. [Jacobson95] Ivar Jacobson, Maria Ericsson, Agenta Jacobson. The Object Advantage. Business Process Reengineering with Object Technology. 1995. Addison-Wesley. 0201422891. [Parnas72] Parnas D. On the Criteria to Be Used in Decomposing Systems into Modules. 1053-1058. 1972. Communications of the ACM. [Gamma95] Erich Gamma, Richard Helm, Ralph Johnson, John Vlissides. Design Patterns. Elements of Object-Oriented Software. 1995. Addison-Wesley Professional. 0201633612. [Larman98] Craig Larman. Applying UML and Patterns. An Introduction to Object-Oriented Analysis and Design. 1998. Prentice-Hall. 0137488807. [Neter73] John Neter, William Wasserman, G. A. Whitmore. Fundamental Statistics for Business and Economics. 4. 1973. Allyn and Bacon, Inc.. 020503853. [vanRossum04] Guido van Rossum, Fred L. Drake. Python Documentation. 2004. Python Labs. [Silberstang05] Edwin Silberstang. The Winners Guide to Casino Gambling. 4th. 2005. Owl Books. 0805077650. [Skiena01] Steven Skiena. Calculated Bets. Computers, Gambling, and Mathematical Modeling to Win. 2001. Cambridge University Press. 0521009626. [Shackleford04] Michael Shackleford. The Wizard Of Odds. 2004.

307

Intrusion Detection Honeypots
From Everand
Intrusion Detection Honeypots
Chris Sanders
3/5 (2)
Audio, Video, and Media in the Ministry
From Everand
Audio, Video, and Media in the Ministry
Clarence Floyd Richmond
No ratings yet
Introduction To Game Design Prototyping Development Jeremy Gibson PDF
No ratings yet
Introduction To Game Design Prototyping Development Jeremy Gibson PDF
104 pages
Building Skills in Object-Oriented Design 2.1.1 Java
No ratings yet
Building Skills in Object-Oriented Design 2.1.1 Java
311 pages
ChatGPT CheatSheet: 400 Powerful Examples That Turn You Into a ChatGPT Expert
From Everand
ChatGPT CheatSheet: 400 Powerful Examples That Turn You Into a ChatGPT Expert
Igor Pogany
No ratings yet
Gray Hat Hacking the Ethical Hacker's
From Everand
Gray Hat Hacking the Ethical Hacker's
Çağatay Şanlı
5/5 (1)
Anne of green gables
From Everand
Anne of green gables
L. M. Montgomery
No ratings yet
The Unstuck Book
From Everand
The Unstuck Book
Conor McCarthy
No ratings yet
The Development and Implementation of A Domino Game: Yuelong Huang
No ratings yet
The Development and Implementation of A Domino Game: Yuelong Huang
52 pages
Yang20 MSC
No ratings yet
Yang20 MSC
61 pages
ModelSim SE Tutorial Software Version 6.2b
0% (1)
ModelSim SE Tutorial Software Version 6.2b
180 pages
Plant Simulation Book - Robinson
100% (3)
Plant Simulation Book - Robinson
339 pages
Software Engineering Research For Computer Games: A Systematic Review
No ratings yet
Software Engineering Research For Computer Games: A Systematic Review
14 pages
Algorithms and Networking for Computer Games 1st Edition Jouni Smed - Experience the full ebook by downloading it now
100% (2)
Algorithms and Networking for Computer Games 1st Edition Jouni Smed - Experience the full ebook by downloading it now
47 pages
Shane Torbert - Applied Computer Science 2011
100% (2)
Shane Torbert - Applied Computer Science 2011
212 pages
Applied Computer Science
No ratings yet
Applied Computer Science
212 pages
Practical Game Development With Unity and Blender
100% (3)
Practical Game Development With Unity and Blender
353 pages
RoboCup@Work Rules Book
No ratings yet
RoboCup@Work Rules Book
65 pages
Modelsim Se Tut
No ratings yet
Modelsim Se Tut
205 pages
Onlinebookstore by Prabaht
No ratings yet
Onlinebookstore by Prabaht
28 pages
Modelsim User
No ratings yet
Modelsim User
685 pages
Modelsim User
No ratings yet
Modelsim User
687 pages
Modelsim User Guaide.
No ratings yet
Modelsim User Guaide.
703 pages
Top-Down Method at Work in Analog IC Design
No ratings yet
Top-Down Method at Work in Analog IC Design
180 pages
mbcmodel
No ratings yet
mbcmodel
428 pages
Simulation-Based Decision Support Tool For Early Stages of Zero-Energy Building Design
No ratings yet
Simulation-Based Decision Support Tool For Early Stages of Zero-Energy Building Design
14 pages
ExpressJS Reference
No ratings yet
ExpressJS Reference
246 pages
Jersey Documentation 1.0.3.1 User Guide
No ratings yet
Jersey Documentation 1.0.3.1 User Guide
34 pages
Maven Quick Reference Card
No ratings yet
Maven Quick Reference Card
8 pages
Picturing Programs An Introduction To Computer Programming
No ratings yet
Picturing Programs An Introduction To Computer Programming
480 pages