Design and Analysis of Algorithms

Dynamic Programming
Dr. Lê Nguyên Khôi
The VNU University of Engineering and Technology
Contents

 Bottom-up design technique

 Typical problems

Design and Analysis of Algorithms 1

Divide & Conquer – Re-visit

 Algorithm design technique

 Top-down design
 Divide the problem into sub-problems
 Solve sub-problems recursively
 Combine solution to sub-problems to construct the
solution to the original problem
 Example
 Merge sort
 Quick sort
 Compute Fibonacci number

Design and Analysis of Algorithms 2

Divide & Conquer – Fibonacci Numbers

0 1 1 2 3 5 8 13 21 …
Fibonacci
1 if return
2 if return
3 return Fibonacci
+ Fibonacci
Complexity: (exponential growth)
for – golden ratio

Design and Analysis of Algorithms 3

Divide & Conquer – Fibonacci Numbers

 Use binary tree representation to illustrate the

computational process
 What is the height of the tree
 How many leaves are there
 What is the bound for the number of leaves
 Number of additions
 Complexity
 Top-down design technique
 Solve similar sub-problems

Design and Analysis of Algorithms 4

Fibonacci Numbers – Other Technique

 Work backward from leaves of the left-most

sub-tree
 Compute the value (key) of the corresponding
parent (internal node)
 Work upward to the root of the tree
 Bottom-up design technique

Design and Analysis of Algorithms 5

Dynamic Programming

 Algorithm design technique

 Bottom-up design
 Solve sub-problems from the smallest size ( ) to
the biggest size (the original problem)
 Construct solutions to the sub-problem based on
solutions of the smaller size sub-problems
 Example
 Compute Fibonacci number
 Insertion sort

Design and Analysis of Algorithms 6

Dynamic Programming – Usage

Properties of the problem:

 Overlapping sub-problems
 A recursive solution contains a “small” number of
distinct sub-problems repeated many times
 Sub-problems share sub-subproblems
 Optimal structure
 An optimal solution to a problem contains optimal
solutions to sub-problems
 The optimal solutions to the sub-problem could be
used to construct the optimal solutions to the original
problem

Design and Analysis of Algorithms 7

Dynamic Programming – Usage

 Apply to optimization problems

 Have many possible solutions
 Each solution has a objective value
 Find a solution with the optimal value
 There could be more than 1 optimal solution

Design and Analysis of Algorithms 8

Design Steps
1. Characterize the structure of a solution
2. The method to combine solutions to sub-problems to
construct solutions to the original problem
3. Design tables to store solutions to sub-problems for
latter use
4. Compute solutions to sub-problems (from small sub-
problems to bigger sub-problems)
5. Determine the solution to the original problem based
on solutions in the table

Dynamic programming is a tabular method

Design and Analysis of Algorithms 9

Divide & Conquer – Dynamic Programming

 Determine the correct design technique

 Apply the divide & conquer technique for
problems with overlapping subproblems
 Re-compute solutions to sub-problems
 Inefficient
 Example: compute Fibonacci number

Design and Analysis of Algorithms 10

Problems

 Compute Fibonacci number

 Binomial coefficients
 Cutting rod
 Longest increasing subsequence
 Knapsack
 Longest common subsequence

Design and Analysis of Algorithms 11

Compute Fibonacci number

 Divide & conquer

 Overlapping sub-problems
 Dynamic programming
 Table to store values of

Design and Analysis of Algorithms 12

Binomial Coefficients

 Divide & conquer

 Overlapping sub-problems
 Dynamic programming
 Table to store values of

Design and Analysis of Algorithms 13

Cutting Rod

 Read section 15.1 p.360

Design and Analysis of Algorithms 14

Longest Increasing Subsequence
 S = { 14, 1, 17, 2, 16, 17, 3, 15, 4, 1, 5,
18, 13, 6, 7, 19, 8, 12, 1, 9, 10, 8 }
 Increasing subsequence
 S’ = { 14, 1, 17, 2, 16, 17, 3, 15, 4, 1, 5,
18, 13, 6, 7, 19, 8, 12, 1, 9, 10, 8 }
 Longest increasing subsequence
 S = { 14, 1, 17, 2, 16, 17, 3, 15, 4, 1, 5,
18, 13, 6, 7, 19, 8, 12, 1, 9, 10, 8 }

Design and Analysis of Algorithms 15

Longest Increasing Subsequence

 Bài toán con: Tìm dãy con tăng dài nhất

kết thúc tại phần tử thứ k
 Thuật toán
 Nếu tất cả các phần tử trước k đều lớn hơn
hoặc bằng S[k], trả về dãy con chỉ chứa S[k]
 Nếu có t phần tử đứng trước k nhỏ hơn S[k],
gọi W là dãy dài nhất trong các dãy con tăng
kết thúc tại các phần tử này. Trả về W S[k]

Design and Analysis of Algorithms 16

Longest Increasing Subsequence
s1 14 {14}
S = { 14, 1, 17, 2, 16, s2 1 {1}
s3 17 {14|1, 17}
17, 3, 15, 4, 1, 5, 18, s4 2 {1, 2}
s5 16
13, 6, 7, 19, 8, 12, 1, s6 17
9, 10, 8 } s7
s8
3
15
s9 4
s10 1
s11 5
Use all previously s12 18
constructed Si to s13
s14
13
6
construct Sj (j>i) s15
s16
7
19
s17 8
s18 12
s19 1
s20 9
s21 10
s22 8
Design and Analysis of Algorithms 17
Longest Increasing Subsequence
s1 14 {14}
S = { 14, 1, 17, 2, 16, s2 1 {1}
s3 17 {14|1, 17}
17, 3, 15, 4, 1, 5, 18, s4 2 {1, 2}
s5 16 {1, 2, 16}
13, 6, 7, 19, 8, 12, 1, s6 17 {1, 2, 16, 17}
9, 10, 8 } s7
s8
3
15
{1, 2, 3}
{1, 2, 3, 15}
s9 4 {1, 2, 3, 4}
s10 1 {1}
s11 5 {1, 2, 3, 4, 5}
Use all previously s12 18 {1, 2, 3, 4, 5, 18}
constructed Si to s13
s14
13
6
{1, 2, 3, 4, 5, 13}
{1, 2, 3, 4, 5, 6}
construct Sj (j>i) s15
s16
7
19
{1, 2, 3, 4, 5, 6, 7}
{1, 2, 3, 4, 5, 6, 7, 19}
s17 8 {1, 2, 3, 4, 5, 6, 7, 8}
s18 12 {1, 2, 3, 4, 5, 6, 7, 8, 12}
s19 1 {1}
s20 9 {1, 2, 3, 4, 5, 6, 7, 8, 9}
s21 10 {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}
s22 8 {1, 2, 3, 4, 5, 6, 7, 8}
Design and Analysis of Algorithms 18
Knapsack
There are N items, item i has weight wi and profit ti
A knapsack has the weight limit of M kg
Find a subset of items such that their total weight
does not exceed the weight limit and the profit of
the knapsack is maximized

Example
N = 5, M = 10
i A B C D E
wi 1 3 5 7 9
ti $100 $200 $301 $400 $500

Design and Analysis of Algorithms 19

 Knapsack
for all cell[x, y]
if x = 0 & y = 0 then
cell[x, y] = 0kg $0 {}
Item Weight Profit
else
A 1kg $100
B 3kg $200 Let m be the additional item at column y
C 5kg $301 Let w be the weight of item m
D 7kg $400 if w > maximum weight of the row
E 9kg $500 cell[x, y] = cell[x, y-1]
else
cell[x, y] = max {cell[x, y-1], (cell[x-w, y-1] U {m})}

empty A A/B A/B/C A/B/C/D A/B/C/D/E

kg ≤ 0 0kg $0 {} 0kg $0 {} 0kg $0 {} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 1 0kg $0 {} 1kg $100 {A} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 2 0kg $0 {} 1kg $100 {A} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 3 0kg $0 {} 1kg $100 {A} 3kg $200 {B} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 4 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 5 0kg $0 {} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 6 0kg $0 {} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 7 0kg $0 {} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 8 0kg $0 {} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 9 0kg $0 {} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {} ?kg $? {}
kg ≤ 10 0kg $0 {} 1kg $100 {A} ?kg $? {} ?kg $? {} ?kg $? {} ?kg $? {}

Design and Analysis of Algorithms 20

 Knapsack
for all cell[x, y]
if x = 0 & y = 0 then
cell[x, y] = 0kg $0 {}
Item Weight Profit
else
A 1kg $100
B 3kg $200 Let m be the additional item at column y
C 5kg $301 Let w be the weight of item m
D 7kg $400 if w > maximum weight of the row
E 9kg $500 cell[x, y] = cell[x, y-1]
else
cell[x, y] = max {cell[x, y-1], (cell[x-w, y-1] U {m})}

empty A A/B A/B/C A/B/C/D A/B/C/D/E

kg ≤ 0 0kg $0 {} 0kg $0 {} 0kg $0 {} 0kg $0 {} 0kg $0 {} 0kg $0 {}
kg ≤ 1 0kg $0 {} 1kg $100 {A} 1kg $100 {A} 1kg $100 {A} 1kg $100 {A} 1kg $100 {A}
kg ≤ 2 0kg $0 {} 1kg $100 {A} 1kg $100 {A} 1kg $100 {A} 1kg $100 {A} 1kg $100 {A}
kg ≤ 3 0kg $0 {} 1kg $100 {A} 3kg $200 {B} 3kg $200 {B} 3kg $200 {B} 3kg $200 {B}
kg ≤ 4 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} 4kg $300 {A, B} 4kg $300 {A, B} 4kg $300 {A, B}
kg ≤ 5 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} 5kg $301 {C} 5kg $301 {C} 5kg $301 {C}
kg ≤ 6 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} 6kg $401 {A,C} 6kg $401 {A,C} 6kg $401 {A,C}
kg ≤ 7 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} 6kg $401 {A,C} 6kg $401 {A,C} 6kg $401 {A,C}
kg ≤ 8 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} 8kg $501 {B,C} 8kg $501 {B,C} 8kg $501 {B,C}
kg ≤ 9 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} 9kg $601 {A, B, C} 9kg $601 {A, B, C} 9kg $601 {A, B, C}
kg ≤ 10 0kg $0 {} 1kg $100 {A} 4kg $300 {A, B} 9kg $601 {A, B, C} 9kg $601 {A, B, C} 9kg $601 {A, B, C}

Design and Analysis of Algorithms 21

Longest Common Subsequence

Given 2 sequence and

Find the longest common subsequence

: A B C B D A B
: B D C A B A

Common subsequence: AB / BCDB / BCBA

LCS( , ) = BCBA

Design and Analysis of Algorithms 22

Longest Common Subsequence

 Check if all subsequences of is

the subsequence of
 Analysis
 Check subsequence for each
subsequence
 There are subsequence of .
 Complexity

Design and Analysis of Algorithms 23