Python and Algorithms

Mari Wahl, [email protected]

University of New York at Stony Brook
May 24, 2013
 Theres nothing to fear but the fear itself.
Thats called recursion, and that would lead you
to infinite fear.
 Hello, human! Welcome to my book on Python and algorithms! If you
are reading this you probably agree with me that those two can be a
lot of fun together (or you might be lost, and in this case I
suggest you give it a try anyway!). Also, many of the examples
shown here are available in my git repository, together with several
other (more advanced) examples for abstract data structures, trees,
graphs, and solutions for the Euler Project and the Topcoder
website. Dont forget to check them out!

This text was written purely for fun (I know, I know, this is a
broad definition of the word fun...) with no pretensions for
anything big, so please forgive me (or better, let me know) if you
find any typo or mistake. I am not a computer scientist by
formation (I am actually an almost-I-swear-it-is-close-Ph.D. in
Physics) so this maybe makes things a little less usual (or risky?).
I hope you have fun!

Mari, Stony Brook, NY

Summer/2013
4
Contents

I Flying with Python 9

1 Numbers 11
1.1 Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Floats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 The fractions Module . . . . . . . . . . . . . . . . . . . . . 14
1.5 The decimal Module . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Other Representations . . . . . . . . . . . . . . . . . . . . . . 15
1.7 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 16

2 Built-in Sequence Types 25

2.1 Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Tuples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4 Bytes and Byte Arrays . . . . . . . . . . . . . . . . . . . . . . 43

3 Collection Data Structures 45

3.1 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Dictionaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3 Pythons collection Data Types . . . . . . . . . . . . . . . 54
3.4 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 58

4 Pythons Structure and Modules 63

4.1 Modules in Python . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Control Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.3 File Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Multiprocessing and Threading . . . . . . . . . . . . . . . . . 79
4.5 Error Handling in Python . . . . . . . . . . . . . . . . . . . . 81
4.6 Debugging and Profiling . . . . . . . . . . . . . . . . . . . . . 83

5
6 CONTENTS

4.7 Unit Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

5 Object-Oriented Design 89
5.1 Classes and Objects . . . . . . . . . . . . . . . . . . . . . . . 90
5.2 Principles of OOP . . . . . . . . . . . . . . . . . . . . . . . . 91
5.3 Python Design Patterns . . . . . . . . . . . . . . . . . . . . . 94
5.4 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 96

II Algorithms are Fun 99

6 Additional Abstract Data Structures 101

6.1 Stacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2 Queues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.3 Deques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.4 Priority Queues and Heaps . . . . . . . . . . . . . . . . . . . 110
6.5 Linked Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.6 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 120

7 Asymptotic Analysis 133

7.1 Complexity Classes . . . . . . . . . . . . . . . . . . . . . . . . 133
7.2 Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.3 Runtime in Functions . . . . . . . . . . . . . . . . . . . . . . 136

8 Sorting 139
8.1 Quadratic Sort . . . . . . . . . . . . . . . . . . . . . . . . . . 139
8.2 Linear Sort . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.3 Loglinear Sort . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
8.4 Comparison Between Sorting Methods . . . . . . . . . . . . . 148
8.5 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 149

9 Searching 153
9.1 Sequential Search . . . . . . . . . . . . . . . . . . . . . . . . . 153
9.2 Binary Search . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
9.3 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 156

10 Dynamic Programming 163

10.1 Memoization . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
10.2 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 165
CONTENTS 7

III Climbing Graphs and Trees 169

11 Introduction to Graphs 171

11.1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 171
11.2 The Neighborhood Function . . . . . . . . . . . . . . . . . . . 173
11.3 Introduction to Trees . . . . . . . . . . . . . . . . . . . . . . . 176

12 Binary Trees 179

12.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . 179
12.2 Representing Binary Trees . . . . . . . . . . . . . . . . . . . . 179
12.3 Binary Search Trees . . . . . . . . . . . . . . . . . . . . . . . 183
12.4 Self-Balancing BST . . . . . . . . . . . . . . . . . . . . . . . . 186
12.5 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 193

13 Traversals and Problems on Graphs and Trees 207

13.1 Depth-First Search . . . . . . . . . . . . . . . . . . . . . . . . 207
13.2 Breadth-First Search . . . . . . . . . . . . . . . . . . . . . . 208
13.3 Representing Tree Traversals . . . . . . . . . . . . . . . . . . 209
13.4 Additional Exercises . . . . . . . . . . . . . . . . . . . . . . . 211
8 CONTENTS
 Part I

Flying with Python

9
Chapter 1

Numbers

When you learn a new language, the first thing you usually do (after our
dear hello world) is to play with some arithmetic operations. Numbers
can be integers, float point number, or complex. They are usually given
decimal representation but can be represented in any bases such as binary,
hexadecimal, octahedral. In this section we will learn how Python deals
with numbers.

1.1 Integers
Python represents integers (positive and negative whole numbers) using the
int (immutable) type. For immutable objects, there is no difference between
a variable and an object difference.
The size of Pythons integers is limited only by the machine memory, not
by a fixed number of bytes (the range depends on the C or Java compiler
that Python was built with). Usually plain integers are at least 32-bit long
(4 bytes)1 .To see how many bytes a integer needs to be represented, starting
in Python 3.1, the int.bit length() method is available:
>>> (999).bit_length()
10

To cast a string to an integer in some base or to change a the base of an

integer, we simply use int(s, base):
>>> s = 11
1
To have an idea of how much this means, think that, 1K of disk memory has 1024
8 bits = 210 bytes.

11
12 CHAPTER 1. NUMBERS

>>> d = int(s)
>>> print(d)
11
>>> b = int(s, 2)
>>> print(b)
3

It will raise a ValueError2 on failure (for example, if we had s=12

and tried to find the binary representation). The optional base argument
must be an integer between 2 and 36 inclusive.

1.2 Floats
Numbers with a fractional part are represented by the immutable type
float. In the case of single precision, a 32-bit float is represented by 1
bit for sign (negative being 1, positive being 0) + 23 bits for the significant
digits (or mantissa) + 8 bits for the exponent. In case of a double precision,
the mantissa will have 53 bits instead. Also, the exponent is usually rep-
resented using the biased notation, where you add the number 127 to the
original value3 .

Comparing Floats
We should never compare floats for equality nor subtract them. The reason
for this is that floats are represented in binary fractions and there are many
numbers that are exact in a decimal base but not exact in a binary base (for
example, the decimal 0.1). Equality tests should instead be done in terms
of some predefined precision. For example, we can use the same approach
that Pythons unittest module has with assert AlmostEqual:
>>> def a(x , y, places=7):
... return round(abs(x-y), places) == 0

Float numbers can also be compared by their bit patterns in memory.

First we need to handle sign comparison separately: if both numbers are
negative, we may compare them by flipping their signs, returning the oppo-
site answer. Patterns with the same exponent are compared according to
their mantissa.
2
We will learn about exceptions and errors in Python in following chapters.
3
Try to figure out why!
1.3. COMPLEX NUMBERS 13

Methods for Floats and Integers

In Python, the division operator / always returns a float. A floor division

(truncation) is made with the operator //. A module (remainder) operation
is given by the operator %. In addition, the method divmod(x,y) returns
both the quotient and remainder when dividing x by y:
>>> divmod(45,6)
(7, 3)

The method round(x, n) returns x rounded to n integral digits if n is a

negative int or returns x rounded to n decimal places if n is a positive int.
The returned value has the same type as x:
>>> round(100.96,-2)
100.0
>>> round(100.96,2)
100.96

The method as integer ratio() gives the integer fractional represen-

tation of a float:
>>> 2.75.as_integer_ratio()
(11, 4)

1.3 Complex Numbers

The complex data type is an immutable type that holds a pair of floats:
z = 3 + 4j, with methods such as: z.real, z.imag, and z.conjugate().
Complex numbers are imported from the cmath module, which provides
complex number versions of most of the trigonometric and logarithmic func-
tions that are in the math module, plus some complex number-specific func-
tions such: cmath.phase(), cmath.polar(), cmath.rect(), cmath.pi, and
cmath.e.
14 CHAPTER 1. NUMBERS

1.4 The fractions Module

Python has the fraction module to deal with parts of a fraction. For
instance, the following snippet shows the basics methods of this module:4
[general_problems/numbers/testing_floats.py]

from fractions import Fraction

def rounding_floats(number1, places):

some operations with float()
return round(number1, places)

def float_to_fractions(number):
return Fraction(*number.as_integer_ratio())

def get_denominator(number1, number2):

a = Fraction(number1, number2)
return a.denominator

def get_numerator(number1, number2):

a = Fraction(number1, number2)
return a.numerator

def test_testing_floats(module_name=this module):

number1 = 1.25
number2 = 1
number3 = -1
number4 = 5/4
number6 = 6
assert(rounding_floats(number1, number2) == 1.2)
assert(rounding_floats(number1*10, number3) == 10)
assert(float_to_fractions(number1) == number4)
assert(get_denominator(number2, number6) == number6)
assert(get_numerator(number2, number6) == number2)

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
4
All the codes shown in this book show a directory structure of where you can find it
in my git repository. Also notice that, when you write your own codes, that the PEP 8
(Python Enhancement Proposal) guidelines recommend four spaces per level of indenta-
tion, and only spaces (no tabs). This is not explicit here because of the way Latex format
the text.
1.5. THE DECIMAL MODULE 15

test_testing_floats()

1.5 The decimal Module

When we need exact decimal floating-point numbers, Python has an addi-
tional immutable float type, the decimal.Decimal. This method can take
any integer or even a string as argument (and starting from Python 3.1,
also floats, with the decimal.Decimal.from float() function). This an
efficient alternative when we do not want to deal with the rounding, equal-
ity, and subtraction problems that floats have:
>>> sum (0.1 for i in range(10)) == 1.0
False
>>> from decimal import Decimal
>>> sum (Decimal ("0.1") for i in range(10)) == Decimal("1.0")
True

While The math and cmath modules are not suitable for the decimal
module, its built-in functions such as decimal.Decimal.exp(x) are enough
to most of the problems.

1.6 Other Representations

The bin(i) method returns the binary representation of int i as a string:
>>> bin(999)
0b1111100111

The hex(i) method returns the hexadecimal representation of i as a

string:
>>> hex(999)
0x3e7

The oct(i) method returns the octal representation of i as a string:

>>> oct(999)
0o1747
16 CHAPTER 1. NUMBERS

1.7 Additional Exercises

Functions to Convert Between Different Bases
We can write our own functions to change bases in numbers. For example,
the snippet bellow converts a number in any base smaller than 10 to the
decimal base:
[general_problems/numbers/convert_to_decimal.py]

def convert_to_decimal(number, base):

multiplier, result = 1, 0
while number > 0:
result += number%10*multiplier
multiplier *= base
number = number//10
return result

def test_convert_to_decimal():
number, base = 1001, 2
assert(convert_to_decimal(number, base) == 9)
print(Tests passed!)

if __name__ == __main__:
test_convert_to_decimal()

By swapping all the occurrences of 10 with any other base in our previous
method we can create a function that converts from a decimal number to
another number (2 base 10):
[general_problems/numbers/convert_from_decimal.py]

def convert_from_decimal(number, base):

multiplier, result = 1, 0
while number > 0:
result += number%base*multiplier
multiplier *= 10
number = number//base
return result

def test_convert_from_decimal():
number, base = 9, 2
assert(convert_from_decimal(number, base) == 1001)
print(Tests passed!)
1.7. ADDITIONAL EXERCISES 17

if __name__ == __main__:
test_convert_from_decimal()

If the base is above 10 then we must use non-numeric characters to

represent these digits. We can let A stand for 10, B stand for 11 and so
on. The following code will convert a number from a decimal base to any
other base (up to 20):
[general_problems/numbers/convert_from_decimal_larger_bases.py]

def convert_from_decimal_larger_bases(number, base):

strings = "0123456789ABCDEFGHIJ"
result = ""
while number > 0:
digit = number%base
result = strings[digit] + result
number = number//base
return result

def test_convert_from_decimal_larger_bases():
number, base = 31, 16
assert(convert_from_decimal_larger_bases(number, base) == 1F)
print(Tests passed!)

if __name__ == __main__:
test_convert_from_decimal_larger_bases()

Finally, a very general base-conversion module, using a recursive method:

[general_problems/numbers/convert_dec_to_any_base_rec.py]

def convert_dec_to_any_base_rec(number, base):

convert an integer to a string in any base
convertString = 012345679ABCDEF
if number < base: return convertString[number]
else:
return convert_dec_to_any_base_rec(number//base, base) +
convertString[number%base]

def test_convert_dec_to_any_base_rec(module_name=this module):

number = 9
base = 2
assert(convert_dec_to_any_base_rec(number, base) == 1001)
18 CHAPTER 1. NUMBERS

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_convert_dec_to_any_base_rec()

Greatest Common Divisor

The following module calculates the greatest common divisor (gcd) between
two given integers:
[general_problems/numbers/finding_gcd.py]

def finding_gcd(a, b):

implements the greatest common divider algorithm
while(b != 0):
result = b
a, b = b, a % b
return result

def test_finding_gcd():
number1 = 21
number2 = 12
assert(finding_gcd(number1, number2) == 3)
print(Tests passed!)

if __name__ == __main__:
test_finding_gcd()

The Random Module

The follow snippet runs some tests on the Pythons random module:
[general_problems/numbers/testing_random.py]

import random

def testing_random():
testing the module random
values = [1, 2, 3, 4]
1.7. ADDITIONAL EXERCISES 19

print(random.choice(values))
print(random.choice(values))
print(random.choice(values))
print(random.sample(values, 2))
print(random.sample(values, 3))

shuffle in place
random.shuffle(values)
print(values)

create random integers

print(random.randint(0,10))
print(random.randint(0,10))

if __name__ == __main__:
testing_random()

Fibonacci Sequences
The module bellow shows how to find the nth number in a Fibonacci sequence
in three ways: (a) with a recursive O(2n ) runtime; (b) with a iterative O(n2 )
runtime; and (c) using a formula that gives a O(1) runtime but is not precise
after around the 70th element:
[general_problems/numbers/find_fibonacci_seq.py]

import math

def find_fibonacci_seq_rec(n):
if n < 2: return n
return find_fibonacci_seq_rec(n - 1) + find_fibonacci_seq_rec(n
- 2)

def find_fibonacci_seq_iter(n):
if n < 2: return n
a, b = 0, 1
for i in range(n):
a, b = b, a + b
return a

def find_fibonacci_seq_form(n):
20 CHAPTER 1. NUMBERS

sq5 = math.sqrt(5)
phi = (1 + sq5) / 2
return int(math.floor(phi ** n / sq5))

def test_find_fib():
n = 10
assert(find_fibonacci_seq_rec(n) == 55)
assert(find_fibonacci_seq_iter(n) == 55)
assert(find_fibonacci_seq_form(n) == 55)
print(Tests passed!)

if __name__ == __main__:
test_find_fib()

Primes
The following program finds whether a number is a prime in three ways:
(a) brute force; (b) rejecting all the candidates up to the square root of the
number; and (c) using the Fermats theorem with probabilistic tests:
[general_problems/numbers/finding_if_prime.py]

import math
import random

def finding_prime(number):
num = abs(number)
if num < 4 : return True
for x in range(2, num):
if num % x == 0:
return False
return True

def finding_prime_sqrt(number):
num = abs(number)
if num < 4 : return True
for x in range(2, int(math.sqrt(num)) + 1):
if number % x == 0:
return False
return True
1.7. ADDITIONAL EXERCISES 21

def finding_prime_fermat(number):
if number <= 102:
for a in range(2, number):
if pow(a, number- 1, number) != 1:
return False
return True
else:
for i in range(100):
a = random.randint(2, number - 1)
if pow(a, number - 1, number) != 1:
return False
return True

def test_finding_prime():
number1 = 17
number2 = 20
assert(finding_prime(number1) == True)
assert(finding_prime(number2) == False)
assert(finding_prime_sqrt(number1) == True)
assert(finding_prime_sqrt(number2) == False)
assert(finding_prime_fermat(number1) == True)
assert(finding_prime_fermat(number2) == False)
print(Tests passed!)

if __name__ == __main__:
test_finding_prime()

The following program uses Pythons random module to generate n-bit

prime numbers:
[general_problems/numbers/generate_prime.py]

import math
import random
import sys
from finding_prime import finding_prime_sqrt

def generate_prime(number=3):
while 1:
p = random.randint(pow(2, number-2), pow(2, number-1)-1)
p = 2 * p + 1
if finding_prime_sqrt(p):
return p
22 CHAPTER 1. NUMBERS

if __name__ == __main__:
if len(sys.argv) < 2:
print ("Usage: generate_prime.py number")
sys.exit()
else:
number = int(sys.argv[1])
print(generate_prime(number))

The NumPy Module

The NumPy module provides array sequences that can store numbers or
characters in a space-efficient way. Arrays in NumPy can have any ar-
bitrary dimension. They can be generated from a list or a tuple with the
array-method, which transforms sequences of sequences into two dimensional
arrays:
>>> x = np.array( ((11,12,13), (21,22,23), (31,32,33)) )
>>> print x
[[11 12 13]
[21 22 23]
[31 32 33]]

The attribute ndim tells us the number of dimensions of an array:

>>> x = np.array( ((11,12,13), (21,22,23) ))
>>> x.ndim
2

Further examples are shown below:

[general_problems/numbers/testing_numpy.py]

import numpy as np

def testing_numpy():
tests many features of numpy
ax = np.array([1,2,3])
ay = np.array([3,4,5])
print(ax)
print(ax*2)
print(ax+10)
print(np.sqrt(ax))
1.7. ADDITIONAL EXERCISES 23

print(np.cos(ax))
print(ax-ay)
print(np.where(ax<2, ax, 10))

m = np.matrix([ax, ay, ax])

print(m)
print(m.T)

grid1 = np.zeros(shape=(10,10), dtype=float)

grid2 = np.ones(shape=(10,10), dtype=float)
print(grid1)
print(grid2)
print(grid1[1]+10)
print(grid2[:,2]*2)

if __name__ == __main__:
testing_numpy()

NumPy arrays are also much more efficient than Pythons lists, as we
can see in the benchmark tests below:
[general_problems/numbers/testing_numpy_speed.py]

import numpy
import time

def trad_version():
t1 = time.time()
X = range(10000000)
Y = range(10000000)
Z = []
for i in range(len(X)):
Z.append(X[i] + Y[i])
return time.time() - t1

def numpy_version():
t1 = time.time()
X = numpy.arange(10000000)
Y = numpy.arange(10000000)
Z = X + Y
return time.time() - t1

if __name__ == __main__:
print(trad_version())
print(numpy_version())
24 CHAPTER 1. NUMBERS

Results:
3.23564291
0.0714290142059

Chapter 2

Built-in Sequence Types

The next step in our studies is learning how Python represents sequence
data types. A sequence type has the following properties:
? membership operator (for example, using in);
? a size method (given by len(seq));
? slicing properties (for example, seq[:-1]); and
? iterability (we can iterate the data in loops).
Python has five built-in sequence types: strings, tuples, lists, byte
arrays, and bytes:1
>>> l = []
>>> type(l)
<type list>
>>> s =
>>> type(s)
<type str>
>>> t = ()
>>> type(t)
<type tuple>
>>> ba = bytearray(b)
>>> type(ba)
<type bytearray>
>>> b = bytes([])
>>> type(byte)
<type type>
1
A named tuple is also available in the standard library, under the collections
package.

25
26 CHAPTER 2. BUILT-IN SEQUENCE TYPES

The Slicing Operator

In sequences, the slicing operator has the following syntax:
seq[start]
seq[start:end]
seq[start:end:step]

The index can be negative, to start counting from the right:

>>> word = "Let us kill some vampires!"
>>> word[-1]
!
>>> word[-2]
s
>>> word[-2:]
s!
>>> word[:-2]
Let us kill some vampire
>>> word[-0]
L

Mutability
Another propriety that any data type holds is mutability. Numbers are
obviously immutable; however, when it comes to sequence types, we can have
mutable types too. For instance, tuple, strings, and bytes are immutable,
while lists and byte arrays are mutable. Immutable types are more efficient
than mutable and some collection data types2 can only work with immutable
data types.
Since any variable is an object reference in Python, copying mutable
objects can be tricky. When you say a = b you are actually pointing a to
where b points. Therefore, to make a deep copy in Python you need to use
special procedures:
To make a copy of a list:
>>> newList = myList[:]
>>> newList2 = list(myList2)

To make a copy of a set (we will see in the next chapter), use:
>>> people = {"Buffy", "Angel", "Giles"}
2
Collection data types are the subject in the next chapter, and it includes, for example,
sets and dictionaries.
2.1. STRINGS 27

>>> slayers = people.copy()

>>> slayers.discard("Giles")
>>> slayers.remove("Angel")
>>> slayers
{Buffy}
>>> people
{Giles, Buffy, Angel}

To make a copy of a dict (also in the next chapter), use the

following:
>>> newDict = myDict.copy()

To make a copy of some other object, use the copy module:

>>> import copy
>>> newObj = copy.copy(myObj) # shallow copy
>>> newObj2 = copy.deepcopy(myObj2) # deep copy

2.1 Strings
Python represents strings, i.e. a sequence of characters, using the im-
mutable str type. In Python, all objects have two output forms: while
string forms are designed to be human-readable, representational forms are
designed to produce an output that if fed to a Python interpreter, repro-
duces the represented object. In the future, when we write our own classes,
it will be important to defined the string representation of our our objects.

Unicode Strings
Pythons Unicode encoding is used to include a special characters in the
string (for example, whitespace). Starting from Python 3, all strings are
now Unicode, not just plain bytes. To create a Unicode string, we use the
u prefix:
>>> uGoodbye\u0020World !
Goodbye World !

In the example above, the escape sequence indicates the Unicode charac-
ter with the ordinal value 0x0020. It is also useful to remember that in
general ASCII representations are given by only 8-bits while the Unicode
representation needs 16-bits.
28 CHAPTER 2. BUILT-IN SEQUENCE TYPES

Methods to Add and Format Strings

The join(list1) Method:

Joins all the strings in a list into one string. While we could use + to
concatenate these strings, when a large volume of data is involved, this
method becomes much less efficient than using join():
>>> slayer = ["Buffy", "Anne", "Summers"]
>>> " ".join(slayer)
Buffy Anne Summers
>>> "-<>-".join(slayer)
Buffy-<>-Anne-<>-Summers
>>> "".join(slayer)
BuffyAnneSummers

join() can also be used with the built-in reversed() method:

>>> "".join(reversed(slayer))
SummersAnneBuffy

The rjust(width[, fillchar]) and ljust(width[, fillchar]) Meth-

ods:

Some formation (aligning) can be obtained with the methods rjust() (add
only at the end), ljust() (add only at the start):
>>> name = "Agent Mulder"
>>> name.rjust(50, -)
-----------------------------Agent Mulder

The format() Method:

Used to format or add variable values to a string:

>>> "{0} {1}".format("Im the One!", "Im not")
"Im the One! Im not"
>>> "{who} turned {age} this year!".format(who="Buffy", age=17)
She turned 88 this year
>>> "The {who} was {0} last week".format(12, who="boy")
Buffy turned 17 this year!
2.1. STRINGS 29

From Python 3.1 it is possible to omit field names, in which case Python
will in effect put them in for us, using numbers starting from 0. For example:
>>> "{} {} {}".format("Python", "can", "count")
Python can count

However, using the operator + would allow a more concise style here. This
method allows three specifiers: s to force string form, r to force represen-
tational form, and a to force representational form but only using ASCII
characters:
>>> import decimal
>>> "{0} {0!s} {0!r} {0!a}".format(decimal.Decimal("99.9"))
"99.9 99.9 Decimal(99.9) Decimal(99.9)"

String (Mapping) Unpacking

The mapping unpacking operator is ** and it produces a key-value list

suitable for passing to a function. The local variables that are currently in
scope are available from the built-in locals() and this can be used to feed
the format() method:
>>> hero = "Buffy"
>>> number = 999
>>> "Element {number} is a {hero}".format(**locals())
Element 999 is a Buffy

Splitting Methods for Strings

The splitlines(f) Method:

Returns the list of lines produced by splitting the string on line terminators,
stripping the terminators unless f is True:
>>> slayers = "Buffy\nFaith"
>>> slayers.splitlines()
[Buffy, Faith]
30 CHAPTER 2. BUILT-IN SEQUENCE TYPES

The split(t, n) Method:

Returns a list of strings splitting at most n times on string t. If n is not
given, it splits as many times as possible. If t is not given, it splits on
whitespace:
>>> slayers = "Buffy*Slaying-Vamps*16"
>>> fields = slayers.split("*")
>>> fields
[Buffy, Slaying-Vamps, 16]
>>> job = fields[1].split("-")
>>> job
[Slaying, Vamps]

We can use split() to write our own method for erasing spaces from
strings:
>>> def erase_space_from_string(string):
... s1 = string.split(" ")
... s2 = "".join(s1)
... return s2

A similar method, rsplit(), splits the string from right to left.

Strip Methods for Strings

The strip(chars) Method:
Returns a copy of the string with leading and trailing whitespace (or the
characters chars) removed:
>>> slayers = "Buffy and Faith999"
>>> slayers.strip("999")
Buffy and Faith

The program bellow uses strip() to list every word and the number of
the times they occur in alphabetical order for some file:3
[general_problems/strings/count_unique_words.py]

import string
import sys
3
A similar example is shown in the Default Dictionaries section.
2.1. STRINGS 31

def count_unique_word():
words = {} # create an empty dictionary
strip = string.whitespace + string.punctuation + string.digits +
"\""
for filename in sys.argv[1:]:
with open(filename) as file:
for line in file:
for word in line.lower().split():
word = word.strip(strip)
if len(word) > 2:
words[word] = words.get(word,0) +1
for word in sorted(words):
print("{0} occurs {1} times.".format(word, words[word]))

Similar methods are: lstrip(), which return a copy of the string with
all whitespace at the beginning of the string stripped away; and rstrip(),
which returns a copy of the string with all whitespace at the end of the
string stripped away.

Methods for Changing the Case

The swapcase() method returns a copy of the string with uppercase char-
acters lowercased and lowercase characters uppercased.
>>> slayers = "Buffy and Faith"
>>> slayers.swapcase()
bUFFY AND fAITH

In the same way:

? capitalize() returns a copy of the string with only the first character
in uppercase;

? lower() returns a copy of the original string, but with all characters
in lowercase;

? upper() returns a copy of the original string, but with all characters
in uppercase.
32 CHAPTER 2. BUILT-IN SEQUENCE TYPES

Methods for Searching

The index(x) and find(x) Methods:
There is two methods to find the position of one string inside another. One
is index(x), which returns the index position of the substring x, or raises a
ValueError exception on failure. The other is find(x), which returns the
index position of the substring x, or -1 on failure:
>>> slayers = "Buffy and Faith"
>>> slayers.find("y")
4
>>> slayers.find("k")
-1
>>> slayers.index("k")
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ValueError: substring not found
>>> slayers.index("y")
4

An adaptation of the previous methods are: rfind(string), which re-

turns the index within the string of the last (from the right) occurrence of
string; and rindex(string), which returns the index within the string of
the last (from the right) occurrence of string, causing an error if it cannot
be found.

The count(t, start, end) Method:

Returns the number of occurrences of the string t in the string s:
>>> slayer = "Buffy is Buffy is Buffy"
>>> slayer.count("Buffy", 0, -1)
2
>>> slayer.count("Buffy")
3

The replace(t, u, n) Method:

Returns a copy of the string with every (or a maximum of n if given) occur-
rences of string t replaced with string u:
>>> slayer = "Buffy is Buffy is Buffy"
>>> slayer.replace("Buffy", "who", 2)
2.2. TUPLES 33

who is who is Buffy

2.2 Tuples
A tuple is an immutable sequence type consisting of values separated by
commas:
>>> t1 = 1234, hello!
>>> t1[0]
1234
>>> t1
(12345, hello!)
>>> t2 = t2, (1, 2, 3, 4, 5) # nested
>>> u
((1234, hello!), (1, 2, 3, 4, 5))

It its possible to create tuples that contain mutable objects, such as

lists. Where strings have a character at every position, tuples have an object
reference at each position. Empty tuples are constructed by an empty pair
of parentheses. A tuple with one item is constructed by following a value
with a comma (it is not sufficient to enclose a single value in parentheses):
>>> empty = ()
>>> t1 = hello,
>>> len(empty)
0
>>> len(t1)
1
>>> t1
(hello,)

Methods for Tuples

The count(x) method counts how many times x appears in the tuple:
>>> t = 1, 5, 7, 8, 9, 4, 1, 4
>>> t.count(4)
2

The index(x) method returns the index position of the element x:

34 CHAPTER 2. BUILT-IN SEQUENCE TYPES

>>> t = 1, 5, 7
>>> t.index(5)
1

Tuple Unpacking
In Python, any iterable can be unpacked using the sequence unpacking op-
erator, *. When used with two or more variables on the left-hand side of an
assignment, one of which preceded by *, items are assigned to the variables,
with all those left over assigned to the starred variable:
>>> x, *y = (1, 2, 3, 4)
>>> x
1
>>> y
[2, 3, 4]

Named Tuples
Pythons package collections4 contains a sequence data type called named
tuple. This behaves just like the built-in tuple, with the same performance
characteristics, but it also carries the ability to refer to items in the tuple
by name as well as by index position. This allows the creation of aggregates
of data items:
>>> import collections
>>> MonsterTuple = collections.namedtuple("Monsters","name age
power")
>>> MonsterTuple = (Vampire, 230, immortal)
>>> MonsterTuple
(Vampire, 230, immortal)

The first argument to collections.namedtuple is the name of the custom

tuple data type to be created. The second argument is a string of space-
separated names, one for each item that the custom tuple will take. The
first argument and the names in the second argument must be valid Python
identifiers.
The example bellow shows a structured way of using named tuples to
organize a data structure:
4
We are going to use collections a lot...
2.3. LISTS 35

[general_problems/tuples/namedtuple_example.py]

from collections import namedtuple

def namedtuple_example():
show an example for named tuples
>>> namedtuple_example()
slayer

sunnydale = namedtuple(name, [job, age])
buffy = sunnydale(slayer, 17)
print(buffy.job)

if __name__ == __main__:
namedtuple_example()

2.3 Lists
In computer science, arrays are a very simple data structure where elements
are sequentially stored in continued memory and linked lists are structures
where several separated nodes link to each other. Iterating over the contents
of the data structure is equally efficient for both kinds, but directly accessing
an element at a given index has O(1) (complexity) runtime5 in an array,
while it is O(n) in a linked list with n nodes (where you would have to
transverse the list from the beginning). Furthermore, in a linked list, once
you know where you want to insert something, insertion is O(1), no matter
how many elements the list has. For arrays, an insertion would have to move
all elements that are to the right of the insertion point or moving all the
elements to a larger array if needed, being then O(n).
In Python, the closest object to an array is a list, which is a dynamic re-
sizing array and it does not have anything to do with linked lists. Why men-
tion linked lists? Linked lists are a very important abstract data structure
(we will see more about them in a following chapter) and it is fundamental
to understand what makes it so different from arrays (or Pythons lists) for
when we need to select the right data structure for a specific problem.
5
The Big-O notation is a key to understand algorithms! We will learn more about this
in the following chapters and use the concept extensively in our studies. For now just keep
in mine that O(1) times  O(n)  O(n2 ), etc...
36 CHAPTER 2. BUILT-IN SEQUENCE TYPES

Lists in Python are created by comma-separated values, between square

brackets. List items do not need to have all the same data type. Unlike
strings which are immutable, it is possible to change individual elements of
a list (lists are mutable):
>>> q = [2, 3]
>>> p = [1, q, 4]
>>> p[1].append("buffy")
>>> p
[1, [2, 3, buffy], 4]
>>> q
[2, 3, buffy]
>>> q
[2, 3, buffy]

To insert items, lists perform best (O(1)) when items are added or re-
moved at the end, using the methods append() and pop(), respectively. The
worst performance (O(n)) occurs when we perform operations that need to
search for items in the list, for example, using remove() or index(), or
using in for membership testing.6
If fast searching or membership testing is required, a collection type such
as a set or a dictionary may be a more suitable choice (as we will see in the
next chapter). Alternatively, lists can provide fast searching if they are kept
in order by being sorted (we will see searching methods that perform on
O(log n) for sorted sequences, particular the binary search, in the following
chapters).

Adding Methods for Lists

The append(x) Method:
Adds a new element at the end of the list. It is equivalent to list[len(list):]=
[x]:
>>> people = ["Buffy", "Faith"]
>>> people.append("Giles")
>>> people
[Buffy, Faith, Giles]
>>> people[len(people):] = ["Xander"]
>>> people
[Buffy, Faith, Giles, Xander]
6
This explains why append() is so much more efficient than insert().
2.3. LISTS 37

The extend(c) Method:

This method is used to extend the list by appending all the iterable items
in the given list. Equivalent to a[len(a):]=L or using +=:
>>> people = ["Buffy", "Faith"]
>>> people.extend("Giles")
>>> people
[Buffy, Faith, G, i, l, e, s]
>>> people += "Willow"
>>> people
[Buffy, Faith, G, i, l, e, s, W, i, l, l, o,
w]
>>> people += ["Xander"]
>>> people
[Buffy, Faith, G, i, l, e, s, W, i, l, l, o,
w, Xander]

The insert(i, x) Method:

Inserts an item at a given position i: the first argument is the index of the
element before which to insert:
>>> people = ["Buffy", "Faith"]
>>> people.insert(1, "Xander")
>>> people
[Buffy, Xander, Faith]

Removing Methods for Lists

The remove() Method:
Removes the first item from the list whose value is x. Raises a ValueError
exception if not found:
>>> people = ["Buffy", "Faith"]
>>> people.remove("Buffy")
>>> people
[Faith]
38 CHAPTER 2. BUILT-IN SEQUENCE TYPES

>>> people.remove("Buffy")
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ValueError: list.remove(x): x not in list

The pop() Method:

Removes the item at the given position in the list, and then returns it. If
no index is specified, pop() returns the last item in the list:
>>> people = ["Buffy", "Faith"]
>>> people.pop()
Faith
>>> people
[Buffy]

The del Method:

It deletes the object reference, not the contend, i.e., it is a way to remove
an item from a list given its index instead of its value. This can also be used
to remove slices from a list:
>>> a = [-1, 4, 5, 7, 10]
>>> del a[0]
>>> a
[4, 5, 7, 10]
>>> del a[2:3]
>>> a
[4, 5, 10]
>>> del a # also used to delete entire variable

When an object reference is deleted and if no other object refers to its

data, Python schedules the data item to be garbage-collected.7

7
Garbage is a memory occupied by objects that are no longer referenced and garbage
collection is a form of automatic memory management, freeing the memory occupied by
the garbage.
2.3. LISTS 39

Searching and Sorting Methods for Lists

The index(x) Method:

Returns the index in the list of the first item whose value is x:
>>> people = ["Buffy", "Faith"]
>>> people.index("Buffy")
0

The count(x) Method:

Returns the number of times x appears in the list:

>>> people = ["Buffy", "Faith", "Buffy"]
>>> people.count("Buffy")
2

The sort() Method:

Sorts the items of the list, in place:

>>> people = ["Xander", "Faith", "Buffy"]
>>> people.sort()
>>> people
[Buffy, Faith, Xander]

The reverse() Method:

Reverses the elements of the list, in place:

>>> people = ["Xander", "Faith", "Buffy"]
>>> people.reverse()
>>> people
[Buffy, Faith, Xander]

List Unpacking
Similar to tuple unpacking:
40 CHAPTER 2. BUILT-IN SEQUENCE TYPES

>>> first, *rest = [1,2,3,4,5]

>>> first
1
>>> rest
[2, 3, 4, 5]

Python also has a related concept called starred arguments, that can be
used as a passing argument for a function:
>>> def example_args(a, b, c):
... return a * b * c # here * is the multiplication operator
>>> L = [2, 3, 4]
>>> example_args(*L)
24
>>> example_args(2, *L[1:])
24

List Comprehensions
A list comprehension is an expression and loop (with an optional condition)
enclosed in brackets:8
[item for item in iterable]
[expression for item in iterable]
[expression for item in iterable if condition]

Some examples of list comprehensions are shown below:

>>> a = [y for y in range(1900, 1940) if y%4 == 0]
>>> a
[1900, 1904, 1908, 1912, 1916, 1920, 1924, 1928, 1932, 1936]
>>> b = [2**i for i in range(13)]
>>> b
[1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096]
>>> c = [x for x in a if x%2==0]
>>> c
[0, 4, 16, 36, 64]
>>> d = [str(round(355/113.0,i)) for i in range(1,6)]
8
The Google Python Style guide endorses list comprehensions and generator expres-
sions saying that they provide a concise and efficient way to create lists and iterators
without resorting to the use of map(), filter(), or lambda.
2.3. LISTS 41

>>> d
[3.1, 3.14, 3.142, 3.1416, 3.14159]
>>> words = Buffy is awesome and a vampire slayer.split()
>>> e = [[w.upper(), w.lower(), len(w)] for w in words]
>>> for i in e:
... print(i)
...
[BUFFY, buffy, 5]
[IS, is, 2]
[AWESOME, awesome, 7]
[AND, and, 3]
[A, a, 1]
[VAMPIRE, vampire, 7]
[SLAYER, slayer, 6]

The Google Python Style Guide advocates that list comprehensions

should only be used for simple cases, when each portion fits in one line
(no multiple for clauses or filter expressions):
[Good]
result = []
for x in range(10):
for y in range(5):
if x * y > 10:
result.append((x, y))

for x in range(5):
for y in range(5):
if x != y:
for z in range(5):
if y != z:
yield (x, y, z)

return ((x, complicated_transform(x))

for x in long_generator_function(parameter)
if x is not None)

squares = [x * x for x in range(10)]

eat(jelly_bean for jelly_bean in jelly_beans

if jelly_bean.color == black)

[Bad]
result = [(x, y) for x in range(10) for y in range(5) if x * y >
10]
42 CHAPTER 2. BUILT-IN SEQUENCE TYPES

return ((x, y, z)
for x in xrange(5)
for y in xrange(5)
if x != y
for z in xrange(5)
if y != z)

Runtime Analysis for Lists

To understand better the performance of Pythons lists, we can benchmark
the lists methods. In the snippet bellow, we use Pythons timeit module
to create a Timer object whose first parameter is what we want to time
and the second parameter is a statement to set up the test. The timeit
module will time how long it takes to execute the statement some number
of times (one million times by default). When its done, it returns the time
as a floating point value representing the total number of seconds:
[general_problems/lists/runtime_lists_with_timeit_module.py]

def test1():
l = []
for i in range(1000):
l = l + [i]
def test2():
l = []
for i in range(1000):
l.append(i)

def test3():
l = [i for i in range(1000)]

def test4():
l = list(range(1000))

if __name__ == __main__:
import timeit
t1 = timeit.Timer("test1()", "from __main__ import test1")
print("concat ",t1.timeit(number=1000), "milliseconds")
t2 = timeit.Timer("test2()", "from __main__ import test2")
print("append ",t2.timeit(number=1000), "milliseconds")
t3 = timeit.Timer("test3()", "from __main__ import test3")
print("comprehension ",t3.timeit(number=1000), "milliseconds")
2.4. BYTES AND BYTE ARRAYS 43

t4 = timeit.Timer("test4()", "from __main__ import test4")

print("list range ",t4.timeit(number=1000), "milliseconds")

""" The results are:

(concat , 2.366791009902954, milliseconds)
(append , 0.16743111610412598, milliseconds)
(comprehension , 0.06446194648742676, milliseconds)
(list range , 0.021029949188232422, milliseconds)

So we see the following pattern for lists:

Operation Big-O Efficiency

index [] O(1)
index assignment O(1)
append O(1)
pop() O(1)
pop(i) O(n)
insert(i,item) O(n)
del operator O(n)
iteration O(n)
contains (in) O(n)
get slice [x:y] O(k)
del slice O(n)
set slice O(n+k)
reverse O(n)
concatenate O(k)
sort O(n log n)
multiply O(nk)
"""

2.4 Bytes and Byte Arrays

Python provides two data types for handling raw bytes: bytes which is
immutable, and bytearray, which is mutable. Both types hold a sequence
of zero of more 8-bit unsigned integers in the range 0 ... 255. The byte type
is very similar to the string type and the bytearray provides mutating
methods similar to lists.
44 CHAPTER 2. BUILT-IN SEQUENCE TYPES

Bits and Bitwise Operations

Bitwise operations can be very useful to manipulate numbers represented as
bits (for example, reproduce an division without using the division opera-
tion). We can you quickly compute 2x by the left-shifting operation: 1  x.
We can also quickly verify whether a number is a power of 2 by checking
whether x&(x 1) is 0 (if x is not an even power of 2, the highest position
of x with a 1 will also have a 1 in x 1, otherwise, x will be 100...0 and
x 1 will be 011...1; add them together they will return 0).
Chapter 3

Collection Data Structures

Differently from the last chapters sequence data structures, where the data
can be ordered or sliced, collection data structures are containers which
aggregates data without relating them. Collection data structures also have
some proprieties that sequence types have:

? membership operator (for example, using in);

? a size method (given by len(seq)); and

? iterability (we can iterate the data in loops).

In Python, built-in collection data types are given by sets and dicts. In
addition, many useful collection data are found in the package collections,
as we will discuss in the last part of this chapter.

3.1 Sets
In Python, a Set is an unordered collection data type that is iterable, mu-
table, and has no duplicate elements. Sets are used for membership testing
and eliminating duplicate entries. Sets have O(1) insertion, so the runtime
of union is O(m + n). For intersection, it is only necessary to transverse the
smaller set, so the runtime is O(n). 1
1
Pythons collection package has supporting for Ordered sets. This data type enforces
some predefined comparison for their members.

45
46 CHAPTER 3. COLLECTION DATA STRUCTURES

Frozen Sets
Frozen sets are immutable objects that only support methods and opera-
tors that produce a result without affecting the frozen set or sets to which
they are applied.

Adding Methods for Sets

The add(x) Method:
Adds the item x to set if it is not already in the set:
>>> people = {"Buffy", "Angel", "Giles"}
>>> people.add("Willow")
>>> people
{Willow, Giles, Buffy, Angel}

The s.update(t) or s| = t Methods:

They both return a set s with elements added from t.

The s.union(t) or s|t Methods:

They both perform union of the two sets.

The s.intersection(t) or s&t Methods:

They both return a new set that has each item from the sets:
>>> people = {"Buffy", "Angel", "Giles", "Xander"}
>>> people.intersection({"Angel", "Giles", "Willow"})
{Giles, Angel}

Subtracting Methods for Sets

The s.difference(t) or s t Methods:
They both return a new set that has every item that is not in the second
set:
3.1. SETS 47

>>> people = {"Buffy", "Angel", "Giles", "Xander"}

>>> vampires = {"Spike", "Angel", "Drusilia"}
>>> people.difference(vampires)
{Xander, Giles, Buffy}

The clear() Method:

Removes all the items in the set:
>>> people = {"Buffy", "Angel", "Giles"}
>>> people.clear()
>>> people
set()

The discard(x), remove(x), and pop() Methods:

discard(x) removes the item x from the set. remove(x) removes the item
x from the set or raises a KeyError exception if the element is not in the set.
pop() returns and removes a random item from the set or raises a KeyError
exception if the set is empty.

Sets with Lists and Dictionaries

Sets can be casted from lists. For example, the snippet below shows some
of the available set operations on lists:
[general_problems/sets/set_operations_with_lists.py]

def difference(l1):
""" return the list with duplicate elements removed """
return list(set(l1))

def intersection(l1, l2):

""" return the intersection of two lists """
return list(set(l1) & set(l2))

def union(l1, l2):

""" return the union of two lists """
return list(set(l1) | set(l2))

def test_sets_operations_with_lists():
l1 = [1,2,3,4,5,9,11,15]
48 CHAPTER 3. COLLECTION DATA STRUCTURES

l2 = [4,5,6,7,8]
l3 = []
assert(difference(l1) == [1, 2, 3, 4, 5, 9, 11, 15])
assert(difference(l2) == [8, 4, 5, 6, 7])
assert(intersection(l1, l2) == [4,5])
assert(union(l1, l2) == [1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 15])
assert(difference(l3) == [])
assert(intersection(l3, l2) == l3)
assert(sorted(union(l3, l2)) == sorted(l2))
print(Tests passed!)

if __name__ == __main__:
test_sets_operations_with_lists()

We can also use sets proprieties in dictionaries:2

[general_problems/dicts/set_operations_dict.py]

from collections import OrderedDict

def set_operations_with_dict():
pairs = [(a, 1), (b,2), (c,3)]
d1 = OrderedDict(pairs)
print(d1) # (a, 1), (b, 2), (c, 3)

d2 = {a:1, c:2, d:3, e:4}

print(d2) # {a: 1, c: 2, e: 4, d: 3}

union = d1.keys() & d2.keys()

print(union) # {a, c}

union_items = d1.items() & d2.items()

print(union_items) # {(a, 1)}

subtraction1 = d1.keys() - d2.keys()

print(subtraction1) # {b}

subtraction2 = d2.keys() - d1.keys()

print(subtraction2) # {d, e}

subtraction_items = d1.items() - d2.items()

2
Sets properties can be used on the dicts attributes items() and keys() attributes,
however values() do not support set operations.
3.2. DICTIONARIES 49

print(subtraction_items) # {(b, 2), (c, 3)}

we can remove keys from a dict doing:

d3 = {key:d2[key] for key in d2.keys() - {c, d}}
print(d3) {a: 1, e: 4}

if __name__ == __main__:
set_operations_with_dict()

3.2 Dictionaries
Dictionaries in Python are implemented using hash tables. Hashing func-
tions compute some random integer value from an arbitrary object in con-
stant time, that can be used as an index into an array:
>>> hash(42)
42
>>> hash("hello")
355070280260770553

A dict is a collection mapping type that is iterable and supports the

membership operator in and the size function len(). Mappings are collec-
tions of key-value items, providing methods for accessing items and their
keys and values. When iterated, unordered mapping types provide their
items in an arbitrary order.
Accessing dictionaries has runtime O(1) so they are used to keep counts
of unique items (for example, counting the number of each unique word in a
file) and for fast membership test. Dictionaries are mutable, so we can easily
add or remove items, but since they are unordered, they have no notion of
index position (so that they cannot be sliced or striped):
>>> tarantino = {}
>>> tarantino[name] = Quentin Tarantino
>>> tarantino[job] = director
>>> tarantino
{job: director, name: Quentin Tarantino}
>>>
>>> sunnydale = dict({"name":"Buffy", "age":16, "hobby":"slaying"})
>>> sunnydale
{hobby: slaying, age: 16, name: Buffy}
>>>
50 CHAPTER 3. COLLECTION DATA STRUCTURES

>>> sunnydale = dict(name="Giles", age=45, hobby="Watch")

>>> sunnydale
{hobby: Watch, age: 45, name: Giles}
>>>
>>> sunnydale = dict([("name", "Willow"), ("age",15), ("hobby",
"nerding")])
>>> sunnydale
{hobby: nerding, age: 15, name: Willow}

Adding Methods for Dictionaries

The setdefault(key[, default]) Method:

The setdefault() method is used when we want to access a key in the

dictionary without being sure that this key exists (if we simply try to ac-
cess a non-existent key in a dictionary, we will get an exception). With
setdefault(), if key is in the dictionary, we get the value to it. If not, we
successfully insert the new key with the value of default:
[general_problems/dicts/setdefault_example.py]

def usual_dict(dict_data):
newdata = {}
for k, v in dict_data:
if k in newdata:
newdata[k].append(v)
else:
newdata[k] = [v]
return newdata

def setdefault_dict(dict_data):
newdata = {}
for k, v in dict_data:
newdata.setdefault(k, []).append(v)
return newdata

def test_setdef(module_name=this module):

dict_data = ((key1, value1),
(key1, value2),
(key2, value3),
3.2. DICTIONARIES 51

(key2, value4),
(key2, value5),)
print(usual_dict(dict_data))
print(setdefault_dict(dict_data))

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_setdef()

The update([other]) Method:

Updates the dictionary with the key/value pairs from other, overwriting
existing keys. .

Searching Methods for Dictionaries

The get(key) Method:
Returns the keys associated value or None if the key is not in the dictionary:
>>> sunnydale = dict(name="Xander", age=17, hobby="winning")
>>> sunnydale.get("hobby")
winning

The items(), values(), and keys() Methods:

The items(), keys(), and values() methods all return dictionary views.
A dictionary view is effectively a read-only iterable object that appears to
hold the dictionarys items or keys or values:
>>> sunnydale = dict(name="Xander", age=17, hobby="winning")
>>> sunnydale.items()
dict_items([(hobby, winning), (age, 17), (name, Xander)])
>>> sunnydale.values()
dict_values([winning, 17, Xander])
>>> sunnydale.keys()
dict_keys([hobby, age, name])
52 CHAPTER 3. COLLECTION DATA STRUCTURES

Removing Methods for Dictionaries

The pop() and popitem() Methods:
The pop() method removes an arbitrary item from the dictionary, return-
ing it. The popitem() method removes an arbitrary (key, value) from the
dictionary, also returning it.

The clear() Method:

Removes all the items in the dictionary:
>>> sunnydale.clear()
>>> sunnydale
{}

Runtime Analysis for Dictionaries

We can study the performance of Pythons dictionaries by benchmarking
their available methods. For instance, in the snippet below we were confirm
that the membership operation for lists is O(n) and for dictionaries is O(1)):
[general_problems/dicts/runtime_dicts_with_timeit_module.py]

import timeit
import random

for i in range(10000,1000001,20000):
t = timeit.Timer("random.randrange(%d) in x"%i, "from __main__
import random,x")
x = list(range(i))
lst_time = t.timeit(number=1000)
x = {j:None for j in range(i)}
d_time = t.timeit(number=1000)
print("%d,%10.3f,%10.3f" % (i, lst_time, d_time))

""" There results are:

10000, 0.192, 0.002
30000, 0.600, 0.002
50000, 1.000, 0.002
70000, 1.348, 0.002
90000, 1.755, 0.002
3.2. DICTIONARIES 53

110000, 2.194, 0.002

130000, 2.635, 0.002
150000, 2.951, 0.002
170000, 3.405, 0.002
190000, 3.743, 0.002
210000, 4.142, 0.002
230000, 4.577, 0.002
250000, 4.797, 0.002
270000, 5.371, 0.002
290000, 5.690, 0.002
310000, 5.977, 0.002

So we can see the linear tile for lists, and constant for dict!

Big-O Efficiency of Python Dictionary Operations

Operation Big-O Efficiency
copy O(n)
get item O(1)
set item O(1)
delete item O(1)
contains (in) O(1)
iteration O(n)
"""

Iterating over Dictionaries

A loop over a dictionary iterates over its keys by default. The keys will
appear in an arbitrary order but we can use sorted() to iterate over the
items in a sorted way. This also works for the attributes keys(), values(),
and items():
>>> for key in sorted(dict.keys()):
... print key, dict[key]

An useful tool to create a list of key-items for a dictionary is by using

generators:
def items_in_key_order(d):
for key in sorted(d):
yield key, d[key]
54 CHAPTER 3. COLLECTION DATA STRUCTURES

Dictionaries also support reverse iteration using reversed(). In addi-

tion, it is good to note that the Google Python Style guide advices that
default iterators should be used for types that support them:
[Good] for key in adict: ...
if key not in adict: ...

[Bad] for key in adict.keys(): ...

if not adict.has_key(key): ...

Branching using Dictionaries

We can use dictionaries as a simple way to write a branching menu, where
if action == "a":
add_to_dict(db)
elif action == "e":
edit_dict(db)

can be reduced to
functions = dict(a=add_to_dict, e=edit_dict,...)
functions[actions](db)

3.3 Pythons collection Data Types

Pythons collections module implements specialized container data types
providing high-performance alternatives to the general purpose built-in con-
tainers.

Default Dictionaries
Default dictionaries are an additional unordered mapping type provide
by Pythons collections.defaultdict. They have all the operators and
methods that a built-in dictionary provide, but they also gracefully handle
missing keys:
[general_examples/dicts/defaultdict_example.py]
3.3. PYTHONS COLLECTION DATA TYPES 55

from collections import defaultdict

def defaultdict_example():
show some examples for defaultdicts
pairs = {(a, 1), (b,2), (c,3)}

d1 = {}
for key, value in pairs:
if key not in d1:
d1[key] = []
d1[key].append(value)
print(d1)

d2 = defaultdict(list)
for key, value in pairs:
d2[key].append(value)
print(d2)

if __name__ == __main__:
defaultdict_example()

Ordered Dictionaries
Ordered dictionaries are an ordered mapping type provided by Pythons
collections.OrderedDict. They have all the methods and properties of a
built-in dict, but in addition they store items in the insertion order:
[general_examples/dicts/OrderedDict_example.py]

from collections import OrderedDict

pairs = [(a, 1), (b,2), (c,3)]

d1 = {}
for key, value in pairs:
if key not in d1:
d1[key] = []
d1[key].append(value)
for key in d1:
print(key, d1[key])

d2 = OrderedDict(pairs)
for key in d2:
56 CHAPTER 3. COLLECTION DATA STRUCTURES

print(key, d2[key])

if __name__ == __main__:
OrderedDict_example()

"""
a [1]
c [3]
b [2]
a 1
b 2
c 3
"""

We can create ordered dictionaries incrementally:

>>> tasks = collections.OrderedDict()
>>> tasks[8031] = "Backup"
>>> tasks[4027] = "Scan Email"
>>> tasks[5733] = "Build System"
>>> tasks
OrderedDict([(8031, Backup), (4027, Scan Email), (5733, Build
System)])

If we change a key value, the order is not changed. To move an item to

the end we should delete and re-insert it. We can also call popitem() to
remove and return the last key-value item in the ordered dictionary. Or we
can call do it in a FIFO order3
In general, using an ordered dictionary to produce a sorted dictionary
makes sense only if we expect to iterate over the dictionary multiple times,
and if we do not expect to do any insertions (or very few).

Counter Dictionaries
A specialised Counter type (subclass for counting hashable objects) is pro-
vided by Pythons collections.Counter:
[general_examples/dicts/Counter_example.py]

from collections import Counter

3
In computer science, FIFO means first-in first-out. Pythons lists append and pop
items by the end so they are LIFO, last-in, last-out.
3.3. PYTHONS COLLECTION DATA TYPES 57

def Counter_example():
show some examples for Counter
it is a dictionary that maps the items to the number of
occurrences
seq1 = [1, 2, 3, 5, 1, 2, 5, 5, 2, 5, 1, 4]
seq_counts = Counter(seq1)
print(seq_counts)

we can increment manually or use the update() method

seq2 = [1, 2, 3]
seq_counts.update(seq2)
print(seq_counts)

seq3 = [1, 4, 3]
for key in seq3:
seq_counts[key] += 1
print(seq_counts)

also, we can use set operations such as a-b or a+b

seq_counts_2 = Counter(seq3)
print(seq_counts_2)
print(seq_counts + seq_counts_2)
print(seq_counts - seq_counts_2)

if __name__ == __main__:
Counter_example()

"""
Counter({5: 4, 1: 3, 2: 3, 3: 1, 4: 1})
Counter({1: 4, 2: 4, 5: 4, 3: 2, 4: 1})
Counter({1: 5, 2: 4, 5: 4, 3: 3, 4: 2})
Counter({1: 1, 3: 1, 4: 1})
Counter({1: 6, 2: 4, 3: 4, 5: 4, 4: 3})
Counter({1: 4, 2: 4, 5: 4, 3: 2, 4: 1})
"""
58 CHAPTER 3. COLLECTION DATA STRUCTURES

3.4 Additional Exercises

Counting Frequency of Items
In the example bellow we use collections.Counter()s most common()
method to find the top N recurring words in a sequence:
[general_problems/dicts/find_top_N_recurring_words.py]

from collections import Counter

def find_top_N_recurring_words(seq, N):

dcounter = Counter()
for word in seq.split():
dcounter[word] += 1
return dcounter.most_common(N)

def test_find_top_N_recurring_words(module_name=this module):

seq = buffy angel monster xander a willow gg buffy the monster
super buffy angel
N = 3
assert(find_top_N_recurring_words(seq, N) == [(buffy, 3),
(monster, 2), (angel, 2)])

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_find_top_N_recurring_words()

The program below counts all the unique words in a file:

[general_problems/dicts/count_unique_words.py]

import collections
import string
import sys

def count_unique_word():
words = collections.defaultdict(int)
strip = string.whitespace + string.punctuation + string.digits +
"\""
3.4. ADDITIONAL EXERCISES 59

for filename in sys.argv[1:]:

with open(filename) as file:
for line in file:
for word in line.lower().split():
word = word.strip(strip)
if len(word) > 2:
words[word] = +1
for word in sorted(words):
print("{0} occurs {1} times.".format(word, words[word]))

Anagrams
The following program finds whether two words are anagrams. Since sets
do not count occurrence, and sorting a list is O(n log n), hash tables can
be the best solution in this case. The procedure we use is: we scan the
first string and add all the character occurrences. Then we scan the second
string, decreasing all the character occurrences. In the end, if all the entries
are zero, the string is an anagram:
[general_problems/dicts/verify_two_strings_are_anagrams.py]

def verify_two_strings_are_anagrams(str1, str2):

ana_table = {key:0 for key in string.ascii_lowercase}

for i in str1:
ana_table[i] += 1

for i in str2:
ana_table[i] -= 1

# verify whether all the entries are 0

if len(set(ana_table.values())) < 2: return True
else: return False

def test_verify_two_strings_are_anagrams():
str1 = marina
str2 = aniram
assert(verify_two_strings_are_anagrams(str1, str2) == True)
str1 = google
str2 = gouglo
assert(verify_two_strings_are_anagrams(str1, str2) == False)
print(Tests passed!)
60 CHAPTER 3. COLLECTION DATA STRUCTURES

if __name__ == __main__:
test_verify_two_strings_are_anagrams()

Another way to find whether two words are anagrams is using the hashing
functions proprieties, where every different amount of characters should
give a different result. In the following program, ord() returns an integer
representing the Unicode code point of the character when the argument is
a unicode object, or the value of the byte when the argument is an 8-bit
string:
[general_problems/dicts/find_anagram_hash_function.py]

def hash_func(astring, tablesize):

sum = 0
for pos in range(len(astring)):
sum = sum + ord(astring[pos])
return sum%tablesize

def find_anagram_hash_function(word1, word2):

tablesize = 11
return hash_func(word1, tablesize) == hash_func(word2,
tablesize)

def test_find_anagram_hash_function(module_name=this module):

word1 = buffy
word2 = bffyu
word3 = bffya
assert(find_anagram_hash_function(word1, word2) == True)
assert(find_anagram_hash_function(word1, word3) == False)

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_find_anagram_hash_function()
3.4. ADDITIONAL EXERCISES 61

Sums of Paths
The following program uses two different dictionary containers to determine
the number of ways two dices can sum to a certain value:
[general_problems/dicts/find_dice_probabilities.py]

from collections import Counter, defaultdict

def find_dice_probabilities(S, n_faces=6):

if S > 2*n_faces or S < 2: return None

cdict = Counter()
ddict = defaultdict(list)

for dice1 in range(1, n_faces+1):

for dice2 in range(1, n_faces+1):
t = [dice1, dice2]
cdict[dice1+dice2] += 1
ddict[dice1+dice2].append( t)

return [cdict[S], ddict[S]]

def test_find_dice_probabilities(module_name=this module):

n_faces = 6
S = 5
results = find_dice_probabilities(S, n_faces)
print(results)
assert(results[0] == len(results[1]))

if __name__ == __main__:
test_find_dice_probabilities()

Finding Duplicates
The program below uses dictionaries to find and delete all the duplicate
characters in a string:
[general_problems/dicts/delete_duplicate_char_str.py]

import string
def delete_unique_word(str1):
table_c = { key : 0 for key in string.ascii_lowercase}
62 CHAPTER 3. COLLECTION DATA STRUCTURES

for i in str1:
table_c[i] += 1
for key, value in table_c.items():
if value > 1:
str1 = str1.replace(key, "")
return str1

def test_delete_unique_word():
str1 = "google"
assert(delete_unique_word(str1) == le)
print(Tests passed!)

if __name__ == __main__:
test_delete_unique_word()
Chapter 4

Pythons Structure and

Modules

4.1 Modules in Python

In Python, modules are defined using the built-in name def. When def is
executed, a function object is created together with its object reference. If
we do not define a return value, Python automatically returns None (like
in C, we call the function a procedure when it does not return a value).
An activation record happens every time we invoke a method: informa-
tion is put in the stack to support invocation. Activation records process in
the following order:

Activation Records
1. the actual parameters of the method are pushed onto the stack,

2. the return address is pushed onto the stack,

3. the top-of-stack index is incremented by the total amount required by

the local variables within the method,

4. a jump to the method.

The process of unwinding an activation record happens in the following

order:

1. the top-of-stack index is decremented by the total amount of memory

consumed,

63
64 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

2. the returned address is popped off the stack,

3. the top-of-stack index is decremented by the total amount of memory

by the actual parameters.

Default Values in Modules

Whenever you create a module, remember that mutable objects should not
be used as default values in the function or method definition:
[Good]
def foo(a, b=None):
if b is None:
b = []
[Bad]
def foo(a, b=[]):

The init .py File

A package is a directory that contains a set of modules and a file called
init .py . This is required to make Python treat the directories as
containing packages, preventing directories with a common name (such as
string) from hiding valid modules that occur later on the module search
path:
>>> import foldername.filemodulename

In the simplest case, it can just be an empty file, but it can also execute
initialization code for the package or set the all variable: init .py
to:
__all__ = ["file1", ...]

(with no .py in the end).

Moreover, the statement:
from foldername import *

means importing every object in the module, except those whose names
begin with , or if the module has a global all variable, the list in it.
4.1. MODULES IN PYTHON 65

Checking the Existence of a Module

To check the existence of a module, we use the flag -c:

$ python -c "import decimas"
Traceback (most recent call last):
File "<string>", line 1, in <module>
ImportError: No module named decimas

The name Variable

Whenever a module is imported, Python creates a variable for it called
name , and stores the modules name in this variable. In this case, every-
thing below the statement
if __name__ == __main__:

will not be executed. In the other hand, if we run the .py file directly,
Python sets name to main , and every instruction following the above
statement will be executed.

Byte-coded Compiled Modules

Byte-compiled code, in form of .pyc files, is used by the compiler to speed-up
the start-up time (load time) for short programs that use a lot of standard
modules.
When the Python interpreter is invoked with the -O flag, optimized
code is generated and stored in .pyo files. The optimizer removes assert
statements. This also can be used to distribute a library of Python code
in a form that is moderately hard to reverse engineer.

The sys Module

The variable sys.path is a list of strings that determines the interpreters
search path for modules. It is initialized to a default path taken from the en-
vironment variable PYTHONPATH, or from a built-in default. You can modify
it using standard list operations:
>>> import sys
>>> sys.path.append( /buffy/lib/ python )
66 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

The variables sys.ps1 and sys.ps2 define the strings used as primary
and secondary prompts. The variable sys.argv allows us to use the argu-
ments passed in the command line inside our programs:
import sys

def main():
print command line arguments
for arg in sys.argv[1:]:
print arg

if __name__ == "__main__":
main()

The built-in method dir() is used to find which names a module defines
(all types of names: variables, modules, functions). It returns a sorted list
of strings:
>>> import sys
>>> dir(sys)
[ __name__ , argv , builtin_module_names , copyright ,
exit , maxint , modules , path , ps1 ,
ps2 , setprofile , settrace , stderr , stdin ,
stdout , version ]

It does not list the names of built-in functions and variables. Therefore,
we can see that dir() is useful to find all the methods or attributes of an
object.

4.2 Control Flow

if
The if statement substitutes the switch or case statements in other lan-
guages:1
>>> x = int(input("Please enter a number: "))
>>> if x < 0:
... x = 0
... print "Negative changed to zero"
1
Note that colons are used with else, elif, and in any other place where a suite is to
follow.
4.2. CONTROL FLOW 67

>>> elif x == 0:
... print "Zero"
>>> elif x == 1:
... print "Single"
>>> else:
... print "More"

for
The for statement in Python differs from C or Pascal. Rather than always
iterating over an arithmetic progression of numbers (like in Pascal), or giving
the user the ability to define both the iteration step and halting condition
(as C), Pythons for statement iterates over the items of any sequence (e.g.,
a list or a string), in the order that they appear in the sequence:
>>> a = ["buffy", "willow", "xander", "giles"]
>>> for i in range(len(a)):
... print(a[i])
buffy
willow
xander
giles

False and True in Python

False is defined by the predefined constant False, or the special object None,
or by an empty sequence of collection (empty string , list [], or tuple
()). Anything else is True. It is also possible to assign the result of a
comparison or other Boolean expression to a variable:
>>> string1, string2, string3 = , monsters, aliens
>>> non_null = string1 or string2 or string3
>>> non_null
monsters

The Google Python Style guide sets the following rules for using implicit
False in Python:

? Never use == or != to compare singletons, such as the built-in variable

None. Use is or is not instead.

? Beware of writing if x: when you really mean if x is not None.

68 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

? Never compare a boolean variable to False using ==. Use if not x:

instead. If you need to distinguish False from None then chain the
expressions, such as if not x and x is not None:.

? For sequences (strings, lists, tuples), use the fact that empty sequences
are False, so if not seq: or if seq: is preferable to if len(seq):
or if not len(seq):.

? When handling integers, implicit False may involve more risk than
benefit, such as accidentally handling None as 0:

[Good]
if not users: print no users
if foo == 0: self.handle_zero()
if i % 10 == 0: self.handle_multiple_of_ten()

[Bad]
if len(users) == 0: print no users
if foo is not None and not foo: self.handle_zero()
if not i % 10: self.handle_multiple_of_ten()

yield vs. return

The difference between yield and return is that the former returns each
value to the caller and then only returns to the caller when all values to
return have been exhausted, and the latter causes the method to exit and
return control to the caller.
One great feature in Python is how it handles iterability. An iterator is a
container object that implements the iterator protocol and is based on two
methods: next, which returns the next item in the container, and iter
which returns the iterator itself. Since all methods in Python are virtual,
we are free to modify how to iterability works in our functions (and classes)
the way we like.
The yield paradigm becomes important in the context of generators,
which are a powerful tool for creating iterators. Generators are like regular
functions but instead of returning a final value in the end, they use the
yield statement to return data during execution. In other words, values
are extracted from the iterator one at time by calling its next () method
and at each of these calls, the yield expressions value is returned. This
happens until the final call, when a StopIteration is raised:
4.2. CONTROL FLOW 69

>>> def f(a):

... while a:
... yield a.pop()

Generators are very robust and efficient and they should considered every
time you deal with a function that returns a sequence or creates a loop. For
example, the following program implements a Fibonacci sequence using the
iterator paradigm:
def fib_generator():
a, b = 0, 1
while True:
yield b
a, b = b, a+b

if __name__ == __main__:
fib = fib_generator()
print(next(fib))
print(next(fib))
print(next(fib))
print(next(fib))

break vs. continue

The command break, breaks out of the smallest enclosing for or while
loop, and switches control to the statement following the innermost loop in
which the break statement appears (it breaks out of the loop).
In the other hand, continue, continues with the next iteration of the
loop, and switches control to the start of the loop (continues with the next
iteration of the loop).
Loop statements may have an else clause which is executed when the
loop terminates through exhaustion of the list (with for) or when the con-
dition becomes false (with while), but not when the loop is terminated by
a break statement.

The range() Function

This method generates lists containing arithmetic progressions. It is useful
when you need to iterate over a sequence of numbers:
>>> range(10)
70 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
>>> range(4, 10)
[4, 5, 6, 7, 8, 9]
>>> range(0, 10, 3)
[0, 3, 6, 9]

The enumerate() Function

Returns a tuple with the item and the index value from an iterable. For
example, we can use enumerate to write our own grep function, which gets a
word and files from the command line and outputs where the word appears:

[general_problems/modules/grep_word_from_files.py]

import sys

def grep_word_from_files():
word = sys.argv[1]
for filename in sys.argv[2:]:
with open(filename) as file:
for lino, line in enumerate(file, start=1):
if word in line:
print("{0}:{1}:{2:.40}".format(filename, lino,
line.rstrip()))

if __name__ == __main__:
if len(sys.argv) < 2:
print("Usage: grep_word_from_files.py word infile1
[infile2...]")
sys.exit()
else:
grep_word_from_files()

The zip() Function

The zip function takes two or more sequences and creates a new sequence
of tuples where each tuple contains one element from each list:
>>> a = [1, 2, 3, 4, 5]
>>> b = [a, b, c, d, e]
>>> zip(a, b)
4.2. CONTROL FLOW 71

<zip object at 0xb72d65cc>

>>> list(zip(a,b))
[(1, a), (2, b), (3, c), (4, d), (5, e)]

The filter(function, sequence) Function

This method returns a sequence consisting of those items from the sequence
for which function (item) is true:
>>> def f(x): return x % 2 != 0 and x % 3 != 0
>>> f(33)
False
>>> f(17)
True
>>> filter(f, range(2, 25))
[5, 7, 11, 13, 17, 19, 23]

The map(function, list) Function

A convenient way of turning a pair of lists into a list of pairs. It applies a
function to every item of an iterable and then returns a list of the results:
>>> def cube(x): return x*x*x
>>> map(cube, range(1, 11))
[1, 8, 27, 64, 125, 216, 343, 512, 729, 1000]
>>> seq = range(8)
>>> def square(x): return x*x
>>> map(None, seq, map(square, seq))
[(0, 0), (1, 1), (2, 4), (3, 9), (4, 16), (5, 25), (6, 36), (7, 49)]

The lambda Function

The lambda function is a dynamic way of compacting functions inside the
code. For example, the function:
>>> def area(b, h):
... return 0.5 * b * h
...
>>> area(5,4)
10.0
72 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

could be rather written as

>>> area = lambda b, h: 0.5 * b * h
>>> area(5, 4)
10.0

Lambda functions are very useful for creating keys in dictionaries:

>>> import collections
>>> minus_one_dict = collections.defaultdict(lambda: -1)
>>> point_zero_dict = collections.defaultdict(lambda: (0, 0))
>>> message_dict = collections.defaultdict(lambda: "No message")

4.3 File Handling

File handling is very easy in Python. For example, the program below shows
an program that reads a file and delete all the blank lines:
[general_problems/modules/remove_blank_lines.py]

import os
import sys

def read_data(filename):
lines = []
fh = None
try:
fh = open(filename)
for line in fh:
if line.strip():
lines.append(line)
except (IOError, OSError) as err:
print(err)
finally:
if fh is not None:
fh.close()
return lines

def write_data(lines, filename):

fh = None
try:
fh = open(filename, "w")
4.3. FILE HANDLING 73

for line in lines:

fh.write(line)
except (EnvironmentError) as err:
print(err)
finally:
if fh is not None:
fh.close()

def remove_blank_lines():
if len(sys.argv) < 2:
print ("Usage: noblank.py infile1 [infile2...]")

for filename in sys.argv[1:]:

lines = read_data(filename)
if lines:
write_data(lines, filename)

if __name__ == __main__:
remove_blank_lines()

Methods for File Handling

The open(filename, mode) Method:
Returns a file object. The mode argument is optional and read will be
assumed if it is omitted, the other options are:
? r for reading,

? w for writing (an existing file with the same name will be erased),

? a for appending (any data written to the file is automatically added

to the end),

? and r+ for both reading and writing.

fin = open(filename, encoding="utf8")

fout = open(filename, "w", encoding="utf8")

The read(size) Method:

Reads some quantity from the data and returns it as a string. Size is an op-
tional numeric argument, when it is omitted or negative, the entire contents
74 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

of the file will be read and returned. If the end of the file has been reached,
read() will return an empty string:
>>> f.read()
This is the entire file.\n
>>> f.read()

The readline() Method:

Reads a single line from the file. A newline character is left at the end of
the string, and is only omitted on the last line of the file if the file. This
makes the return value unambiguous.

The readlines() Method:

Return a list containing all the lines of data in the file. If given an optional
parameter size, it reads that many bytes from the file and enough more to
complete a line, and returns the lines from that. This is often used to allow
efficient reading of a large file by lines, but without having to load the entire
file in memory. Only complete lines will be returned:
>>> f.readlines()
[This is the first line of the file.\n, Second line of the
file\n]

The write() Method:

Writes the contents of a string to the file, returning None. Write bytes/bytear-
ray object to the file if opened in binary mode or a string object if opened
in text mode:
>>> f.write(This is a test\n)

The tell() and seek() Methods:

The tell() method returns an integer giving the file objects current posi-
tion in the file, measured in bytes from the beginning of the file.
To change the file objects position, use seek(offset, from-what).
The position is computed from adding offset to a reference point and the
4.3. FILE HANDLING 75

reference point is selected by the from-what argument. A from-what value

of 0 measures from the beginning of the file, 1 uses the current file position,
and 2 uses the end of the file as the reference point.

The close() Method:

Closes the file and free up any system resources taken up by the open file.
It returns True if the file is closed.

The input() Method:

Accepts input from the user. This function takes an optional string argu-
ment (which will be printed in the console), then it will wait for the user to
type in a response and to finish by pressing Enter (or Return).
If the user does not type any text but just presses Enter, the function
returns an empty string; otherwise, it returns a string containing what the
user typed, without any line terminator.
>>> def get_int(msg):
... while True:
... try:
... i = int(input(msg))
... return i
... except ValueError as err:
... print(err)
>>> age = get_int("Enter your age: ")

The peek(n) Method:

Returns n bytes without moving the file pointer position.

The fileno() Method:

Returns the underlying files file descriptor (available only for file objects
that have file descriptors).

The shutil Package

This package is useful for manipulating files in the system. For example, the
following snippet gets a file and an extension from the command line and
produces a copy of this file with its extension changed to the given string:
76 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

[general_problems/files/change_ext_file.py]

import os
import sys
import shutil

def change_file_ext():
if len(sys.argv) < 2:
print("Usage: change_ext.py filename.old_ext new_ext")
sys.exit()

name = os.path.splitext(sys.argv[1])[0] + "." + sys.argv[2]

print (name)

try:
shutil.copyfile(sys.argv[1], name)
except OSError as err:
print (err)

if __name__ == __main__:
change_file_ext()

The pickle Module

The pickle module can take almost any Python object (even some forms
of Python code!), and convert it to a string representation. This process is
called pickling, where reconstructing the object from the string representa-
tion is called unpickling.
If you have an object x, and a file object f that has been opened for
writing, the simplest way to pickle the object takes only one line of code:
>>> pickle.dump(x, f)

Then, to unpickle this object:

>>> x = pickle.load(f)

Exporting Data with Pickle

The example below shows an use for pickle:
4.3. FILE HANDLING 77

[general_problems/files/export_pickle.py]

import pickle

def export_pickle(data, filename=test.dat, compress=False):

fh = None
try:
if compress:
fh = gzip.open(filename, "wb") # write binary
else:
fh = open(filename, "wb") # compact binary pickle format
pickle.dump(data, fh, pickle.HIGHEST_PROTOCOL)

except(EnvironmentError, pickle.PickingError) as err:

print("{0}: export error:
{1}".format(os.path.basename(sys.argv[0], err)))
return False

finally:
if fh is not None:
fh.close()

def test_export_pickle():
mydict = {a: 1, b: 2, c: 3}
export_pickle(mydict)

if __name__ == __main__:
test_export_pickle()

In general, booleans, numbers, and strings, can be pickled as can in-

stances of classes and built-in collection types (providing they contain only
pickable objects, .e., their dict is pickable).

Reading Data with Pickle

The example below shows how to read a pickled data:
[general_problems/files/import.py]

import pickle

def import_pickle(filename):
fh = None
try:
78 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

fh = open(filename, "rb")
mydict2 = pickle.load(fh)
return mydict2

except (EnvironmentError) as err:

print ("{0}: import error:
{0}".format(os.path.basename(sys.arg[0]), err))
return false

finally:
if fh is not None:
fh.close()

def test_import_pickle():
pkl_file = test.dat
mydict = import_pickle(pkl_file)
print(mydict)

if __name__ == __main__:
test_import_pickle()

The struct Module

We can convert Python objects to and from suitable binary representation

using struct. This object can handle only strings with a specific length.
struct allows the creation of a function that takes a string and return a
byte object with an integer length count and a sequence of length count
UTF-8 encoded bytes for the text.
Some methods are: struct.pack() (takes a struct format string and val-
ues and returns a byte object), struct.unpack() (takes a format and a bytes
or bytearray object and returns a tuple of values), and struct.calcsize()
(takes a format and returns how many bytes a struct using the format will
occupy):
>>> data = struct.pack("<2h", 11, -9)
>>> items = struct.unpack("<2h", data) # little endian
4.4. MULTIPROCESSING AND THREADING 79

4.4 Multiprocessing and Threading

Each program in the operational system is a separate process. Each process
has one or more threads. If a process has several threads, they appear to
run simultaneously. Two approaches can be used to spread the workload in
programs:

Multiple processes Multiple processes have separate regions of memory

and can only communicate by special mechanisms. The processor
loads and saves a separate set of registers for each thread. It is incon-
venient for communication and data sharing. This is handled by the
subprocess module.

Multiple threads Multiple threads in a single process have access to the

same memory. They communicate simply by sharing data, provid-
ing ensure of one thread at time, handled by the threading module.
Threads share the processs resources, including the heap space. But
each thread still has it own stack.

Although Python has a threading mechanism, it does not support true

parallel execution. However, it is possible is to use parallel processes, which
in modern operating systems are really efficient.

The subprocess Module

Used to create a pair of parent-child programs, where the parent program
is run by the user and this in turn runs as many instances of the child
program, each with different work to do. Using child processing allow us
to take maximum advantage of multicore processor and leaves concurrency
issues to be handled by the operational system.

The threading Module

The complexity arises when we want to separate threads to share data, and
we need to be careful with a policy for locks and for avoiding deadlocks.
Each Python program has at least one thread, the main thread. To create
multiple threads we use the threading module::

? calling threading.thread() and passing a callable object,

? calling threading.threadclass and threading.threadsubclass.

80 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

The queue.queue class can handle all the locking internally: we can rely
on it to serialize accesses, meaning that only one thread at time has access
to the data (FIFO). The program will not terminate while it has any threads
running.
It might create a problem since once the worker threads have done their
work, they are finished but they are technically still running. The solu-
tion is to transform threads into daemons. In this case, the program will
terminate as soon as there is no daemon threads running. The method
queue.queue.join() blocks the end until the queue is empty.

Mutexes and Semaphores

A mutex is like a lock. Mutexes are used in parallel programming to ensure
that only one thread can access a shared resource at a time. For example,
say one thread is modifying an array. When it has gotten halfway through
the array, the processor switches to another thread. If we were not using
mutexes, the thread could try to modify the array as well, which is probably
not what we want.
Conceptually, a mutex is an integer that starts at 1. Whenever a thread
needs to alter the array, it locks the mutex. This causes the thread to wait
until the number is positive and then decreases it by one. When the thread
is done modifying the array, it unlocks the mutex, causing the number to
increase by 1. If we are sure to lock the mutex before modifying the array
and to unlock it when we are done, then we know that no two threads will
modify the array at the same time.
Semaphores are more general than mutexes. A semaphores integer may
start at a number greater than 1. The number at which a semaphore starts
is the number of threads that may access the resource at once. Semaphores
support wait and signal operations, which are analogous to the lock
and unlock operations of mutexes.

Deadlock and Spinlock

Deadlock is a problem that sometimes arises in parallel programming, when
two threads are stuck indefinitely. We can prevent deadlock if we assign an
order to our locks and require that locks will always be acquired in order
(this is a very general and not precise approach).
Spinlock is a form of busy waiting, that can be useful for high-performance
computing (HPC) (when the entire system is dedicated to a single applica-
tion and exactly one thread per core). It takes less time than a semaphore.
4.5. ERROR HANDLING IN PYTHON 81

The Google Python Style Guide for Threading

Google warns to not rely on the atomicity of built-in types (data types such
as dictionaries appear to have atomic operations, but they are actually based
on private methods such as hash or eq ). Queue modules data type
as the preferred way to communicate data between threads. Otherwise, use
the threading module and its locking primitives.

4.5 Error Handling in Python

There are two distinguishable kinds of errors when we compile a program
in Python: syntax errors (parsing errors) and exceptions (errors detected
during execution, not unconditionally fatal). While syntax errors will never
allow the program to be compiled, exceptions can only be detected in some
cases and for this reason they should be carefully handled.

Handling Exceptions
When an exception is raised and not handled, Python outputs a traceback
along with the exceptions error message. A traceback (sometimes called a
backtrace) is a list of all the calls made from the point where the unhandled
exception occurred back to the top of the call stack.
We can handle predictable exceptions by using the try-except-finally
paradigm:
try:
try_suite
except exception1 as variable1:
exception_suite1
...
except exceptionN as variableN:
exception_suiteN

If the statements in the try blocks suite are all executed without raising
an exception, the except blocks are skipped. If an exception is raised inside
the try block, control is immediately passed to the suite corresponding to
the first matching exception. This means that any statements in the suite
that follow the one that caused the exception will not be executed:
while 1:
try:
82 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

x = int(raw_input("Please enter a number: "))

break
except ValueError:
print "Oops! That was no valid number. Try again..."

The raise statement allows the programmer to force a specified excep-

tion to occur:
import string
import sys
try:
f = open(myfile.txt)
s = f.readline()
i = int(string.strip(s))
except IOError, (errno, strerror):
print "I/O error(%s): %s" % (errno, strerror)
except ValueError:
print "Could not convert data to an integer."
except:
print "Unexpected error:", sys.exc_info()[0]
raise

We also can use else:

for arg in sys.argv[1:]:
try:
f = open(arg, r)
except IOError:
print cannot open, arg
else:
print arg, has, len(f.readlines()), lines
f.close()

The Google Python Style Guide for Exceptions

Exceptions must be used carefully and must follow certain conditions:

? Raise exceptions like this: raise MyException(Error message) or

raise MyException. Do not use the two-argument form.

? Modules or packages should define their own domain-specific base ex-

ception class, which should be inherit from the built-in Exception
class. The base exception for a module should be called Error.
4.6. DEBUGGING AND PROFILING 83

class Error(Exception):
pass

? Never use catch-all except: statements, or catch Exception or StandardError,

unless you are re-raising the exception or in the outermost block in
your thread (and printing an error message).

? Minimize the amount of code in a try/except block. The larger the

body of the try, the more likely that an exception will be raised by
a line of code that you didnt expect to raise an exception. In those
cases, the try/except block hides a real error.

? Use the finally clause to execute code whether or not an exception

is raised in the try block. This is often useful for clean-up, i.e., closing
a file.

? When capturing an exception, use as rather than a comma. For ex-

ample:
try:
raise Error
except Error as error:
pass

4.6 Debugging and Profiling

Debugging a Code
Interactive Running:
If you have some code in a source file and you want to explore it interactively,
you can run Python with the -i switch, like this: python -i example.py.

pdb:
The debugger pdb can be used in the command line:
>>> python3 -m pdb program.py

or added as a module as the first statement of the function we want to

examine:
84 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

import pdb
pdb.set_trace()

To perform the inspection, type: s for step, p for print, and c for con-
tinue.

Profiling
If a program runs very slowly or consumes far more memory than we expect,
the problem is most often due to our choice of algorithms or data structures
or due to some inefficient implementation. Some performance verification is
useful though:

? prefer tuples to list with read-only data;

? use generators rather than large lists or tuples to iteration;

? when creating large strings out of small strings, instead of concate-

nating the small, accumulate them all in a list and join the list of
strings in the end. A good examples is given by the Google Python
Style guide:
[Good]
items = [<table>]
for last_name, first_name in employee_list:
items.append(<tr><td>%s, %s</td></tr> %
(last_name, first_name))
items.append(</table>)
employee_table = .join(items)

[Bad]
employee_table = <table>
for last_name, first_name in employee_list:
employee_table += <tr><td>%s, %s</td></tr> %
(last_name, first_name)
employee_table += </table>

The cProfile Package:

Provides a detailed breakdown of call times and can be used to find perfor-
mance bottlenecks.
4.6. DEBUGGING AND PROFILING 85

import cProfile
cProfile.run(main())

You can run it by typing:

$ python3 -m cProfile - o profile.day mymodule.py
$ python3 -m pstats

The timeit Package:

Used for timing small pieces of the code:

>>> import timeit
>>> timeit.timeit("x = 2 + 2")
0.034976959228515625
>>> timeit.timeit("x = sum(range(10))")
0.92387008666992188
> python -m timeit -s "import mymodule as m" "m.myfunction()"

The following code shows a simple example of how to time a function:

[general_problems/modules/using_time_module.py]

import time

def sumOfN2(n):
a simple example of how to time a function
start = time.time()
theSum = 0
for i in range(1,n+1):
theSum = theSum + i
end = time.time()
return theSum,end-start

if __name__ == __main__:
n = 5
print("Sum is %d and required %10.7f seconds"%sumOfN2(n))
n = 200
print("Sum is %d and required %10.7f seconds"%sumOfN2(n))
86 CHAPTER 4. PYTHONS STRUCTURE AND MODULES

4.7 Unit Testing

It is good practice to write tests for individual functions, classes, and meth-
ods, to ensure they behave to the expectations. Pythons standard library
provides two unit testing modules: doctest and unittest. There are also
third part testing tools: nose and py.test.

doctest
Use it when writing the tests inside the modules and functions docstrings.
Then just add three line in the end:
if __name__ = "__main__"
import doctest
doctest.testmod()

To run the programs doctest, there are two approaches:

? Importing the doctest module and then run the program:

$ python3 program.py -v

? Creating a separate test program using the unittest module, which

is modelled on Javas JUnit unittesting library.
import doctest
import unittest
import module_to_be_tested

suite = unittest.testsuite()
suite.addtest(doctest.doctestsuite(module_to_be_tested)
runner = unittest.testtestrunner()
print(runner.run(suite))

Test Nomenclature
Test fixtures The code necessary to set up a test (for example, creating
an input file for testing and deleting afterwards).

Test cases The basic unit of testing.

Test suites Collection of test cases, created by the subclass unittest.testcase,

where each method has a name beginning with test.
4.7. UNIT TESTING 87

Test runner An object that executes one of more test suites.

88 CHAPTER 4. PYTHONS STRUCTURE AND MODULES
Chapter 5

Object-Oriented Design

Suppose we want to define an object in Python to represent a circle. We

could remember about Pythons collections module and create a named
tuple for our circle:
>>> circle = collections.namedtuple("Circle", "x y radius")
>>> circle
<class __main__.Circle>
>>> circle = circle(13, 84, 9)
>>> circle
Circle(x=13, y=84, radius=9)

However, many things are missing here. First, there are no guarantees
that anyone who uses our circle data is not going to type an invalid input
value, such as a negative number for the radius. Second, how could we also
associate to our circle some operations that are proper from it, such as its
area or perimeter?
For the first problem, we can see that the inability to validate when cre-
ating an object is a really bad aspect of taking a purely procedural approach
in programming. Even if we decide to include many exceptions handling
the invalid inputs for our circles, we still would have a data container that
is not intrinsically made and validated for its real purpose. Imagine now if
we had chosen a list instead of the named tuple, how would we handle the
fact that lists have sorting properties?
It is clear from the example above that we need to find a way to create
an object that has only the proprieties that we expect it to have. In other
words, we want to find a way to package data and restrict its methods. That
is what object-oriented programming allows you to do: to create your own

89
90 CHAPTER 5. OBJECT-ORIENTED DESIGN

custom data type, which in this example would be a circle class.

5.1 Classes and Objects

Classes are the way we can gather special predefined data and methods to-
gether. We use them by creating objects, which are instances of a particular
class. The simplest form of a class in Python looks like the following snippet:
class ClassName:
<statement-1>
.g
<statement-N>

>>> x = ClassName() # class instantiation

Class Instantiation
Class instantiation uses function notation to create objects in a known ini-
tial state. The instantiation operation creates an empty object which has
individuality. However, multiple names (in multiple scopes) can be bound
to the same object (also know as aliasing). In Python, when an object is
created, first the special method new () is called (the constructor) and
then init () initializes it.

Attributes
Objects have the attributes from their Classes, which are methods and data.
Method attributes are functions whose first argument is the instance on
which it is called to operate (which in Python is conventionally called self).
Attributes are any name following a dot. References to names in modules
are attribute references: in the expression modname.funcname, modname is
a module object and funcname is one of its attribute. Attributes may be
read-only or writeable. Writeable attributes may be deleted with the del
statement.

Namespaces
A namespace is a mapping from names to objects. Most namespaces are
currently implemented as Python dictionaries. Examples of namespaces
5.2. PRINCIPLES OF OOP 91

are: the set of built-in names, the global names in a module, and the local
names in a function invocation. The statements executed by the top-level
invocation of the interpreter, either reading from a script file or interactively,
are considered part of a module called main , so they have their own global
namespace.

Scope
A scope is a textual region of a Python program where a namespace is
directly accessible. Although scopes are determined statically, they are used
dynamically. Scopes are determined textually: the global scope of a function
defined in a module is that modules namespace. When a class definition is
entered, a new namespace is created, and used as the local scope.

5.2 Principles of OOP

Specialization
Specialization (or inheritance) is the procedure of creating a new class that
inherits all the attributes from the super class (also called base class). Any
method can be overridden (reimplemented) in a subclass (in Python, all
methods are virtual). Inheritance is described as an is-a relationship.
Furthermore Google Python Style Guide advices that if a class inherits
from no other base classes, we should explicitly inherit it from Pythons
highest class, object:
class OuterClass(object):
class InnerClass(object):

Polymorphism
Polymorphism (or dynamic method binding) is the principle where methods
can be redefined inside subclasses. In other words, if we have an object
of a subclass and we call a method that is also defined in the superclass,
Python will use the method defined in the subclass. If, for instance, we need
to recover the superclasss method, we can easily call it using the built-in
super().
For example, all instances of a custom class are hashable by default
in Python. This means that the hash() attribute can be called, allowing
92 CHAPTER 5. OBJECT-ORIENTED DESIGN

them to be used as dictionary keys and to be stored in sets. However, if

we reimplement the attribute eq (), we change this propriety (what can
result on our instances no longer being hashable).

Aggregation
Aggregation (or composition) defines the process where a class includes one
of more instance variables that are from other classes. It is a has-a relation-
ship. In Python, every class uses inheritance (they are all custom classes
from the object base class), and most use aggregation since most classes
have instance variables of various types.

A first Example of a Class

We now have the tools for writing our first class in Python. The example
bellow illustrate how we could write a circle data container1 . using the
object-oriented design paradigm. First, we created a class called Point with
general data and methods attributes. Then we use inheritance to create a
Circle subclass from it:
[general_problems/oop/ShapeClass.py]

import math

class Point:
def __init__(self, x = 0, y = 0):
self.x = x # data attribute
self.y = y

def distance_from_origin(self): # method attribute

return math.hypot(self.x, self.y)

def __eq__(self, other):

return self.x == other.x and self.y == other.y

def __repr__(self):
return "point ({0.x!r}, {0.y!r})".format(self)

def __str__(self):
return "({0.x!r}, {0.y!r})".format(self)

1
containers, which is a generic data structure that permits storage and retrieval of
data items independent of content.
5.2. PRINCIPLES OF OOP 93

class Circle(Point):

def __init__(self, radius, x=0, y=0):

super().__init__(x,y) # creates/initializes
self.radius = radius

def edge_distance_from_origin(self):
return abs(self.distance_from_origin() - self.radius)

def area(self):
return math.pi*(self.radius**2)

def circumference(self):
return 2*math.pi*self.radius

def __eq__(self, other): # avoid infinite recursion

return self.radius == other.radius and
super().__eq__(other)

def __repr__(self):
return "circle ({0.radius!r}, {0.x!r})".format(self)

def __str__(self):
return repr(self)

>>> import ShapeClass as shape

>>> a = shape.Point(3,4)
>>> a
point (3, 4)
>>> repr(a)
point (3, 4)
>>> str(a)
(3, 4)
>>> a.distance_from_origin()
5.0
>>> c = shape.Circle(3,2,1)
>>> c
circle (3, 2)
>>> repr(c)
circle (3, 2)
>>> str(c)
circle (3, 2)
>>> c.circumference()
18.84955592153876
>>> c. edge_distance_from_origin()
94 CHAPTER 5. OBJECT-ORIENTED DESIGN

0.7639320225002102

5.3 Python Design Patterns

Design patterns are an attempt to bring a formal definition for correctly
designed structures to software engineering. There are many different design
patterns to solve different general problems.

Decorator Pattern
Decorators (also know as the @ notation) are a tool to elegantly specify some
transformation on functions and methods. The decorator pattern allows us
to wrap an object that provides core functionality with other objects that
alter that functionality. For example, the snippet bellow was copied from
the Google Python Style guide:
class C(object):
def method(self):
method = my_decorator(method)

can be written as
class C(object):
@my_decorator
def method(self):

A full example of a decorator, for a custom benchmarking function, is

shown bellow, :
[general_problems/oop/do_benchmark.py]

import random

def benchmark(func):
import time
def wrapper(*args, **kwargs):
t = time.clock()
res = func(*args, **kwargs)
print("\t%s" % func.__name__, time.clock()-t)
return res
return wrapper
5.3. PYTHON DESIGN PATTERNS 95

@benchmark
def random_tree(n):
temp = [n for n in range(n)]
for i in range(n+1):
temp[random.choice(temp)] = random.choice(temp)
return temp

if __name__ == __main__:
random_tree(10000)

"""
python3 do_benchmark.py
random_tree 0.04999999999999999
"""

The most common decorators are @classmethod and @staticmethod,

for converting ordinary methods to class or static methods.

Observer Pattern
The observer pattern is useful when we want to have a core object that
maintains certain values, and then having some observers to create serialized
copies of that object. This can be implemented by using the @properties
decorator, placed before our functions (before def). This will control at-
tribute access, for example, to make an attribute to be read-only. Properties
are used for accessing or setting data instead of simple accessors or setters:
@property
def radius(self):
return self.__radius

Singleton Pattern
A class follows the singleton pattern if it allows exactly one instance of a
certain object to exist. Since Python does not have private constructors,
we use the new class method to ensure that only one instance is ever
created. When we override it, we first check whether our singleton instance
was created. If not, we create it using a super class call:
>>> class SinEx:
96 CHAPTER 5. OBJECT-ORIENTED DESIGN

... _sing = None

... def __new__(self, *args, **kwargs):
... if not self._sing:
... self._sing = super(SinEx,
self).__new__(self, *args, **kwargs)
... return self._sing

>>> x = SinEx()
>>> x
<__main__.SinEx object at 0xb72d680c>
>>> y = SinEx()
>>> x == y
True
>>> y
<__main__.SinEx object at 0xb72d680c>

The two objects are equal and are in the same address, so they are the
same object.

5.4 Additional Exercises

A Hash Table Class
The example bellow shows how to create a map abstract data with two lists:
[general_problems/oop/Hash_Table.py]

class HashTable:
def __init__(self):
self.size = 11
self.slots = [None] * self.size
self.data = [None] * self.size

def put(self,key,data):
hashvalue = self.hashfunction(key,len(self.slots))

if self.slots[hashvalue] == None:
self.slots[hashvalue] = key
self.data[hashvalue] = data
else:
if self.slots[hashvalue] == key:
self.data[hashvalue] = data
else:
5.4. ADDITIONAL EXERCISES 97

nextslot = self.rehash(hashvalue,len(self.slots))
while self.slots[nextslot] != None and \
self.slots[nextslot] != key:
nextslot = self.rehash(nextslot,len(self.slots))

if self.slots[nextslot] == None:
self.slots[nextslot]=key
self.data[nextslot]=data
else:
self.data[nextslot] = data

def hashfunction(self,key,size):
return key%size

def rehash(self,oldhash,size):
return (oldhash+1)%size

def get(self,key):
startslot = self.hashfunction(key,len(self.slots))

data = None
stop = False
found = False
position = startslot
while self.slots[position] != None and \
not found and not stop:
if self.slots[position] == key:
found = True
data = self.data[position]
else:
position=self.rehash(position,len(self.slots))
if position == startslot:
stop = True
return data

def __getitem__(self,key):
return self.get(key)

def __setitem__(self,key,data):
self.put(key,data)

def test_HashTable(module_name=this module):

98 CHAPTER 5. OBJECT-ORIENTED DESIGN

H = HashTable()
H[54]="buffy"
H[26]="xander"
H[17]="giles"
print(H.slots)
print(H.data)
s = Tests in {name} have {con}!
print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_HashTable()
 Part II

Algorithms are Fun

99
Chapter 6

Additional Abstract Data

Structures

An abstract data type (ADT) is a mathematical model for a certain class

of data structures that have similar behavior. Different classes of abstract
data types have many different but functionally equivalent data structures
that implement them.
Data structures can be classified as either contiguous or linked, depend-
ing whether they are based on arrays or pointers. In Python, for instance,
contiguously-allocated structures (composed of single slabs of memory) in-
clude strings, lists, tuples, and dictionaries. In the following sections we will
see examples of some more specialised continuous structures and examples
of some linked data structures (distinct chunks of memory bound together
by pointers).

6.1 Stacks
A stack is a linear data structure that can be accessed only at one of its
ends (which we will refers as the top) for either storing or retrieving. In
other words, array access of elements in a stack is restricted and they are
an example of a last-in-first-out (LIFO) structure. You can think of a stack
as a huge pile of books on your desk. Stacks need to have the following
operations running at O(1):

push Insert an item at the top of the stack.

pop Remove an item from the top of the stack.

101
102 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

top/peek Look up the element on the top.

empty/size Check whether the stack is empty or return its size.

Stacks in Python can be implemented with lists and the methods append()
and pop() (without an explicit index):
[adt/stacks/stack.py]

class Stack(list):
define the stack class
def __init__(self):
self.items = []

def isEmpty(self):
return self.items == []

def push(self, items):

self.items.append(items)

def pop(self):
if not self.isEmpty():
return self.items.pop()
else:
raise Exception(Stack is empty!)

def peek(self):
return self.items[-1]

def size(self):
return len(self.items)

def main():
stack = Stack()
stack.push(1)
stack.push(2)
stack.push(3)
print(stack.size())
print(stack.peek())
print(stack.pop())
print(stack.peek())

if __name__ == __main__:
main()
6.1. STACKS 103

Another approach we can use to implement a stack (or any abstract

structure) is by thinking of it as a container of nodes (objects) following a
LIFO order:1
class Node(object):
def __init__(self, value=None):
self.value = value
self.next = None

and then building the stack as a collection of nodes:

[adt/stacks/linked_stack.py]

class Node(object):
def __init__(self, value=None):
self.value = value
self.next = None

class StackwithNodes(object):
Define a Stack with nodes
def __init__(self):
self.top = None

def isEmpty(self):
return bool(self.top)

def pop(self):
node = self.top
if node:
self.top = node.next
return node.value
else:
raise Exception(Stack is empty.)

def push(self, value):

node = Node(value)
node.next = self.top
self.top = node

def size(self):
node = self.top
if node not None: num_nodes = 1
else: return 0

1
We will use similar a Node Class in many examples in the rest of these notes.
104 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

node = node.next
while node:
num_nodes += 1
node = node.next
return num_nodes

def peek(self):
return self.top.value

def main():
stack = StackwithNodes()
stack.push(1)
stack.push(2)
stack.push(3)
print(stack.size())
print(stack.peek())
print(stack.pop())
print(stack.peek())

if __name__ == __main__:
main()

Stacks are suitable for depth-first traversal algorithms in graphs, as we

will see in future chapters.

6.2 Queues
A queue, differently of a stack, is a structure where the first enqueued ele-
ment (at the back) will be the first one to be dequeued (when it is at the
front), i.e., a queue is a first-in-first-out (FIFO) structure. You can think of
a queue as a line of people waiting for a roller-coaster ride. Array access of
elements in queues is also restricted and queues should have the following
operations running at O(1):
enqueue Insert an item at the back of the queue.
dequeue Remove an item from the front of the queue.
peek/front Retrieve an item at the front of the queue without removing
it.
empty/size Check whether the queue is empty or give its size.
6.2. QUEUES 105

The example bellow shows a class for a queue in Python:

[adt/queues/queue.py]

class Queue(object):
a class for a queue
def __init__(self):
self.items = []

def isEmpty(self):
return self.items == []

def enqueue(self, item):

self.items.insert(0, item)

def dequeue(self):
return self.items.pop()

def size(self):
return len(self.items)

def peek(self):
if not self.isEmpty():
return self.items[-1]
else:
raise Exception(Queue is empty.)

def size(self):
return len(self.items)

def main():
queue = Queue()
queue.enqueue(1)
queue.enqueue(2)
queue.enqueue(3)
print(queue.size())
print(queue.peek())
print(queue.dequeue())
print(queue.peek())

if __name__ == __main__:
main()
106 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

However, we have learned that the method insert() for lists in Python
is very inefficient (remember, lists only work on O(1) when we append or
pop at/from their end, because otherwise all of the other elements would
have to be shifted in memory). We can be smarter than that and write an
efficient queue using two stacks (two lists) instead of one:
[adt/queues/queue_from_two_stacks.py]

class Queue(object):
an example of a queue implemented from 2 stacks
def __init__(self):
self.in_stack = []
self.out_stack = []

def enqueue(self, item):

return self.in_stack.append(item)

def dequeue(self):
if self.out_stack:
return self.out_stack.pop()
while self.in_stack:
self.out_stack.append(self.in_stack.pop())
if not self.out_stack:
raise Exception("Queue empty!")
return self.out_stack.pop()

def size(self):
return len(self.in_stack) + len(self.out_stack)

def peek(self):
if self.out_stack:
return self.out_stack[-1]
while self.in_stack:
self.out_stack.append(self.in_stack.pop())
if self.out_stack:
return self.out_stack[-1]
else:
return None

def main():
queue = Queue()
queue.enqueue(1)
queue.enqueue(2)
queue.enqueue(3)
6.2. QUEUES 107

print(queue.size())
print(queue.peek())
print(queue.dequeue())
print(queue.peek())

if __name__ == __main__:
main()

Another approach is to implement a queue as a container for nodes, as

we have done for stacks, but now the nodes are inserted and removed in a
FIFO order:
[adt/queues/linked_queue.py]

class Node(object):
def __init__(self, value):
self.value = value
self.next = None

class LinkedQueue(object):
Queue acts as a container for nodes (objects) that are
inserted and removed according FIFO
def __init__(self):
self.front = None
self.back = None

def isEmpty(self):
return bool(self.front) and bool(self.back)

def dequeue(self):
if self.front:
value = self.front.value
self.front = self.front.next
return value
raise Exception(Queue is empty, cannot dequeue.)

def enqueue(self, value):

node = Node(value)
if self.front:
self.back.next = node
else:
self.front = node
self.back = node
return True
108 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

def size(self):
node = self.front
if node:
num_nodes = 1
node = node.next
while node:
num_nodes += 1
node = node.next
return num_nodes

def peek(self):
return self.front.value

def main():
queue = LinkedQueue()
queue.enqueue(1)
queue.enqueue(2)
queue.enqueue(3)
print(queue.size())
print(queue.peek())
print(queue.dequeue())
print(queue.peek())

if __name__ == __main__:
main()

Queues are necessary for breath-first traversal algorithms for graphs, as

we will see in future chapters.

6.3 Deques
A deque is a double-ended queue, which can roughly be seen as an union of
a stack and a queue:
[adt/queues/dequeue.py]

class Deque(object):
a class for a double ended queue
def __init__(self):
self.items = []

def isEmpty(self):
6.3. DEQUES 109

return self.items == []

def addFront(self, item):

self.items.append(item)

def addRear(self, item):

self.items.insert(0, item)

def removeFront(self):
return self.items.pop()

def removeRear(self):
return self.items.pop(0)

def size(self):
return len(self.items)

def __repr__(self):
return {}.format(self.items)

def main():
dq = Deque()
dq.addFront(1)
dq.addFront(2)
dq.addFront(3)
dq.addRear(40)
dq.addRear(50)
print(dq.size())
print(dq)

if __name__ == __main__:
main()

We see again the problem of inserting/removing items in Pythons lists

in any positions that are not the end. The good news is that Python is
your friend and its collections.deque gives us an efficient deque, with
fast appending and popping from both ends:
>>> from collections import deque
>>> q = deque(["buffy", "xander", "willow"])
>>> q
deque([buffy, xander, willow])
>>> q.append("giles")
>>> q
110 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

deque([buffy, xander, willow, giles])

>>> q.popleft()
buffy
>>> q.pop()
giles
>>> q
deque([xander, willow])
>>> q.appendleft(angel)
>>> q
deque([angel, xander, willow])

Note that we can also specify the size of our deque. For example, we
could have written q = deque(maxlen = 4) in the example above. Another
interesting method for deques is rotate(n), which rotated the deque n steps
to the right or, if n is negative, to the left.
Interestingly, deques in Python are based on a doubly linked list,2 not in
dynamic arrays. It means that operations such as inserting an item anywhere
are fast (O(1)), but arbitrary index accessing can be slow (O(n)).

6.4 Priority Queues and Heaps

A priority queue is an abstract data type which is similar to a regular queue
or stack, but where each element has a priority associated with it. If two
elements have the same priority, they are served according to their order in
the queue.
A sensible implementation of a priority queue is given by a heap data
structure and we will use it for most of our examples.

Heaps
Conceptually, a heap is a binary tree where each node is smaller (larger) than
its children. We will learn about trees in the next chapters but we should
already keep in mind that when modifications are made, in a balanced tree,
we can repair its structure with O(logn) runtimes. Heaps are generally useful
for applications that repeatedly access the smallest (largest) element in the
list. Moreover min-(max-)heap will let you to find the smallest (largest)
element in O(1) and to extract/add/replace it in O(ln n).
2
Linked lists are another abstract data structure that we will learn about at the end
of this chapter. Doubly here means that their nodes have links to the next and to the
previous node.
6.4. PRIORITY QUEUES AND HEAPS 111

Pythons heapq Package

A very efficient heap implementation in Python is given by the heapq mod-
ule, which provides functions to insert and remove items while keeping the
sequence as a heap.
We can use the heapq.heapify(x) method to transform a list into a
heap, in-place, and in O(n) time:
>>> list1 = [4, 6, 8, 1]
>>> heapq.heapify(list1)
>>> list1
[1, 4, 8, 6]

Once we have a heap, the heapq.heappush(heap, item) method is used

to push the item onto it:
>>> import heapq
>>> h = []
>>> heapq.heappush(h, (1, food))
>>> heapq.heappush(h, (2, have fun))
>>> heapq.heappush(h, (3, work))
>>> heapq.heappush(h, (4, study))
>>> h
[(1, food), (2, have fun), (3, work), (4, study)]

The method heapq.heappop(heap) is used to pop and return the small-

est item from the heap:
>>> list1
[1, 4, 8, 6]
>>> heapq.heappop(list1)
1
>>> list1
[4, 6, 8]

The method heapq.heappushpop(heap, item) is used to push the item

on the heap, then it pops and returns the smallest item from the heap. In
a similar way, heapq.heapreplace(heap, item) will pop and return the
smallest item from the heap, and then push the new item. The heap size
does not change in any of these methods and they are more efficient than
using each method separately.
In addition, many operations can be made using the heaps propriety.
For example heapq.merge(*iterables) will merge multiple sorted inputs
112 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

into a single sorted output (returning a iterator):

>>> for x in heapq.merge([1,3,5],[2,4,6]):
... print(x,end="\n")
...
1
2
3
4
5
6

The methods heapq.nlargest(n, iterable[, key]) and heapq.nsmallest(n,

iterable[, key]) will return a list with the n largest and smallest elements
from the dataset defined by iterable.

A Class for a Heap

If we want to write ourselves a class for a heap, the fist thing to do is defining
a method to create the heap propriety (heapify). The following code does
that, implementing a max-heap for a given list:
[heap/heapify.py]

class Heapify(object):
def __init__(self, data=None):
self.data = data or []
for i in range(len(data)//2, -1, -1):
self.__max_heapify__(i)

def __repr__(self):
return {}.format(self.data)

def parent(self, i):

return i >> 1

def left_child(self, i):

return (i << 1) + 1

def right_child(self, i):

return (i << 1) + 2 # +2 instead of +1 because its
0-indexed.
6.4. PRIORITY QUEUES AND HEAPS 113

def __max_heapify__(self, i):

largest = i
left = self.left_child(i)
right = self.right_child(i)
n = len(self.data)
largest = (left < n and self.data[left] > self.data[i]) and
left or i
largest = (right < n and self.data[right] >
self.data[largest]) and right or largest
if i is not largest:
self.data[i], self.data[largest] = self.data[largest],
self.data[i]
self.__max_heapify__(largest)

def extract_max(self):
n = len(self.data)
max_element = self.data[0]
self.data[0] = self.data[n - 1]
self.data = self.data[:n - 1]
self.__max_heapify__(0)
return max_element

def test_Heapify():
l1 = [3, 2, 5, 1, 7, 8, 2]
h = Heapify(l1)
assert(h.extract_max() == 8)
print ("Tests Passed!")

if __name__ == __main__:
test_Heapify()

A Class for a Priority Queue

To conclude this section, the following example shows how to use the heapq
package to implement a priority queue class:
[heap/PriorityQueueClass.py]

import heapq

class PriorityQueue(object):
implements a priority queue class
114 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

def __init__(self):
self._queue = []
self._index = 0 # comparying same priority level

def push(self, item, priority):

heapq.heappush(self._queue, (-priority, self._index, item))
self._index += 1

def pop(self):
return heapq.heappop(self._queue)[-1]

class Item:
def __init__(self, name):
self.name = name
def __repr__(self):
return "Item({!r})".format(self.name)

def test_PriorityQueue():
push and pop are all O(logN)
q = PriorityQueue()
q.push(Item(test1), 1)
q.push(Item(test2), 4)
q.push(Item(test3), 3)
assert(str(q.pop()) == "Item(test2)")
print(Tests passed!.center(20,*))

if __name__ == __main__:
test_PriorityQueue()

6.5 Linked Lists

A linked list is like a stack (elements added to the head) or a queue (elements
added to the tail), except that we can peek any element in the structure on
O(1), not only the elements at the ends. In general, a linked list is simply
a linear list of nodes containing a value and a pointer (a reference) to the
next node (except for the last), where the reference points to None:
>>> class Node(object):
... def __init__(self, value=None, next=None):
... self.value = value
... self. next = next
6.5. LINKED LISTS 115

>>> L = Node("a", Node("b", Node("c", Node("d"))))

>>> L.next.next.value
c

We can adapt this node class accept some get and set methods:
class Node(object):
def __init__(self, value):
self.value = value
self.next = None

def getData(self):
return self.value

def getNext(self):
return self.next

def setData(self, newdata):

self.value = newdata

def setNext(self, newnext):

self.next = newnext

We obtain a LIFO linked list as a collection of these nodes:

[adt/linked_lists/likedlist_lifo.py]

class Node(object):
def __init__(self, value = None, next = None):
self.value = value
self.next = next

class LinkList(object):
def __init__(self):
self.head = None
self.lenght = 0

def addNode(self, value):

temp = self.head
node = Node(value)
node.next = temp
self.head = node
self.lenght += 1

def printList(self):
116 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

node = self.head
while node:
print(node.value)
node = node.next

def deleteNode(self, index):

prev = None
node = self.head
i = 0
while node and i<index:
prev = node
node = node.next
i += 1
if index == i:
self.lenght -= 1
if prev == None:
self.head = node.next
else:
prev.next = node.next
else:
print(Index not found)

def main():
ll = LinkList()
print(ll.lenght)
ll.addNode(1)
ll.addNode(2)
ll.addNode(3)
print(ll.lenght)
ll.printList()
ll.deleteNode(4)
ll.printList()
print(ll.lenght)

if __name__ == __main__:
main()

We can also write a class for FIFO linked list:

[adt/linked_lists/likedlist_fifo.py]

class Node(object):
def __init__(self, value = None, next = None):
self.value = value
6.5. LINKED LISTS 117

self.next = next

class LinkList(object):
def __init__(self):
self.head = None
self.tail = None
self.length = 0

def addNode(self, value):

node = Node(value)
if not self.head:
self.head = node
if self.tail:
self.tail.next = node
self.tail = node
self.length += 1

def printList(self):
node = self.head
while node:
print(node.value)
node = node.next

def deleteNode(self, index):

prev = None
node = self.head
i = 0
while node and i < index:
prev = node
node = node.next
i += 1
if i == index:
if not prev:
self.head = node.next
else:
prev.next = node.next
self.length -= 1
else:
print(Index not found!)

def removeDupl(self):
prev = None
node = self.head
aux_dict = Counter()
while node:
118 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

value_here = node.value
if aux_dict[value_here] == 0:
aux_dict[value_here] = 1
else:
if prev == None:
self.head = node.next
else:
prev.next = node.next
self.length -= 1
prev = node
node = node.next

def removeDupl_no_buf(self):
node = self.head
while node:
pivot = node.value
pointer = node.next
prev = node
while pointer:
value_here = pointer.value
if value_here == pivot:
prev.next = pointer.next
self.length -= 1
prev = pointer
pointer = pointer.next
node = node.next

from collections import Counter

def main():

ll = LinkList()
for i in range(1, 10):
ll.addNode(i)
ll.addNode(i+1)
print(Linked List with duplicates:)
ll.printList()
print(Length before deleting duplicate is: , ll.length)
ll.removeDupl()
ll.printList()
print(Lenght after deleting duplicates is: , ll.length)

ll = LinkList()
for i in range(1, 10):
6.5. LINKED LISTS 119

ll.addNode(i)
ll.addNode(i+1)
print(Linked List with duplicates:)
ll.printList()
print(Length before deleting duplicate is: , ll.length)
ll.removeDupl_no_buf()
ll.printList()
print(Lenght after deleting duplicates is: , ll.length)

if __name__ == __main__:
main()

Linked lists have a dynamic size at runtime and they are good for when
you have an unknown number of items to store. Insertion is O(1) but dele-
tion and searching can be O(n) because locating an element in a linked list
is slow and is it done by a sequential search. Traversing backward or sort-
ing a linked list are even worse, being both O(n2 ). A good trick to obtain
deletion of a node i at O(1) is copying the data from i + 1 to i and then to
deleting the node i + 1.
120 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

6.6 Additional Exercises

Stacks
Stacks are very useful when data has to be sorted and retrieved in reverse
order. In the example bellow we use our Stack class to reverse a string:
[adt/stacks/reverse_string_with_stack.py]

import sys
import stack

def reverse_string_with_stack(str1):
s = stack.Stack()
revStr =
for c in str1:
s.push(c)
while not s.isEmpty():
revStr += s.pop()
return revStr

def test_reverse_string_with_stack():
str1 = Buffy is a Slayer!
assert(reverse_string_with_stack(str1) == !reyalS a si yffuB)
print(Tests passed!)

if __name__ == __main__:
test_reverse_string_with_stack()

The following example uses a stack to balance parenthesis in a string:

[adt/stacks/balance_parenthesis_str_stack.py]

from stack import Stack

def balance_par_str_with_stack(symbolString):
s = Stack()
balanced = True
index = 0
while index < len(symbolString) and balanced:
symbol = symbolString[index]
if symbol == "(":
s.push(symbol)
else:
6.6. ADDITIONAL EXERCISES 121

if s.isEmpty():
balanced = False
else:
s.pop()
index = index + 1

if balanced and s.isEmpty():

return True
else:
return False

def test_balance_par_str_with_stack(module_name=this module):

print(balance_par_str_with_stack(((()))))
print(balance_par_str_with_stack((()))
s = Tests in {name} have {con}!
print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_balance_par_str_with_stack()

The following example uses a stack to transform a decimal number to

binary number:
[atf/stacks/dec2bin_with_stack.py]

from stack import Stack

def dec2bin_with_stack(decnum):
s = Stack()
str_aux =
while decnum > 0:
dig = decnum % 2
decnum = decnum//2
s.push(dig)
while not s.isEmpty():
str_aux += str(s.pop())
return str_aux

def test_dec2bin_with_stack(module_name=this module):

decnum = 9
assert(dec2bin_with_stack(decnum) == 1001)
s = Tests in {name} have {con}!
122 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_dec2bin_with_stack()

The following example implements a stack that has O(1) minimum lookup:
[adt/stacks/stack_with_min.py]

class Stack(list):
def push(self, value):
if len(self) > 0:
last = self[-1]
minimum = self._find_minimum(value, last)
else:
minimum = value
self.minimum = minimum
self.append(NodeWithMin(value, minimum))

def _find_minimum(self, value, last_value):

if value < last_value.minimum:
return value
return last_value.minimum

def min(self):
return self.minimum

class NodeWithMin(object):
def __init__(self, value, minimum):
self.value = value
self.minimum = minimum

def __repr__(self):
return str(self.value)

def min(self):
return self.minimum

def main():
stack = Stack()
stack.push(1)
stack.push(2)
stack.push(3)
node = stack.pop()
6.6. ADDITIONAL EXERCISES 123

print(node.minimum)
stack.push(0)
stack.push(4)
node = stack.pop()
print(node.min())
print(stack.min())
print(stack)

if __name__ == __main__:
main()

The following example implements a set of stacks, composed of several

stacks. It creates a new stack when the previous stack exceeds capacity.
The push and pop methods are identical to a single stack:
[adt/stacks/set_of_stacks.py]

class SetOfStacks(list):
def __init__(self, capacity=4):
self.stacks = []
self.last_stack = []
self.capacity = capacity
self.stacks.append(self.last_stack)

def __repr__(self):
return str(self.stacks)

def push(self, value):

last_stack = self.last_stack
if len(last_stack) is self.capacity:
last_stack = []
self.last_stack = last_stack
self.stacks.append(last_stack)
last_stack.append(value)

def pop(self):
last_stack = self.last_stack
value = last_stack.pop()
if len(last_stack) is 0:
self.stacks.pop()
self.last_stack = self.stacks[-1]
return value
124 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

def main():
stack = SetOfStacks()
stack.push(1)
stack.push(2)
stack.push(3)
stack.push(4)
stack.push(5)
stack.push(6)
print(stack)
stack.pop()
stack.pop()
stack.pop()
print(stack)

if __name__ == __main__:
main()

Queues
The example bellow uses the concepts of a queue to rotate an array from
right to left for a given number n:3
[adt/queues/rotating_array.py]

def rotating_array(seq, n):

myqueue = []
for i in range(n):
myqueue.append(seq.pop())
myqueue.reverse()
return myqueue + seq

def test_rotating_array(module_name=this module):

seq = [1, 2, 3, 4, 5, 6, 7]
n = 4
assert(rotating_array(seq, N) == [4, 5, 6, 7, 1, 2, 3])

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
3
We could get the same effect using collections.deque with the method rotate(n).
6.6. ADDITIONAL EXERCISES 125

test_rotating_array()

Deques
A nice application for a double-ended queue is verifying whether a string is
a palindrome:
[adt/queues/palindrome_checker_with_deque.py]

import sys
import string
import collections

def palindrome_checker_with_deque(str1):
d = collections.deque()
eq = True
strip = string.whitespace + string.punctuation + "\""
for s in str1.lower():
if s not in strip: d.append(s)
while len(d) > 1 and eq:
first = d.pop()
last = d.popleft()
if first != last:
eq = False
return eq

def test_palindrome_checker_with_deque():
str1 = Madam Im Adam
str2 = Buffy is a Slayer
assert(palindrome_checker_with_deque(str1) == True)
assert(palindrome_checker_with_deque(str2) == False)
print(Tests passed!)

if __name__ == __main__:
test_palindrome_checker_with_deque()

Priority Queues and Heaps

The example bellow uses Pythons heapq package to find the N largest and
smallest items in a sequence:
126 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

[adt/heap/find_N_largest_smallest_items_seq.py]

import heapq

def find_N_largest_items_seq(seq, N):

return heapq.nlargest(N,seq)

def find_N_smallest_items_seq(seq, N):

return heapq.nsmallest(N, seq)

def find_smallest_items_seq_heap(seq):
find the smallest items in a sequence using heapify first
heap[0] is always the smallest item
heapq.heapify(seq)
return heapq.heappop(seq)

def find_smallest_items_seq(seq):
if it is only one item, min() is faster
return min(seq)

def find_N_smallest_items_seq_sorted(seq, N):

N ~ len(seq), better sort instead
return sorted(seq)[:N]

def find_N_largest_items_seq_sorted(seq, N):

N ~ len(seq), better sort instead
return sorted(seq)[len(seq)-N:]

def test_find_N_largest_smallest_items_seq(module_name=this
module):
seq = [1, 3, 2, 8, 6, 10, 9]
N = 3
assert(find_N_largest_items_seq(seq, N) == [10, 9, 8])
assert(find_N_largest_items_seq_sorted(seq, N) == [8, 9, 10])
assert(find_N_smallest_items_seq(seq, N) == [1,2,3])
assert(find_N_smallest_items_seq_sorted(seq, N) == [1,2,3])
assert(find_smallest_items_seq(seq) == 1)
assert(find_smallest_items_seq_heap(seq) == 1)

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
6.6. ADDITIONAL EXERCISES 127

test_find_N_largest_smallest_items_seq()

The following example uses Pythons heapq package to merge a two

sorted sequences with little overhead:4
[adt/heap/merge_sorted_seqs.py]

import heapq

def merge_sorted_seqs(seq1, seq2):

result = []
for c in heapq.merge(seq1, seq2):
result.append(c)
return result

def test_merge_sorted_seq(module_name=this module):

seq1 = [1, 2, 3, 8, 9, 10]
seq2 = [2, 3, 4, 5, 6, 7, 9]
seq3 = seq1 + seq2
assert(merge_sorted_seq(seq1, seq2) == sorted(seq3))

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_merge_sorted_seq()

Linked List
The following example implements a linked list class from stack methods:
[adt/linked_lists/linked_list_from_stack.py]

class Node(object):
def __init__(self,data=None,next=None):
self.data = data
self.next = next

def setnext(self,next):
self.next = next
4
Note that the result would not be sorted if we just added both lists.
128 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

def __str__(self):
return "%s" % self.data

class LinkedListStack(object):
def __init__(self, max=0):
self.max = max
self.head = None
self.z = None
self.size = 0

def push(self, data):

if self.size == 0:
self.head = Node(data)
self.size += 1
else:
head = self.head
node = Node(data = data)
self.head = node
node.setnext(head)

def pop(self):
node = self.head.next
self.head = node

def isEmpty(self):
return self.size == 0

def __str__(self):
d = ""
if self.isEmpty(): return ""
else:
temp = self.head
d += "%s\n" % temp
while temp.next != None:
temp = temp.next
d += "%s\n" % temp
return d

def test_ll_from_stack():
ll = LinkedListStack(max = 20)
ll.push("1")
ll.push("2")
ll.push("3")
ll.push("4")
6.6. ADDITIONAL EXERCISES 129

print(ll)
ll.pop()
print(ll)

if __name__ == __main__:
test_ll_from_stack()

The following snippet shows an example of an ordered linked list. In this

case, the list is a collection of items where each item holds a relative position
that is based upon some underlying characteristic of the item. The ordering
is typically either ascending or descending and we assume that list items
have a meaningful comparison operation that is already defined. Many of
the ordered list operations are the same as those from the unordered list:
[adt/linked_lists/ordered_list.py]

from Node import Node

class OrderedList(object):
def __init__(self):
self.head = None

def add(self,item):
this method is different from linked list
current = self.head
previous = None
stop = False
while current != None and not stop:
if current.getData() > item:
stop = True
else:
previous = current
current = current.getNext()
temp = Node(item)
if previous == None:
temp.setNext(self.head)
self.head = temp
else:
temp.setNext(current)
previous.setNext(temp)

def length(self):
current = self.head
130 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES

count = 0
while current != None:
count = count + 1
current = current.getNext()
return count

def search(self,item):
this method is different from linked list
current = self.head
found = False
stop = False
while current != None and not found and not stop:
if current.getData() == item:
found = True
else:
if current.getData() > item:
stop = True
else:
current = current.getNext()
return found

def remove(self,item):
current = self.head
previous = None
found = False
while not found:
if current.getData() == item:
found = True
else:
previous = current
current = current.getNext()
if previous == None:
self.head = current.getNext()
else:
previous.setNext(current.getNext())

def test_OrderedList(module_name=this module):

olist = OrderedList()
olist.add(31)
olist.add(22)
olist.add(10)
assert(olist.search(22) == True)
olist.remove(22)
assert(olist.search(22) == False)
s = Tests in {name} have {con}!
6.6. ADDITIONAL EXERCISES 131

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_OrderedList()
132 CHAPTER 6. ADDITIONAL ABSTRACT DATA STRUCTURES
Chapter 7

Asymptotic Analysis

Asymptotic analysis is a method to describe the limiting behavior and the

performance of algorithms when applied to very large input datasets. To
understand why asymptotic analysis is important, suppose you have to sort
a billion of numbers (n = 109 )1 in a common desktop computer. Suppose
that this computer has a CPU clock time of 1 GHz, which roughly means
that it executes 109 processor cycles (or operations) per second.2 Then, for
an algorithm that has a runtime of O(n2 ), it would take approximately one
billion of seconds to finish the sorting (in the worst case) which means one
entire year!
Another way of visualizing the importance of asymptotic analysis is di-
rectly looking to the functions behaviour. In the Fig. 7 we have many
classes of functions plotted together and it is clear that when n increases,
the number of operations for any polynomial or exponential algorithm is
infeasible.

7.1 Complexity Classes

A complexity class is a set of problems with related complexity. A reduction
is a transformation of one problem into another problem which is at least
as difficult as the original problem. The most commonly used reduction
1
Remember that for memory gigabytes means 10243 = 230 bytes and for storage it
means 10003 = 109 bytes. Also, integers usually take 2 or 4 bytes each. However, for this
example we are simplifying this by saying that a number has 1 byte.
2
In this exercise we are not considering other factors that would make the processing
slower, such as RAM latency, copy cache operations, etc.

133
134 CHAPTER 7. ASYMPTOTIC ANALYSIS

Figure 7.1: Asymptotic behaviour of many classes of functions.

is a polynomial-time reduction, meaning that the reduction process takes

polynomial time. A problem is hard for a class of problems if every problem
in it can be reduced to the original problem.

P
The complexity class of decision problems that can be solved on a determin-
istic Turing machine in polynomial time (in the worst case). If we can turn
a problem into a decision problem, the result would belong to P.

NP
The complexity class of decision problems that can be solved on a non-
deterministic Turing machine (NTM) in polynomial time. In other words,
it includes all decision problems whose yes instances can be solved in poly-
nomial time with the NTM. A problem is called complete if all problems
in the class are reduced to it. Therefore, the subclass called NP-complete
(NPC) contains the hardest problems in all of NP.
Any problem that is at least as hard (determined by polynomial-time
reduction) as any problem in NP, but that need not itself be in NP, is
called NP-hard. For example, finding the shortest route through a graph,
7.2. RECURSION 135

which is called the Travelling Salesman (or Salesrep) problem (TSP).

P=NP?
The class co-NP is the class of the complements of NP problems. For every
yes answer, we have the no, and vice versa. If NP is truly asymmetric,
then these two classes are different. Although there is overlap between them
because all of P lies in their intersection: both the yes and no instances in
P can be solved in polynomial time with an NTM.
What would happen if a NPC was found in a intersection of N and
co-NP? First, it would mean that all of NP would be inside co-NP, so we
would show NP = co-NP and the asymmetry would disappear. Second,
since all of P is in this intersection, P = NP. If P = NP, we could solve
any (decision) problem that had a practical (verifiable) solution.
However, it is (strongly) believed that NP and co-NP are different. For
instance, no polynomial solution to the problem of factoring numbers was
found, and this problem is in both NP and co-NP.

7.2 Recursion
The three laws of recursion are:

1. A recursive algorithm must have a base case.

2. A recursive algorithm must change its state and move toward the base
case.

3. A recursive algorithm must call itself, recursively.

For every recursive call, the recursive function has to allocate memory
on the stack for arguments, return address, and local variables, costing time
to push and pop these data onto the stack. Recursive algorithms take at
least O(n) space where n is the depth of the recursive call.
Recursion is very costly when there are duplicated calculations and/or
there are overlap among subproblems. In some cases this can cause the stack
to overflow. For this reason, where subproblems overlap, iterative solutions
might be a better approach. For example, in the case of the Fibonacci
series, the iterative solution runs on O(n) while the recursive solution runs
on exponential runtime.
136 CHAPTER 7. ASYMPTOTIC ANALYSIS

Recursive Relations
To describe the running time of recursive functions, we use recursive rela-
tions:
T (n) = a T (g(n)) + f (n),
where a represents the number of recursive calls, g(n) is the size of each
subproblem to be solved recursively, and f (n) is any extra work done in the
function. The following table shows examples of recursive relations:
T (n) = T (n 1) + 1 O(n) Processing a sequence
T (n) = T (n 1) + n O(n2 ) Handshake problem
T (n) = 2T (n 1) + 1 O(2n ) Towers of Hanoi
T (n) = T (n/2) + 1 O(ln n) Binary search
T (n) = T (n/2) + n O(n) Randomized select
T (n) = 2T (n/2) + 1 O(n) Tree transversal
T (n) = 2T (n/2) + n O(n ln n) Sort by divide and conquer

Divide and Conquer Algorithms

Recurrences for the divide and conquer algorithms have the form:

T (n) = a T (n/b) + f (n),

where we have a recursive calls, each with a percentage 1/b of the dataset.
Summing to this, the algorithm does f (n) of work. To reach the problem of
T(1) = 1 in the final instance (leaf, as we will learn when we study trees),
the height is defined as h = lnb n, Fig. 7.2.

7.3 Runtime in Functions

We are now ready to estimate algorithm runtimes. First of all, if the al-
gorithm does not have any recursive calling, we only need to analyse its
data structures and flow blocks. In this case, complexities of code blocks
executed one after the other are just added and complexities of nested loops
are multiplied.
If the algorithm has recursive calls, we can use the recursive functions
from the previous section to find the runtime. When we write a recurrence
relation for a function, we must write two equations, one for the general case
and one for the base case (that should be O(1), so that T (1) = 1). Keeping
this in mind, let us take a look at the example of the algorithm to find the
nth element in a Fibonacci sequence, which is known as to be exponential:
7.3. RUNTIME IN FUNCTIONS 137

Figure 7.2: Tree illustrating divide and conquer recurrences.

[general_poroblems/numbers/find_fibonacci_seq.py]

def find_fibonacci_seq_rec(n):
if n < 2: return n
return find_fibonacci_seq_rec(n - 1) +
find_fibonacci_seq_rec(n - 2)

Here, g(n)s are n 2 and n 1, a is 2, and f (n) is 1, so the recursive

relation in this case is

T (n) = 2T (n 1) + 1.

Now let us open this equation for each next recursion:

T (n) = 22 T (n 2) + 2 2k T (n k) + k...

We need to make sure that the function have O(1) in the base case,
where it is T (1) = 1, this means that n k = 1 or k = n 1. So plugging
back into the equation, we have:

T (n) = 2n1 + n 1 2n . (7.3.1)

We have indeed proved that this algorithm is exponential! The same

process can be done for each recursive relation and the following table shows
the runtime results for many algorithm:
138 CHAPTER 7. ASYMPTOTIC ANALYSIS

O(n2 ) quadratic insertion, selection sort

O(n ln n) loglinear algorithms breaking problem into smaller chunks
per invocation, and then sticking the results together,
quick and merge sort
O(n) linear iteration over a list
O(ln n) log algorithms breaking problem into smaller chunks
per invocation, searching a binary search tree
O(1) constant hash table lookup/modification
O(nk ) polynomial k-nested for loops over n
O(kn ) exponential producing every subset of n items
O(n!) factorial producing every ordering of n values
Chapter 8

Sorting

The simplest way of sorting a group of items is to start by removing the

smallest item from the group, and putting it first. Then removing the next
smallest, and putting it next and so on. This is clearly an O(n2 ) algo-
rithm, so we need to find a better solution. In this chapter we will look at
many examples of sorting algorithms and analyse their characteristics and
runtimes.
An in-place sort does not use any additional memory to do the sorting
(for example, swapping elements in an array). A stable sort preserves the
relative order of otherwise identical data elements (for example, if two data
elements have identical values, the one that was ahead of the other stays
ahead). In any comparison sort problem, a key is the value (or values) that
determines the sorting order. A comparison sort requires only that there is a
way to determine if a key is less than, equal to, or greater than another key.
Most sorting algorithms are comparison sorts where the worst-case running
time for such sorts can be no better than O(n ln n).

8.1 Quadratic Sort

Insertion Sort
Insertion sort is a simple sorting algorithm with best runtime case runtime
of O(n) and average and worst runtime cases of O(n2 ). It sorts by repeatedly
inserting the next unsorted element in an initial sorted segment of the array.
For small data sets, it can be preferable to more advanced algorithms such
as merge sort or quicksort if the list is already sorted (it is a good way to
add new elements to a presorted list):

139
140 CHAPTER 8. SORTING

[sorting/insertion_sort.py]

def insertion_sort(seq):
for i in range(1, len(seq)):
j = i
while j > 0 and seq[j-1] > seq[j]:
seq[j-1], seq[j] = seq[j], seq[j-1]
j -= 1
return seq

def insertion_sort_rec(seq, i = None):

if i == None: i = len(seq) -1
if i == 0: return i
insertion_sort_rec(seq, i-1)
j = i
while j > 0 and seq[j-i] > seq[j]:
seq[j-1], seq[j] = seq[j], seq[j-1]
j -= 1
return seq

def test_insertion_sort():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2, 5, 4, 1, 5, 3]
assert(insertion_sort(seq) == sorted(seq))
assert(insertion_sort_rec(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_insertion_sort()

Selection Sort
Selection sort is based on finding the smallest or largest element in a list
and exchanging it to the first, then finding the second, etc, until the end is
reached. Even when the list is sorted, it is O(n2 ) (and not stable):
[sorting/selection_sort.py]

def selection_sort(seq):
for i in range(len(seq) -1, 0, -1):
max_j = i
for j in range(max_j):
if seq[j] > seq[max_j]:
8.1. QUADRATIC SORT 141

max_j = j
seq[i], seq[max_j] = seq[max_j], seq[i]
return seq

def test_selection_sort():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2]
assert(selection_sort(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_selection_sort()

Gnome Sort

Gnome sort works by moving forward to find a misplaced value and then
moving backward to place it in the right position:
[sorting/gnome_sort.py]

def gnome_sort(seq):
i = 0
while i < len(seq):
if i ==0 or seq[i-1] <= seq[i]:
i += 1
else:
seq[i], seq[i-1] = seq[i-1], seq[i]
i -= 1
return seq

def test_gnome_sort():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2, 5, 4, 1, 5, 3]
assert(gnome_sort(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_gnome_sort()
142 CHAPTER 8. SORTING

8.2 Linear Sort

Count Sort

Count sort sorts integers with a small value range, counting occurrences
and using the cumulative counts to directly place the numbers in the result,
updating the counts as it goes.
There is a loglinear limit on how fast you can sort if all you know about
your data is that they are greater or less than each other. However, if you
can also count events, sort becomes linear in time, O(n + k):
[sorting/count_sort.py]

from collections import defaultdict

def count_sort_dict(a):
b, c = [], defaultdict(list)
for x in a:
c[x].append(x)
for k in range(min(c), max(c) + 1):
b.extend(c[k])
return b

def test_count_sort():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2, 5, 4, 1, 5, 3]
assert(count_sort_dict(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_count_sort()

If several values have the same key, they will have the original order with
respect with each other, so the algorithm is stable.

8.3 Loglinear Sort

The sort() and sorted() Methods

In Python, we would normally sort a list by using the list.sort() (in-place)

method or any other iterable item by the sorted() function. They are both
8.3. LOGLINEAR SORT 143

a very efficient implementation of the Pythons timsort algorithm1 . With

the sorted() function, the original object is not changed.
The sorted() function can also be customized though optional argu-
ments, for example: reverse=True; key=len; str.lower (treating upper-
case and lowercase the same); or even with a custom sorting function.

Merge Sort
Merge sort divides the list in half to create two unsorted lists. These two
unsorted lists are sorted and merged by continually calling the merge-sort
algorithm, until you get a list of size 1. The algorithm is stable, as well as
fast for large data sets. However, since it is not in-place, it requires much
more memory than many other algorithms. The space complexity is O(n)
for arrays and O(ln n) for linked lists2 . The best, average, and worst case
times are all O(n ln n).
Merge sort is a good choice when the data set is too large to fit into the
memory. The subsets can be written to disk in separate files until they are
small enough to be sorted in memory. The merging is easy, and involves
just reading single elements at a time from each file and writing them to the
final file in the correct order:
[sorting/merge_sort.py]

O(log(n))
def merge_sort(seq):
if len(seq) < 2 : return seq
mid = len(seq)//2
left, right = None, None
if seq[:mid]: left = merge_sort([:mid])
if seq[mid:]: right = merge_sort([mid:])
return merge_n(left,right)

#O(2n)
def merge_2n(left, right):
if not left or not right:
return left or right
result = []
1
Timsort is a hybrid sorting algorithm, derived from merge sort and insertion sort, and
invented by Tim Peters for Python.
2
Never ever consider to sort a linked list tough, it is problem the worst option you have
in terms of runtime complexity.
144 CHAPTER 8. SORTING

while left and right :

if left[-1] >= right[-1]:
result.append(left.pop())
else:
result.append(right.pop())
result.reversed()
return (left or right) + result

#O(n)
def merge_n(left,right):
if not left or not right:
return left or right
result = []
i,j = 0,0
while i < len(left) and j < len(right):
if left[i] <= right[i]:
result.append(left[i])
i+=1
else :
result.append(right[j])
j+=1
if i < len(left) - 1 : result+=left[i:]
elif j < len(right) - 1 : result += right[j:]
return result

def test_merge_sort():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2]
assert(merge_sort(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_merge_sort()

Quick Sort
Quick sort works by choosing a pivot and partitioning the array so that the
elements that are smaller than the pivot goes to the left. Then, it recursively
sorts the left and right parts.
The choice of the pivot value is a key to the performance. It can be
shown that always choosing the value in the middle of the set is the best
choice for already-sorted data and no worse than most other choices for
random unsorted data.
8.3. LOGLINEAR SORT 145

The worst case is O(n2 ) in the rare cases when partitioning keeps pro-
ducing a region of n 1 elements (when the pivot is the minimum value).
The best case produces two n/2-sized lists. This and the average case are
both O(n ln n). The algorithm is not stable.
[sorting/quick_sort.py]

def quick_sort(seq):
if len(seq) < 2 : return seq
mid = len(seq)//2
pi = seq[mid]
seq = seq[:mid] + seq[mid+1:]
lo = [x for x in seq if x <= pi]
hi = [x for x in seq if x > pi]
return quick_sort(lo) + [pi] + quick_sort(hi)

def test_quick_sort():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2]
assert(quick_sort(seq) == sorted(seq))
assert(quick_sort_divided(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_quick_sort()

Heap Sort
Heap sort is similar to a selection sort, except that the unsorted region is a
heap, so finding the largest element n times gives a loglinear runtime.
In a heap, for every node other than the root, the value of the node is at
least (at most) the value of its parent. Thus, the smallest (largest) element is
stored at the root and the subtrees rooted at a node contain larger (smaller)
values than does the node itself.
Although the insertion is only O(1), the performance of validating (the
heap order) is O(ln n). Searching (traversing) is O(n). In Python, a heap
sort can be implemented by pushing all values onto a heap and then popping
off the smallest values one at a time:
[sorting/heap_sort1.py]

import heapq
146 CHAPTER 8. SORTING

def heap_sort1(seq):
heap sort with Pythons heapq
h = []
for value in seq:
heapq.heappush(h, value)
return [heapq.heappop(h) for i in range(len(h))]

def test_heap_sort1():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2]
assert(heap_sort1(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_heap_sort1()

If we decide to use the heap class that we have from the last chapters,
we can write a heap sort simply by:
[sorting/heap_sort2.py]

from heap import Heap

def heap_sort2(seq):
heap = Heap(seq)
res = []
for i in range(len(seq)):
res.insert(0, heap.extract_max())
return res

def test_heap_sort2():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2]
assert(heap_sort2(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_heap_sort2()

Finally, we can also write our heap sort explicitly:

[sorting/heap_sort3.py]
8.3. LOGLINEAR SORT 147

def heap_sort3(seq):
for start in range((len(seq)-2)//2, -1, -1):
siftdown(seq, start, len(seq)-1)
for end in range(len(seq)-1, 0, -1):
seq[end], seq[0] = seq[0], seq[end]
siftdown(seq, 0, end - 1)
return seq

def siftdown(seq, start, end):

root = start
while True:
child = root * 2 + 1
if child > end: break
if child + 1 <= end and seq[child] < seq[child + 1]:
child += 1
if seq[root] < seq[child]:
seq[root], seq[child] = seq[child], seq[root]
root = child
else:
break

def test_heap_sort():
seq = [3, 5, 2, 6, 8, 1, 0, 3, 5, 6, 2]
assert(heap_sort3(seq) == sorted(seq))
print(Tests passed!)

if __name__ == __main__:
test_heap_sort3()
148 CHAPTER 8. SORTING

8.4 Comparison Between Sorting Methods

8.5. ADDITIONAL EXERCISES 149

8.5 Additional Exercises

Quadratic Sort
The following program implements a bubble sort, a very inefficient sorting
algorithm:
[searching/bubble_sort.py]

def bubble_sort(seq):
size = len(seq) -1
for num in range(size, 0, -1):
for i in range(num):
if seq[i] > seq[i+1]:
temp = seq[i]
seq[i] = seq[i+1]
seq[i+1] = temp
return seq

def test_bubble_sort(module_name=this module):

seq = [4, 5, 2, 1, 6, 2, 7, 10, 13, 8]
assert(bubble_sort(seq) == sorted(seq))
s = Tests in {name} have {con}!
print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_bubble_sort()

Linear Sort
The example bellow shows a simple count sort for people ages:
def counting_sort_age(A):
oldestAge = 100
timesOfAge = [0]*oldestAge
ageCountSet = set()
B = []
for i in A:
timesOfAge[i] += 1
ageCountSet.add(i)
for j in ageCountSet:
count = timesOfAge[j]
150 CHAPTER 8. SORTING

while count > 0:

B.append(j)
count -= 1
return B

The example bellow uses quick sort to find the k largest elements in a
sequence:
[sorting/find_k_largest_seq_quicksort.py]

import random

def swap(A, x, y):

tmp = A[x]
A[x] = A[y]
A[y] = tmp

def qselect(A, k, left=None, right=None):

left = left or 0
right = right or len(A) - 1
pivot = random.randint(left, right)
pivotVal = A[pivot]

swap(A, pivot, right)

swapIndex, i = left, left
while i <= right - 1:
if A[i] < pivotVal:
swap(A, i, swapIndex)
swapIndex += 1
i += 1

swap(A, right, swapIndex)

rank = len(A) - swapIndex
if k == rank:
return A[swapIndex]
elif k < rank:
return qselect(A, k, left=swapIndex+1, right=right)
else:
return qselect(A, k, left=left, right=swapIndex-1)

def find_k_largest_seq_quickselect(seq, k):

kth_largest = qselect(seq, k)
result = []
8.5. ADDITIONAL EXERCISES 151

for item in seq:

if item >= kth_largest:
result.append(item)
return result

def test_find_k_largest_seq_quickselect():
seq = [3, 10, 4, 5, 1, 8, 9, 11, 5]
k = 2
assert(find_k_largest_seq_quickselect(seq,k) == [10, 11])

if __name__ == __main__:
test_find_k_largest_seq_quickselect()
152 CHAPTER 8. SORTING
Chapter 9

Searching

The most common searching algorithms are the sequential search and the
binary search. If an input array is not sorted, or the input elements are
accommodated by dynamic containers (such as linked lists), the search has
to be sequential. If the input is a sorted array, the binary search algorithm
is the best choice. If we are allowed to use auxiliary memory, a hash table
might help the search, with which a value can be located in O(1) time with
a key.

9.1 Sequential Search

In the following example we illustrate the runtime of a sequential search for
items stored in a collection. If the item is present, the best case is O(1), the
average case is O(n/2), and the worst case is O(n). However, if the item is
not present, all three cases will be O(n):
[searching/sequential_search.py]

def sequential_search(seq, n):

for item in seq:
if item == n: return True
return False

def test_sequential_search(module_name=this module):

seq = [1, 5, 6, 8, 3]
n1 = 5
n2 = 7
assert(sequential_search(seq, n1) == True)

153
154 CHAPTER 9. SEARCHING

assert(sequential_search(seq, n2) == False)

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_sequential_search()

Now, if we sort the sequence first, we can improve the sequential search
in the case when the item is not present to have the same runtimes as when
the item is present:
[searching/ordered_sequential_search.py]

def ordered_sequential_search(seq, n):

item = 0
for item in seq:
if item > n: return False
if item == n: return True
return False

def test_ordered_sequential_search(module_name=this module):

seq = [1, 2, 4, 5, 6, 8, 10]
n1 = 10
n2 = 7
assert(ordered_sequential_search(seq, n1) == True)
assert(ordered_sequential_search(seq, n2) == False)
s = Tests in {name} have {con}!
print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_ordered_sequential_search()

9.2 Binary Search

A binary search finds the position of a specified input value (the key) within
a sorted array. In each step, the algorithm compares the search key value
with the key value of the middle element of the array. If the keys match
the items index, (position) is returned. Otherwise, if the search key is less
than the middle elements key, the algorithm repeats the process in the left
subarray; or if the search key is greater, on the right subarray. The algorithm
9.2. BINARY SEARCH 155

runs on O(ln n):

[searching/binary_search.py]

def binary_search_rec(seq,key): # Recursive

if not seq : return None
mid = len(seq)//2
if key == seq[mid] : return mid
elif key < seq[mid] : return binary_search_rec(seq[:mid],key)
else : return binary_search_rec(seq[mid+1:],key)

def binary_search_iter(seq,key): # Iterative

hi, lo = len(seq), 0
while lo < hi :
mid = (hi + lo)//2
if key == seq[mid] : return mid
elif key < seq[mid] : hi = mid
else : low = mid + 1
return None

def test_binary_search():
seq = [1,2,5,6,7,10,12,12,14,15]
key = 6
assert(binary_search_iter(seq, key) == 3)
assert(binary_search_rec(seq, key) == 3)
print(Tests passed!)

if __name__ == __main__:
test_binary_search()

The bisect Module

Pythons built-in bisect() module is available for binary search in a sorted
sequence:
>>> from bisect import bisect
>>> list = [0, 3, 4, 5]
>>> bisect(list, 5)
4

Note that the module returns the index after the key, which is where you
should place the new value. Other available functions are bisect right and
bisect left.
156 CHAPTER 9. SEARCHING

9.3 Additional Exercises

Searching in a Matrix
The following module searches an entry in a matrix where the rows and
columns are sorted. In this case, every row is increasingly sorted from left
to right, and every column is increasingly sorted from top to bottom. The
runtime is linear on O(m + n):
[general_problems/numbers/search_entry_matrix.py]

def find_elem_matrix_bool(m1, value):

found = False
row = 0
col = len(m1[0]) - 1
while row < len(m1) and col >= 0:
if m1[row][col] == value:
found = True
break
elif m1[row][col] > value:
col-=1
else:
row+=1
return found

def test_find_elem_matrix_bool(module_name=this module):

m1 = [[1,2,8,9], [2,4,9,12], [4,7,10,13], [6,8,11,15]]
assert(find_elem_matrix_bool(m1,8) == True)
assert(find_elem_matrix_bool(m1,3) == False)
m2 = [[0]]
assert(find_elem_matrix_bool(m2,0) == True)

s = Tests in {name} have {con}!

print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_find_elem_matrix_bool()

The following problem searches an element in a matrix where in every

row, the values increasing from left to right, but the last number in a row
is smaller than the first number in the next row. The naive brute force
solution scans all numbers and costs O(mn). However, since the numbers
are already sorted, the matrix can be viewed as a 1D sorted array and we
can use the binary search algorithm with efficiency O(lognm):
9.3. ADDITIONAL EXERCISES 157

[searching/searching_in_a_matrix.py]

import numpy

def searching_in_a_matrix(m1, value):

rows = len(m1)
cols = len(m1[0])
lo = 0
hi = rows*cols
while lo < hi:
mid = (lo + hi)//2
row = mid//cols
col = mid%cols
v = m1[row][col]
if v == value: return True
elif v > value: hi = mid
else: lo = mid+1
return False

def test_searching_in_a_matrix():
a = [[1,3,5],[7,9,11],[13,15,17]]
b = numpy.array([(1,2),(3,4)])
assert(searching_in_a_matrix(a, 13) == True)
assert(searching_in_a_matrix(a, 14) == False)
assert(searching_in_a_matrix(b, 3) == True)
assert(searching_in_a_matrix(b, 5) == False)
print(Tests passed!)

if __name__ == __main__:
test_searching_in_a_matrix()

Unimodal Arrays
An array is unimodal if it consists of an increasing sequence followed by
a decreasing sequence. The example below shows how to find the locally
maximum of an array using binary search:
[searching/find_max_unimodal_array.py]

def find_max_unimodal_array(A):
if len(A) <= 2 : return None
left = 0
158 CHAPTER 9. SEARCHING

right = len(A)-1
while right > left +1:
mid = (left + right)//2
if A[mid] > A[mid-1] and A[mid] > A[mid+1]:
return A[mid]
elif A[mid] > A[mid-1] and A[mid] < A[mid+1]:
left = mid
else:
right = mid
return None

def test_find_max_unimodal_array():
seq = [1, 2, 5, 6, 7, 10, 12, 9, 8, 7, 6]
assert(find_max_unimodal_array(seq) == 12)
print(Tests passed!)

if __name__ == __main__:
test_find_max_unimodal_array()

Calculating Square Roots

The example bellow implements the square root of a number using binary
search:
[searching/find_sqrt_bin_search.py]

def find_sqrt_bin_search(n, error=0.001):

lower = n < 1 and n or 1
upper = n < 1 and 1 or n
mid = lower + (upper - lower) / 2.0
square = mid * mid
while abs(square - n) > error:
if square < n:
lower = mid
else:
upper = mid
mid = lower + (upper - lower) / 2.0
square = mid * mid
return mid

def test_ind_sqrt_bin_search():
number = 9
assert(find_sqrt_bin_search(number) == 3)
9.3. ADDITIONAL EXERCISES 159

print(Tests passed!)

if __name__ == __main__:
test_ind_sqrt_bin_search()

Counting Frequency of Elements

The following example finds how many times a k element appears in a sorted
list. Since the array is sorted, binary search gives a O(logn) runtime:
[searching/find_time_occurence_list.py]

from binary_search import binary_search

def find_time_occurrence_list(seq, k):

index_some_k = binary_serch(seq, k)
count = 1
sizet = len(seq)
for i in range(index_some_k+1, sizet): # go up
if seq[i] == k: count +=1
else: break
for i in range(index_some_k-1, -1, -1): # go down
if seq[i] == k: count +=1
else: break
return count

def test_find_time_occurrence_list():
seq = [1,2,2,2,2,2,2,5,6,6,7,8,9]
k = 2
assert(find_time_occurrence_list(seq, k) == 6)
print(Tests passed!)

if __name__ == __main__:
test_find_time_occurrence_list()

Intersection of Arrays
The snippet bellow shows three ways to perform the intersection of two
sorted arrays. The simplest way is to use sets, however this will not preserve
the ordering. The second example uses an adaptation of the merge sort. The
160 CHAPTER 9. SEARCHING

third example is suitable when one of the arrays is much larger than other.
In this case, binary search is the best option:
[searching/intersection_two_arrays.py]

def intersection_two_arrays_sets(seq1, seq2):

find the intersection of two arrays using set proprieties

set1 = set(seq1)
set2 = set(seq2)
return set1.intersection(set2) #same as list(set1 & set2

def intersection_two_arrays_ms(seq1, seq2):

find the intersection of two arrays using merge sort
res = []
while seq1 and seq2:
if seq1[-1] == seq2[-1]:
res.append(seq1.pop())
seq2.pop()
elif seq1[-1] > seq2[-1]:
seq1.pop()
else:
seq2.pop()
res.reverse()
return res

from binary_search import binary_search

def intersection_two_arrays_bs(seq1, seq2):
using binary search
if len(seq1) > len(seq2): seq, key = seq1, seq2
else: seq, key = seq2, seq1
intersec = []
for item in key:
if binary_search(seq, item):
intersec.append(item)
return intersec

def test_intersection_two_arrays(module_name=this module):

seq1 = [1,2,3,5,7,8]
seq2 = [3,5,6]
assert(set(intersection_two_arrays_sets(seq1,seq2)) ==
set([3,5]))
9.3. ADDITIONAL EXERCISES 161

assert(intersection_two_arrays_bs(seq1,seq2) == [3,5])
assert(intersection_two_arrays_ms(seq1,seq2) == [3,5])
s = Tests in {name} have {con}!
print(s.format(name=module_name, con=passed))

if __name__ == __main__:
test_intersection_two_arrays()
162 CHAPTER 9. SEARCHING
Chapter 10

Dynamic Programming

Dynamic programming is used to simplify a complicated problem by breaking

it down into simpler subproblems by means of recursion. If a problem has
an optimal substructure and overlapping subproblems, it may be solved by
dynamic programming.
Optimal substructure means that the solution to a given optimization
problem can be obtained by a combination of optimal solutions to its sub-
problems. The first step to utilize dynamic programming is to check whether
the problem exhibits such optimal substructure. The second step is having
overlapping problems by solving subproblems once and then storing the solu-
tion to be retrieved. A choice is made at each step in dynamic programming,
and the choice depends on the solutions to subproblems, bottom-up manner,
from smaller subproblems to larger subproblems.

10.1 Memoization
Dynamically Solving the Fibonacci Series
High-level languages such as Python can implement the recursive formula-
tion directly, caching return values. Memoization is a method where if a call
is made more than once with the same arguments, and the result is returned
directly from the cache.
For example, we can dynamically solve the exponential Fibonacci series
by using a memo function designed as an algorithm that uses nested scopes
to give the wrapped function memory:
[dynamic_programming/memo.py]

163
164 CHAPTER 10. DYNAMIC PROGRAMMING

from functools import wraps

def memo(func):
cache = {}
@wraps(func)
def wrap(*args):
if args not in cache:
cache[args] = func(*args)
return cache[args]
return wrap

>>> fibonacci = memo(fibonacci)

>>> fibonacci(130)
1066340417491710595814572169

We could even use the decorator directly in the function:

@memo
def fib(i):
if i < 2: return 1
return fib(i-1) + fib(i-2)
>>> fibonacci(130)
1066340417491710595814572169
10.2. ADDITIONAL EXERCISES 165

10.2 Additional Exercises

Longest Increasing Subsequence
Another interesting application of memoization is to the problem of finding
the longest increasing subsequence1 of a given sequence. A naive recur-
sive solution gives a O(2n ) runtime. However, the iterative solution can be
solved in loglinear time using dynamic programming.
In the example below we benchmark these functions with an array with
40 elements to find that the memoized version takes less than one second to
run, while without dynamic programming, the function can take over 120
seconds to run:
[dynamic_programming/memoized_longest_inc_subseq.py]

from itertools import combinations

from bisect import bisect
from memo import memo
from do_benchmark import benchmark

def naive_longest_inc_subseq(seq):
exponential solution to the longest increasing subsequence
problem
for length in range(len(seq), 0, -1):
for sub in combinations(seq, length):
if list(sub) == sorted(sub):
return len(sub)

def longest_inc_subseq1(seq):
iterative solution for the longest increasing subsequence
problem
end = []
for val in seq:
idx = bisect(end, val)
if idx == len(end): end.append(val)
else: end[idx] = val
return len(end)

def longest_inc_subseq2(seq):
another iterative algorithm for the longest increasing
subsequence problem
1
See other versions of this problem in the end of the chapter about lists in Python.
166 CHAPTER 10. DYNAMIC PROGRAMMING

L = [1] * len(seq)
for cur, val in enumerate(seq):
for pre in range(cur):
if seq[pre] <= val:
L[cur] = max(L[cur], 1 + L[pre])
return max(L)

def memoized_longest_inc_subseq(seq):
memoized recursive solution to the longest increasing
subsequence problem
@memo
def L(cur):
res = 1
for pre in range(cur):
if seq[pre] <= seq[cur]:
res = max(res, 1 + L(pre))
return res
return max(L(i) for i in range(len(seq)))

@benchmark
def test_naive_longest_inc_subseq():
print(naive_longest_inc_subseq(s1))

benchmark
def test_longest_inc_subseq1():
print(longest_inc_subseq1(s1))

@benchmark
def test_longest_inc_subseq2():
print(longest_inc_subseq2(s1))

@benchmark
def test_memoized_longest_inc_subseq():
print(memoized_longest_inc_subseq(s1))

if __name__ == __main__:
from random import randrange
s1 = [randrange(100) for i in range(40)]
print(s1)
test_naive_longest_inc_subseq()
test_longest_inc_subseq1()
test_longest_inc_subseq2()
10.2. ADDITIONAL EXERCISES 167

test_memoized_longest_inc_subseq()
168 CHAPTER 10. DYNAMIC PROGRAMMING
 Part III

Climbing Graphs and Trees

169
Chapter 11

Introduction to Graphs

11.1 Basic Definitions

A graph is an abstract network, consisting of nodes (or vertices, V ) connected
by edges (or arcs, E). A graph can be defined as a pair of sets, G = (V, E),
where the node set V is any finite set, and the edge set E is a set of node
pairs. For example, some graph can be simply represented by the node set
V = {a, b, c, d} and the edge set E = {{a, b}, {b, c}, {c, d}, {d, a}}.

Direction of a Graph
If a graph has no direction, it is referred as undirected. In this case, nodes
with an edge between them are adjacent and adjacent nodes are neighbors.
If the edges have a direction, the graph is directed (digraph). In this
case, the graph has leaves. The edges are no longer unordered: an edge
between nodes u and v is now either an edge (u, v) from u to v, or an edge
(v, u) from v to u. We can say that in a digraph G, the function E(G) is a
relation over V (G).

Subgraphs
A subgraph of G consists of a subset of V and E. A spanning subgraph
contains all the nodes of the original graph.

Completeness of a Graph
If all the nodes in a graph are pairwise adjacent, the graph is called complete.

171
172 CHAPTER 11. INTRODUCTION TO GRAPHS

Degree in a Node

The number of undirected edges incident on a node is called degree. Zero-

degree graphs are called isolated. For directed graphs, we can split this num-
ber into in-degree (incoming edges/parents) and out-degree/children (outgo-
ing edges).

Paths, Walks, and Cycle

A path in G is a subgraph where the edges connect the nodes in a sequence,

without revisiting any node. In a directed graph, a path has to follow the
directions of the edges.
A walk is an alternating sequence of nodes and edges that allows nodes
and edges to be visited multiple times.
A cycle is like a path except that the last edge links the last node to the
first.

Length of a Path

The length of a path or walk is the value given by its edge count.

Weight of an Edge
Associating weights with each edge in G gives us a weighted graph. The
weight of a path or cycle is the sum of its edge weights. So, for unweighted
graphs, it is simply the number of edges.

Planar Graphs

A graph that can be drawn on the plane without crossing edges is called
planar. This graph has regions, which are areas bounded by the edges.The
Eulers formula for connected planar graphs says that V E + F = 2, where
V, E, F are the number of nodes, edges, and regions, respectively.

Graph Traversal

A traversal is a walk through all the connected components of a graph. The

main difference between graph traversals is the ordering of the to-do list
among the unvisited nodes that have been discovered.
11.2. THE NEIGHBORHOOD FUNCTION 173

Connected and Strongly Connected Components

An undirected graph is connected if there is a path from every node to every
other node. A directed graph is connected if its underlying undirected graph
is connected.
A connected component is a maximal subgraph that is connected. Con-
nected components can be found using traversal algorithms such as depth-
first searching (DFS) or breath-first searching (BFS), as we will see in fol-
lowing chapters.
If there is a path from every node to every other node in a directed
graph, the graph is called strongly connected. A strongly connected compo-
nent (SCC) is a maximal subgraph that is strongly connected.

Trees and Forests

A forest is a cycle-free graph. A tree is an acyclic, connected, and directed
graph. A forest consists of one of more trees.
In a tree, any two two nodes are connected by exactly one path. Adding
any new edge to it creates a cycle and removing any edge yields unconnected
components.

11.2 The Neighborhood Function

A graphs neighborhood function, N (V ), is a container (or iterable object) of
all neighbors of V . The most well-known data structures used to represent
them are adjacent lists and adjacent matrices.

Adjacent Lists
For each node in an adjacent list, we have access to a list (or set or container
or iterable) of its neighbor. Supposing we have n nodes, each adjacent (or
neighbor) list is just a list of such numbers. We place the lists into a main
list of size n, indexable by the node numbers, where the order is usually
arbitrary.

Using Sets as Adjacent Lists:

We can use Pythons set type to implement adjacent lists:
>>> a,b,c,d,e,f = range(6) # nodes
174 CHAPTER 11. INTRODUCTION TO GRAPHS

>>> N = [{b,c,d,f}, {a,d,f}, {a,b,d,e}, {a,e}, {a,b,c}, {b,c,d,e}]

>>> b in N[a] # membership
True
>>> b in N[b] # membership
False
>>> len(N[f]) # degree
4

Using Lists as Adjacent Lists:

We can also use Pythons lists to implement adjacent lists, which let you
efficiently iterate N (V ) over any node V . Replacing sets with lists makes
membership checking to be O(n). If all that your algorithm does is iterating
over neighbors, using list may be preferential. However if the graph is dense
(many edges), adjacent sets are a better solution:
>>> a,b,c,d,e,f = range(6) # nodes
>>> N = [[b,c,d,f], [a,d,f], [a,b,d,e], [a,e], [a,b,c], [b,c,d,e]]
>>> b in N[a] # membership
True
>>> b in N[b] # membership
False
>>> len(N[f]) # degree
4

Deleting objects from the middle of a Python list is O(n), but deleting
from the end is only O(1). If the order of neighbors is not important, you
can delete an arbitrary neighbor in O(1) time by swapping it in to the last
item in the list and then calling pop().

Using Dictionaries as Adjacent Lists:

Finally, we can use dictionaries as adjacent lists. In this case, the neighbors
would be the keys, and we are able to associate each of them with some
extra value, such as an edge weight:
>>> a,b,c,d,e,f = range(6) # nodes
>>> N = [{b:2,c:1,d:4,f:1}, {a:4,d:1,f:4}, {a:1,b:1,d:2,e:4},
{a:3,e:2}, {a:3,b:4,c:1}, {b:1,c:2,d:4,e:3}]
>>> b in N[a] # membership
True
>>> len(N[f]) # degree
4
11.2. THE NEIGHBORHOOD FUNCTION 175

>>> N[a][b] # edge weight for (a,b)

2

A more flexible approach for node labels is to use dictionaries as a main

structure only. For instance, we can use a dictionary with adjacency sets:
>>> a,b,c,d,e,f = range(6) # nodes
>>> N = { a:set(bcdf), b:set(adf), c:set(abde),
d:set(ae), e:set(abc), f:set(bcde)}
>>> b in N[a] # membership
True
>>> len(N[f]) # degree
4

Adjacent Matrices
In adjacent matrices, instead of listing all the neighbors for each node, we
have one row with one position for each possible neighbor, filled with True
and False values. The simplest implementation of adjacent matrices is given
by nested lists. Note that the diagonal is always False:
>>> a,b,c,d,e,f = range(6) # nodes
>>> N = [[0,1,1,1,0,1], [1,0,0,1,0,1], [1,1,0,1,1,0], [1,0,0,0,1,0],
[1,1,1,0,0,0], [0,1,1,1,1,0]]
>>> N[a][b] # membership
1
>>> N[a][e]
0
>>> sum(N[f]) # degree
4

An adjacent matrix for an undirected graph will always be symmetric.

To add weight to adjacent matrices, we can replace True and False by
values. In this case, non-existent edges can be represented by infinite weights
(float(inf), or None, -1, or very large values):
>>> _ = float(inf) # nodes
>>> N = [[_,2,1,4,_,1], [4,_,_,1,_,4], [1,1,_,2,4,_], [3,_,_,_,2,_],
[3,4,1,_,_,_], [1,2,_,4,3,_]]
>>> N[a][b] < _ # membership
True
>>> sum(1 for w in N[f] if w < _) # degree
4
176 CHAPTER 11. INTRODUCTION TO GRAPHS

Looking up an edge in an adjacent matrix is O(1) while iterating over a

nodes neighbor is O(n).

11.3 Introduction to Trees

While in a graph there may be multiple references to any given node; in a
tree each node (data element) is referenced only by at most one other node,
the parent node. The root node is the node that has no parent. The nodes
referenced by a parent node are called children. A tree is said to be full and
complete if all of its leaves are at the bottom and all of the non-leaf nodes
have exactly two children.

Height or Depth of a Tree

The height (or depth) of a tree is the length of the path from the root to the
deepest node in the tree. It is equal to the maximum level of any node in
the tree. The depth of the root is zero. If the height of a tree is represented
as the log of the number of leaves, the integer number from the log may be
also called depth.

Level or Depth of a Node

The level (or depth) of a node is the length of path from the root to this
node. The set of all nodes at a given depth in a tree is also called the level
of the tree.

Representing Trees
The simplest way of representing a tree is by a nested lists:
>>> T = [a, [b, [d, f]], [c, [e, g]]]
>>> T[0]
a
>>> T[1][0]
b
>>> T[1][1][0]
d
>>> T[1][1][1]
f
>>> T[2][0]
11.3. INTRODUCTION TO TREES 177

c
>>> T[2][1][1]
g

However, this becomes very hard to handle if we we simply add a couple

more branches. The only good way to work with trees is representing them
as a collection of nodes. Let us start with a simple example, where we
define a simple tree class with an attribute for value, another for a children
(or next), and a method for printing the result:
[trees/trees/simple_tree/tree.py]

class SimpleTree(object):
def __init__(self, value=None, children=None):
self.children = children or []
self.value = value

def __repr__(self, level=0):

ret = "\t"*level+repr(self.value)+"\n"
for child in self.children:
ret += child.__repr__(level+1)
return ret

def main():
"""
a
b
d
e
c
h
g
"""
st = SimpleTree(a, [SimpleTree(b, [SimpleTree(d),
SimpleTree(e)] ), SimpleTree(c, [SimpleTree(h),
SimpleTree(g)]) ])
print(st)

if __name__ == __main__:
main()

In the next chapter we will learn how to improve this class, including
many features and methods that a tree can hold. For now, it is useful to
178 CHAPTER 11. INTRODUCTION TO GRAPHS

keep in mind that when we are prototyping data structures such as trees, we
should always be able to come up with a flexible class to specify arbitrary
attributes in the constructor. The following program implements what is
referred to as a bunch class;, a generic tool that is a specialization of the
Pythons dict class and that let you create and set arbitrary attributes on
the fly:
[trees/simple_trees/bunchclass.py]

class BunchClass(dict):
def __init__(self, *args, **kwds):
super(BunchClass, self).__init__(*args, **kwds)
self.__dict__ = self

def main():
{right: {right: Xander, left: Willow}, left:
{right: Angel, left: Buffy}}
bc = BunchClass # notice the absence of ()
tree = bc(left = bc(left="Buffy", right="Angel"), right =
bc(left="Willow", right="Xander"))
print(tree)

if __name__ == __main__:
main()

In the example above, the functions arguments *args and **kwds can
hold an arbitrary number of arguments and an arbitrary number of keywords
arguments, respectively.
Chapter 12

Binary Trees

12.1 Basic Concepts

Binary trees are tree data structures where each node has at most two child
nodes: the left and the right. Child nodes may contain references to their
parents. The root of a tree (the ancestor of all nodes) can exist either inside
or outside the tree.
Binary trees can be seen as a way of passing an initial number n of tokens
down, meaning that at any point in the tree the sum of all the horizontal
nodes will be n. The degree of every node is maximum two. Supposing that
an arbitrary rooted tree has m internal nodes and each internal node has
exactly two children, if the tree has n leaves, the degree of the tree is n 1:

2m = n + m 1 m = n 1,

i.e a tree with n nodes has exactly n 1 branches or degree.

12.2 Representing Binary Trees

The simplest (and silliest) way to represent a binary tree is using Pythons
lists. The following module constructs a list with a root and two empty
sublists for the children. To add a left subtree to the root of a tree, we
insert a new list into the second position of the root list. Note that this
algorithm is not very efficient due to the restrictions that Pythons lists
have on inserting and popping in the middle::

179
180 CHAPTER 12. BINARY TREES

Figure 12.1: The height (h) and width (number of leaves) of a (perfectly
balanced) binary tree.

[trees/binary_trees/BT_lists.py]

def BinaryTreeList(r):
return [r, [], []]

def insertLeft(root, newBranch):

t = root.pop(1)
if len(t) > 1:
root.insert(1,[newBranch,t,[]])
else:
root.insert(1,[newBranch, [], []])
return root

def insertRight(root, newBranch):

t = root.pop(2)
if len(t) > 1:
root.insert(2,[newBranch,[],t])
else:
root.insert(2,[newBranch,[],[]])
return root

def getRootVal(root):
return root[0]

def setRootVal(root, newVal):

12.2. REPRESENTING BINARY TREES 181

root[0] = newVal

def getLeftChild(root):
return root[1]

def getRightChild(root):
return root[2]

def main():

3
[5, [4, [], []], []]
[7, [], [6, [], []]]

r = BinaryTreeList(3)
insertLeft(r,4)
insertLeft(r,5)
insertRight(r,6)
insertRight(r,7)
print(getRootVal(r))
print(getLeftChild(r))
print(getRightChild(r))

if __name__ == __main__:
main()

However this method is not very practical when we have many branches
(or at least it needs many improvements, for example, how it manages the
creation of new lists and how it displays or searches for new elements).
A more natural way to handle binary trees is (again) by representing
it as a collection of nodes. A simple node in a binary tree should carry
attributes for value and for left and right children, and it can have a method
to identify leaves:
[trees/binary_trees/BT.py]

class BT(object):
def __init__(self, value):
self.value = value
self.left = None
self.right = None

def is_leaf(self):
182 CHAPTER 12. BINARY TREES

return self.left is None and self.right is None

def insert_left(self, new_node):

if not self.left:
self.left = BT(new_node)
else:
t = BT(self.left)
t.left = new_node
self.left = t

def insert_right(self, new_node):

if not self.right:
self.right = BT(new_node)
else:
t = BT(self.right)
t.right = new_node
self.right = t

def __repr__(self):
return {}.format(self.value)

def tests_BT():
"""
1
2 3
4 5 6 7
"""
tree = BT(1)
tree.insert_left(2)
tree.insert_right(3)
tree.left().insert_left(4)
tree.left().insert_right(5)
tree.right().insert_left(6)
tree.right().insert_right(7)
print(tree.right().right())
tree.right().right().value(8)
print(tree.right().right())
assert(tree.right().is_leaf() == False)
assert(tree.right().right().is_leaf() == True)
print("Tests Passed!")

if __name__ == __main__:
tests_BT()
12.3. BINARY SEARCH TREES 183

12.3 Binary Search Trees

A binary search tree (BST) is a node-based binary tree data structure which
has the following properties:

1. The left subtree of a node contains only nodes with keys less than the
nodes key.

2. The right subtree of a node contains only nodes with keys greater than
the nodes key.

3. Both the left and right subtrees must also be a binary search tree.

4. There must be no duplicate nodes.

If the binary search tree is balanced, the following operations are O(ln n):
(i) finding a node with a given value (lookup), (ii) finding a node with
maximum or minimum value, and (iii) insertion or deletion of a node.

Representing Binary Search Trees

The following code implements a class for a binary search tree using our
previous binary tree class as a superclass. The main difference now is that
we can only insert a new node under the binary search tree conditions, which
naturally gives us a method for finding an element in the tree. To be able
to achieve this, we use polymorphism on the superclasss insert right and
insert left methods, and we unify them to the proper insert method:
[trees/binary_trees/BST.py]

from BT import BT

class BST(BT):
def __init__(self, value=None):
self.value = value
self.left = None
self.right = None

def insert_left(self, new_node):

self.insert(value)

def insert_right(self, new_node):

self.insert(value)
184 CHAPTER 12. BINARY TREES

def insert(self, value):

if self.value == None:
self.value = value
else:
if value > self.value:
self.right = self.right and self.right.insert(value)
or BST(value)
else:
self.left = self.left and self.left.insert(value) or
BST(value)
return self

def find(self, value):

if value == self.value:
return self
elif value > self.value and self.right:
return self.right.find(value)
elif self.left:
return self.left.find(value)
return None

def main():
"""
4
2 6
1 3 5 7
"""
tree = BST()
tree.insert(4)
tree.insert(2)
tree.insert(6)
tree.insert(1)
tree.insert(3)
tree.insert(7)
tree.insert(5)
print(tree.get_right())
print(tree.get_right().get_left())
print(tree.get_right().get_right())
print(tree.get_left())
print(tree.get_left().get_left())
print(tree.get_left().get_right())
assert(tree.find(30) == None)
12.3. BINARY SEARCH TREES 185

if __name__ == __main__:
main()

There are many other ways that a tree can be created. We could, for
instance, think of two classes, one simply for nodes, and a second one that
controls these nodes. This is not much different from the previous example
(and in the end of this chapter we will see a third hybrid example of these
two):
[trees/binary_trees/BST_with_Nodes.py]

class Node(object):
def __init__(self, value):
self.value = value
self.left = None
self.right = None

def __repr__(self):
return {}.format(self.value)

class BSTwithNodes(object):
def __init__(self):
self.root = None

def insert(self, value):

if not self.root:
self.root = Node(value)
else:
current = self.root
while True:
if value < current.value:
if current.left:
current = current.left
else:
current.left = Node(value)
break;

elif value > current.value:

if current.right:
current = current.right
else:
current.right = Node(value)
break;
186 CHAPTER 12. BINARY TREES

else:
break

def main():
"""
BST
4
2 6
1 3 5 7
"""
tree = BSTwithNodes()
l1 = [4, 2, 6, 1, 3, 7, 5]
for i in l1: tree.insert(i)
print(tree.root)
print(tree.root.right)
print(tree.root.right.left)
print(tree.root.right.right)
print(tree.root.left)
print(tree.root.left.left)
print(tree.root.left.right)

if __name__ == __main__:
main()

12.4 Self-Balancing BST

A balanced tree is a tree where the differences of the height of every subtree
is at most equal to 1. A self-balancing binary search tree is any node-based
binary search tree that automatically keeps itself balanced. By applying a
balance condition we ensure that the worst case runtime of common tree
operations will be at most O(ln n).

Balancing Factor of a Tree

A balancing factor can be attributed to each internal node in a tree, being
the difference between the heights of the left and right subtrees. There
are many balancing methods for trees, but they are usually based on two
operations:

? Node splitting (and merging): nodes are not allowed to have more
than two children, so when a node become overfull it splits into two
12.4. SELF-BALANCING BST 187

subnodes.

? Node rotations: process of switching edges. If x is the parent of y, we

make y the parent of x and x must take over one of the children of y.

AVL Trees
An AVL tree is a binary search tree with a self-balancing condition where
the difference between the height of the left and right subtrees cannot be
more than one.
To implement an AVL tree, we can start by adding a self-balancing
method to our BST classes, called every time we add a new node to the
tree. The method works by continuously checking the height of the tree,
which is added as a new attribute:
def height(node):
if node is None:
return -1
else:
return node.height

def update_height(node):
node.height = max(height(node.left), height(node.right)) + 1

Now we can go ahead and implement the rebalancing method for our
tree. The method will check whether the difference between the new heights
of the right and left subtrees are up to 1. If this is not true, the method will
perform the rotations:
def rebalance(self, node):
while node is not None:
update_height(node)

if height(node.left) >= 2 + height(node.right):

if height(node.left.left) >= height(node.left.right):
self.right_rotate(node)
else:
self.left_rotate(node.left)
self.right_rotate(node)

elif height(node.right) >= 2 + height(node.left):

if height(node.right.right) >=
height(node.right.left):
self.left_rotate(node)
188 CHAPTER 12. BINARY TREES

else:
self.right_rotate(node.right)
self.left_rotate(node)

node = node.value

The rotation methods are straightforward: it takes a node and swaps it

to the right or to the left children node:
def left_rotate(self, x):
y = x.right
y.value = x.value
if y.value is None:
self.root = y
else:
if y.value.left is x:
y.value.left = y
elif y.value.right is x:
y.value.right = y
x.right = y.left
if x.right is not None:
x.right.value = x
y.left = x
x.value = y

def right_rotate(self, x):

y = x.left
y.value = x.value
if y.value is None:
self.root = y
else:
if y.value.left is x:
y.value.left = y
elif y.value.right is x:
y.value.right = y
x.left = y.right
if x.left is not None:
x.left.value = x
y.right = x
x.value = y

We are now ready to write the entire AVL tree class! In the following
code we have used our old BST class as a superclass, together with the
methods we have described above. In addition, two methods for traversals
12.4. SELF-BALANCING BST 189

were used, and we will explain them better in the next chapter. For now, it
is good to keep the example in mind and that this AVL tree indeed supports
insert, find, and delete-min operations at O(ln n) time:
[trees/binary_trees/avl.py]

from BST_with_Nodes import BSTwithNodes, Node

class AVL(BSTwithNodes):
def __init__(self):
self.root = None

def left_rotate(self, x):

y = x.right
y.value = x.value
if y.value is None:
self.root = y
else:
if y.value.left is x:
y.value.left = y
elif y.value.right is x:
y.value.right = y
x.right = y.left
if x.right is not None:
x.right.value = x
y.left = x
x.value = y
update_height(x)
update_height(y)

def right_rotate(self, x):

y = x.left
y.value = x.value
if y.value is None:
self.root = y
else:
if y.value.left is x:
y.value.left = y
elif y.value.right is x:
y.value.right = y
x.left = y.right
if x.left is not None:
x.left.value = x
y.right = x
x.value = y
190 CHAPTER 12. BINARY TREES

update_height(x)
update_height(y)

def insert_item(self, value):

if self.root == None:
self.root = Node(value)
else:
current = self.root
while True:
if value < current.value:
if current.left:
current = current.left
else:
current.left = Node(value)
break;

elif value > current.value:

if current.right:
current = current.right
else:
current.right = Node(value)
break;
else:
break

def insert(self, value):

node = self.insert_item(value)
self.rebalance(node)

def rebalance(self, node):

while node is not None:
update_height(node)
if height(node.left) >= 2 + height(node.right):
if height(node.left.left) >= height(node.left.right):
self.right_rotate(node)
else:
self.left_rotate(node.left)
self.right_rotate(node)
elif height(node.right) >= 2 + height(node.left):
if height(node.right.right) >=
height(node.right.left):
self.left_rotate(node)
else:
self.right_rotate(node.right)
12.4. SELF-BALANCING BST 191

self.left_rotate(node)
node = node.value

def inorder(self, node):

if node is not None:
self.inorder(node.left)
print(node.value)
self.inorder(node.right)

def preorder(self, node):

if node is not None:
print(node.value)
self.preorder(node.left)
self.preorder(node.right)

def postorder(self, node):

if node is not None:
self.postorder(node.left)
self.postorder(node.right)
print(node.value)

def height(node):
if node is None: return -1
else: return node.height

def update_height(node):
node.height = max(height(node.left), height(node.right)) + 1

def main():
tree = AVL()
tree.insert(4)
tree.insert(2)
tree.insert(6)
tree.insert(1)
tree.insert(3)
tree.insert(7)
tree.insert(5)
print(Inorder Traversal:)
tree.inorder(tree.root)

if __name__ == __main__:
main()
192 CHAPTER 12. BINARY TREES

Red-black Trees
Red-black trees are an evolution of a binary search trees that aim to keep the
tree balanced without affecting the complexity of the primitive operations.
This is done by coloring each node in the tree with either red or black and
preserving a set of properties that guarantees that the deepest path in the
tree is not longer than twice the shortest one.
Red-black trees have the following properties:

? Every node is colored with either red or black.

? All leaf (nil) nodes are colored with black; if a nodes child is missing
then we will assume that it has a nil child in that place and this nil
child is always colored black.

? Both children of a red node must be black nodes.

? Every path from a node n to a descendent leaf has the same number
of black nodes (not counting node n). We call this number the black
height of n.

Binary Heaps
Binary heaps are complete balanced binary trees. The heap property makes
it easier to maintain the structure, i.e., the balance of the tree. There is no
need to modify a structure of the tree by splitting or rotating nodes in a
heap: the only operation will be swapping parent and child nodes.
In a binary heap, the root (the smallest or largest element) is always
found in h[0]. Considering a node at index i:
i1
? the parent index is 2 ,

? the left child index is 2i + 1,

? the right child index is 2i + 2.

12.5. ADDITIONAL EXERCISES 193

12.5 Additional Exercises

Another (Complete Example) of a Binary Tree

[trees/binary_trees/binary_tree.py]

Implementation of a binary tree and its properties. For example,

the following bt:

1 ---> level 0
2 3 ---> level 1
4 5 ---> level 2
6 7 ---> level 3
8 9 ---> level 4

has the following properties:

- SIZE OR NUMBER OF NODES: n = 9

- NUMBER OF BRANCHES OR INTERNAL NODES: b = n-1 = 8
- VALUE OF ROOT = 1
- MAX_DEPTH OR HEIGHT: h = 4
- IS BALANCED? NO
- IS BST? NO
- INORDER DFT: 8, 6, 9, 4, 7, 2, 5, 1, 3
- POSTORDER DFT: 8, 9, 6, 7, 4, 5, 2, 3, 1
- PREORDER DFT: 1, 2, 4, 6, 8, 9, 7, 5, 3
- BFT: 1, 2, 3, 4, 5, 6, 7, 8, 9

from collections import deque

class NodeBT(object):
def __init__(self, item=None, level=0):
Construtor for a Node in the Tree
self.item = item
self.level = level
self.left = None
self.right = None
self.traversal = []

METHODS TO MODIFY NODES

194 CHAPTER 12. BINARY TREES

def _addNextNode(self, value, level_here=1):

Aux for self.addNode(value)
self.traversal = []
new_node = NodeBT(value, level_here)
if not self.item:
self.item = new_node
elif not self.left:
self.left = new_node
elif not self.right:
self.right = new_node
else:
self.left = self.left._addNextNode(value, level_here+1)
return self

METHODS TO PRINT/SHOW NODES ATTRIBUTES

def __repr__(self):
Private method for this class string representation
return {}.format(self.item)

def _getDFTpreOrder(self, node):

Traversal Pre-Order, O(n)
if node:
if node.item: self.traversal.append(node.item)
self._getDFTpreOrder(node.left)
self._getDFTpreOrder(node.right)
return self

def _printDFTpreOrder(self, noderoot):

Fill the pre-order traversal array
self.traversal = []
self._getDFTpreOrder(noderoot)
return self.traversal

def _getDFTinOrder(self, node):

Traversal in-Order, O(n)
if node:
self._getDFTinOrder(node.left)
if node.item: self.traversal.append(node.item)
self._getDFTinOrder(node.right)
12.5. ADDITIONAL EXERCISES 195

return self

def _printDFTinOrder(self, noderoot):

Fill the in-order traversal array
self.traversal = []
self._getDFTinOrder(noderoot)
return self.traversal

def _getDFTpostOrder(self, node):

Traversal post-Order, O(n)
if node:
self._getDFTpostOrder(node.left)
self._getDFTpostOrder(node.right)
if node.item: self.traversal.append(node.item)
return self

def _getBFT(self, node):

Traversal bft, O(n)
if node:
queue = deque()
queue.append(node)
while len(queue) > 0:
current = queue.popleft()
if current.item: self.traversal.append(current)
if current.left: queue.append(current.left)
if current.right: queue.append(current.right)
return self

def _printBFT(self, noderoot):

Fill the in-order traversal array
self.traversal = []
self._getBFT(noderoot)
return self.traversal

def _printDFTpostOrder(self, noderoot):

Fill the post-order traversal array
self.traversal = []
self._getDFTpostOrder(noderoot)
return self.traversal

def _searchForNode(self, value):

Traverse the tree looking for the node
if self.item == value: return self
else:
196 CHAPTER 12. BINARY TREES

found = None
if self.left: found = self.left._searchForNode(value)
if self.right: found = found or
self.right._searchForNode(value)
return found

def _findNode(self, value):

Find whether a node is in the tree.
if the traversal was calculated, it is just a membership
checking, which is O(1), otherwise it is necessary to
traverse
the binary tree, so best case is O(1) and worst is O(n).

if self.traversal: return value in self.traversal
else: return self._searchForNode(value)

def _isLeaf(self):
Return True if the node is a leaf
return not self.right and not self.left

def _getMaxHeight(self, level=0):

Get the max height at the node, O(n)
levelr, levell = -1, -1
if self.right:
levelr = self.right._getMaxHeight(level +1)
if self.left:
levell = self.left._getMaxHeight(level +1)
return max(levelr, levell) + 1

def _getMinHeight(self, level=0):

Get the min height at the node, O(n)
levelr, levell = -1, -1
if self.right:
levelr = self.right._getMinHeight(level +1)
if self.left:
levell = self.left._getMinHeight(level +1)
return min(levelr, levell) + 1

def _isBalanced(self):
Find whether the tree is balanced, by calculating heights
first, O(n2)
if self._getMaxHeight() - self._getMinHeight() < 2:
return False
else:
if self._isLeaf():
12.5. ADDITIONAL EXERCISES 197

return True
elif self.left and self.right:
return self.left._isBalanced() and
self.right._isBalanced()
elif not self.left and self.right:
return self.right._isBalanced()
elif not self.right and self.left:
return self.right._isBalanced()

def _isBST(self):
Find whether the tree is a BST, inorder
if self.item:
if self._isLeaf(): return True
elif self.left:
if self.left.item < self.item: return
self.left._isBST()
else: return False
elif self.right:
if self.right.item > self.item: return
self.right._isBST()
else: return False
else:
raise Exception(Tree is empty)

class BinaryTree(object):

>>> bt = BinaryTree()
>>> for i in range(1, 10): bt.addNode(i)
>>> bt.hasNode(7)
True
>>> bt.hasNode(12)
False
>>> bt.printTree()
[1, 2, 4, 6, 8, 9, 7, 5, 3]
>>> bt.printTree(pre)
[1, 2, 4, 6, 8, 9, 7, 5, 3]
>>> bt.printTree(bft)
[1, 2, 3, 4, 5, 6, 7, 8, 9]
>>> bt.printTree(post)
[8, 9, 6, 7, 4, 5, 2, 3, 1]
>>> bt.printTree(in)
[8, 6, 9, 4, 7, 2, 5, 1, 3]
198 CHAPTER 12. BINARY TREES

>>> bt.hasNode(9)
True
>>> bt.hasNode(11)
False
>>> bt.isLeaf(8)
True
>>> bt.getNodeLevel(1)
0
>>> bt.getNodeLevel(8)
4
>>> bt.getSizeTree()
9
>>> bt.isRoot(10)
False
>>> bt.isRoot(1)
True
>>> bt.getHeight()
4
>>> bt.isBST()
False
>>> bt.isBalanced()
False
>>> bt.isBalanced(2)
False
>>> bt.getAncestor(8, 5)
2
>>> bt.getAncestor(8, 5, pre-post)
2
>>> bt.getAncestor(8, 5, post-in)
2

def __init__(self):
Construtor for the Binary Tree, which is a container of
Nodes
self.root = None

METHODS TO MODIFY THE TREE

def addNode(self, value):

12.5. ADDITIONAL EXERCISES 199

Add new node to the tree, by the left first, O(n)

if not self.root: self.root = NodeBT(value)
else: self.root._addNextNode(value)

METHODS TO PRINT/SHOW TREES ATTRIBUTES

def __repr__(self):
Private method for this class string representation
return {}.format(self.item)

def printTree(self, order = pre):

Print Tree in the chosen order
if self.root:
if order == pre: return
self.root._printDFTpreOrder(self.root)
elif order == in: return
self.root._printDFTinOrder(self.root)
elif order == post: return
self.root._printDFTpostOrder(self.root)
elif order == bft: return
self.root._printBFT(self.root)
else: raise Exception(Tree is empty)

def hasNode(self, value):

Verify whether the node is in the Tree
return bool(self.root._findNode(value))

def isLeaf(self, value):

Return True if the node is a Leaf
node = self.root._searchForNode(value)
return node._isLeaf()

def getNodeLevel(self, item):

Return the level of the node, best O(1), worst O(n)
node = self.root._searchForNode(item)
if node: return node.level
else: raise Exception(Node not found)

def getSizeTree(self):
Return how many nodes in the tree, O(n)
return len(self.root._printDFTpreOrder(self.root))

def isRoot(self, value):

200 CHAPTER 12. BINARY TREES

Return the root of the tree

return self.root.item == value

def getHeight(self):
Returns the height/depth of the tree, best/worst O(n)
return self.root._getMaxHeight()

def isBalanced(self, method=1):

Return True if the tree is balanced
if method == 1:
O(n2)
return self.root._isBalanced()
else:
O(n)
return self.root._isBalancedImproved()

def isBST(self, method=1):

Return True if the tree is BST
return self.root._isBST()

def _getAncestorPreIn(self, preorder, inorder, value1, value2):

Return the ancestor of two nodes with pre and in
root = preorder[0]
preorder = preorder[1:]
i = 0
item = inorder[0]
value1left, value2left = False, False
while item != root and i < len(inorder):
if item == value1: value1left = True
elif item == value2: value2left = True
i += 1
item = inorder[i]
if (value1left and not value2left) or (value2left and not
value1left):
return root
else:
return self._getAncestorPreIn(preorder, inorder[:i] +
inorder[i+1:], value1, value2)

def _getAncestorPrePost(self, preorder, postorder, value1,

value2):
Return the ancestor of two nodes with pre and post
root = preorder[0]
preorder = preorder[1:]
postorder = postorder[:-1]
12.5. ADDITIONAL EXERCISES 201

value1right, value2right = False, False

i = len(postorder)-1
itempre = preorder[0]
itempos = postorder[i]
while itempre != itempos and i > 0:
if itempos == value1: value1right = True
elif itempos == value2: value2right = True
i -= 1
itempos = postorder[i]

if (value1right and not value2right) or (value2right and not

value1right):
return root
else:
return self._getAncestorPrePost(preorder, postorder[:i] +
postorder[i+1:], value1, value2)

def _getAncestorInPost(self, inorder, postorder, value1,

value2):
Return the ancestor of two nodes with in and post
root = postorder[-1]
postorder = postorder[:-1]
value1left, value2left = False, False
i = 0
item = inorder[i]
while item != root and i < len(inorder):
if item == value1: value1left = True
elif item == value2: value2left = True
i += 1
item = inorder[i]

if (value1left and not value2left) or (value2left and not

value1left):
return root
else:
return self._getAncestorInPost(postorder, inorder[:i] +
inorder[i+1:], value1, value2)

def _getAncestorBST(self, preorder, value1, value2):

Return the ancestor of two nodes if it is a bst
while preorder:
current = preorder[0]
if current < value1:
try: preorder = preorder[2:]
except: return current
202 CHAPTER 12. BINARY TREES

elif current > value2:

try: preorder = preorder[1:]
except: return current
elif value1 <= current <= value2:
return current
return None

def getAncestor(self, value1, value2, method=pre-in):

Return the commom ancestor for two nodes
if method == pre-in:
Using pre and inorder, best/worst O(n)
preorder = self.root._printDFTpreOrder(self.root)
inorder = self.root._printDFTinOrder(self.root)
return self._getAncestorPreIn(preorder, inorder, value1,
value2)
if method == pre-post:
Using pre and postorder, best/worst O(n)
preorder = self.root._printDFTpreOrder(self.root)
postorder = self.root._printDFTpostOrder(self.root)
return self._getAncestorPrePost(preorder, postorder,
value1, value2)
if method == post-in:
Using in and postorder, best/worst O(n)
inorder = self.root._printDFTinOrder(self.root)
postorder = self.root._printDFTpostOrder(self.root)
return self._getAncestorInPost(inorder, postorder,
value1, value2)
if method == bst:
if self.isBST():
preorder = self.root._printDFTpreOrder(self.root)
return self._getAncestorBST(preorder, value1, value2)
else:
return Exception(The tree is not a BST)

if __name__ == __main__:
import doctest
doctest.testmod()

Another (Complete Example) of a Binary Search Tree

[trees/binary_trees/binary_search_tree.py]
12.5. ADDITIONAL EXERCISES 203

from binary_tree import BinaryTree, NodeBT

Implementation of a binary search tree and its properties.

We use the Binary Tree class and its Node class as superclasses,
and we modify the methods
that are needeed to create a BST (polymorphism). For example,
the following bst:

7 ---> level 0
4 9 ---> level 1
2 5 8 10 ---> level 2
1 6 ---> level 3

has the following properties:

- SIZE OR NUMBER OF NODES: n = 10

- NUMBER OF BRANCHES OR INTERNAL NODES: b = n-1 = 9
- VALUE OF ROOT = 7
- MAX_DEPTH OR HEIGHT: h = 3
- IS BALANCED? YES
- IS BST? YES
- INORDER DFT: 1, 2, 4, 5, 6, 7, 8, 9, 10
- POSTORDER DFT: 1, 2, 6, 5, 4, 8, 10, 9, 7
- PREORDER DFT: 7, 4, 2, 1, 5, 6, 9, 8, 10
- BFT: 7, 4, 9, 2, 5, 8, 10, 1, 6

class NodeBST(NodeBT):

def _addNextNode(self, value, level_here=1):

Aux for self.addNode(value): for BST, best O(1), worst
O(log n)
self.traversal = []
new_node = NodeBST(value, level_here)
if not self.item:
self.item = new_node
elif value < self.item:
self.left = self.left and self.left._addNextNode(value,
level_here+1) or new_node
else:
self.right = self.right and
self.right._addNextNode(value, level_here+1) or
new_node
return self
204 CHAPTER 12. BINARY TREES

def _searchForNode(self, value):

Traverse the tree looking for the node. For BST it is
O(logn)
if self.item == value: return self
elif value > self.item and self.right: return
self.right._findNode(value)
elif value < self.item and self.left: return
self.left._findNode(value)
return None

class BinarySearchTree(BinaryTree):

>>> bst = BinarySearchTree()
>>> l1 = [7, 4, 5, 9, 2, 8, 1, 6, 10]
>>> for i in l1: bst.addNode(i)
>>> bst.hasNode(3)
False
>>> bst.hasNode(10)
True
>>> bst.printTree(pre)
[7, 4, 2, 1, 5, 6, 9, 8, 10]
>>> bst.printTree(post)
[1, 2, 6, 5, 4, 8, 10, 9, 7]
>>> bst.printTree(in)
[1, 2, 4, 5, 6, 7, 8, 9, 10]
>>> bst.printTree(bft)
[7, 4, 9, 2, 5, 8, 10, 1, 6]
>>> bst.getHeight()
3
>>> bst.isBST()
True
>>> bst.isBalanced()
False
>>> bst.isBalanced(2)
False
>>> bst.getAncestor(2, 9)
7
>>> bst.getAncestor(2, 9, bst)
7
>>> bst.getAncestor(2, 9, pre-post)
7
>>> bst.getAncestor(2, 9, post-in)
12.5. ADDITIONAL EXERCISES 205

def addNode(self, value):

Add new node to the tree, by the left first, O(n).
The only difference from the Binary Tree class is the
node class is
NodeBST and not NodeBT
if not self.root: self.root = NodeBST(value)
else: self.root._addNextNode(value)

if __name__ == __main__:
import doctest
doctest.testmod()
206 CHAPTER 12. BINARY TREES
Chapter 13

Traversals and Problems on

Graphs and Trees

Traversals are algorithms used to visit the objects (nodes) in some connected
structure, such as a tree or a graph. Traversal problems can be either visiting
every node or visiting only some specific nodes.

13.1 Depth-First Search

Depth-first traversal, or depth-first search (DFS), are algorithms that searches
deeper first in a graph or a tree. Their difference when in graphs or trees is
that in case of graphs, it is necessary to mark nodes as visited (otherwise
we might be stuck in a loop).
DFS algorithms are called once for every node that is reachable from the
start node, looking at its successors. The runtime is O(number of reachable

Figure 13.1: Binary tree traversals: preorder, inorder, postorder, and

breath-first search.

207
208CHAPTER 13. TRAVERSALS AND PROBLEMS ON GRAPHS AND TREES

nodes + total number of outgoing edges from these nodes) = O(V + E).
DFSs are usually implemented using LIFO structure such as stacks to keep
track of the discovered nodes, and they can be divided in three different
strategies:

Preorder: Visit a node before traversing subtrees (root left right):

def preorder(root):
if root != 0:
yield root.value
preorder(root.left)
preorder(root.right)

Postorder: Visit a node after traversing all subtrees (left right root):
def postorder(root):
if root != 0:
postorder(root.left)
postorder(root.right)
yield root.value

Inorder: Visit a node after traversing its left subtree but before the right
subtree (left root right):
def inorder(root):
if root != 0:
inorder(root.left)
yield root.value
inorder(root.right)

13.2 Breadth-First Search

Breadth-first traversal, or breath-first search (BFS), are algorithms that yields
the values of all nodes of a particular depth before going to any deeper node.
Problems that use BFS usually ask to find the fewest number of steps
(or the shortest path) needed to reach a certain end point from the starting
13.3. REPRESENTING TREE TRAVERSALS 209

point. Traditionally, BFSs are implemented using a list to store the values
of the visited nodes and then a FIFO queue to store those nodes that have
yet to be visited. The total runtime is also O(V + E).

13.3 Representing Tree Traversals

There are many ways we could write traversals. In the following code we
use the BST with nodes class, defined in the last chapter, to implement
each of the traversal algorithms. For the DFS cases, we have also tested two
different methods:
[trees/traversals/BST_with_Nodes_traversal.py]

from BST_with_Nodes import BSTwithNodes, Node

class BSTTraversal(BSTwithNodes):
def __init__(self):
self.root = None
self.nodes_BFS = []
self.nodes_DFS_pre = []
self.nodes_DFS_post = []
self.nodes_DFS_in = []

def BFS(self):
self.root.level = 0
queue = [self.root]
current_level = self.root.level

while len(queue) > 0:

current_node = queue.pop(0)
if current_node.level > current_level:
current_level += 1
self.nodes_BFS.append(current_node.value)

if current_node.left:
current_node.left.level = current_level + 1
queue.append(current_node.left)

if current_node.right:
current_node.right.level = current_level + 1
queue.append(current_node.right)
210CHAPTER 13. TRAVERSALS AND PROBLEMS ON GRAPHS AND TREES

return self.nodes_BFS

def DFS_inorder(self, node):

if node is not None:
self.DFS_inorder(node.left)
self.nodes_DFS_in.append(node.value)
self.DFS_inorder(node.right)
return self.nodes_DFS_in

def DFS_preorder(self, node):

if node is not None:
self.nodes_DFS_pre.append(node.value)
self.DFS_preorder(node.left)
self.DFS_preorder(node.right)
return self.nodes_DFS_pre

def DFS_postorder(self, node):

if node is not None:
self.DFS_postorder(node.left)
self.DFS_postorder(node.right)
self.nodes_DFS_post.append(node.value)
return self.nodes_DFS_post

def main():
tree = BSTTraversal()
l1 = [10, 5, 15, 1, 6, 11, 50]
for i in l1: tree.insert(i)

print(Breadth-First Traversal: , tree.BFS())

print(Inorder Traversal: , tree.DFS_inorder(tree.root))
print(Preorder Traversal: , tree.DFS_preorder(tree.root))
print(Postorder Traversal: , tree.DFS_postorder(tree.root))

if __name__ == __main__:
main()
13.4. ADDITIONAL EXERCISES 211

13.4 Additional Exercises

More Traversals in a BST

As an alternative to the example shown in this chapter, in the following

program we implement traversals for our old BST class (without nodes):
[traversals/BST_traversal.py]

from BST import BST

class TranversalBST(object):
def __init__(self):
self.bst = BST(None)
self.nodes = []

def insert(self, value):

if not self.bst.value:
self.bst.value = value
else:
self.bst.insert(value)

def contains(self, value):

return bool(self.bst.find(value))

def get(self, index):

for i, value in enumerate(self.inorder()):
if i == index:
return value

def inorder(self):
current = self.bst
self.nodes = []
stack = []
while len(stack) > 0 or current is not None:
if current is not None:
stack.append(current)
current = current.left
else:
current = stack.pop()
self.nodes.append(current.value)
current = current.right
return self.nodes

def preorder(self):
212CHAPTER 13. TRAVERSALS AND PROBLEMS ON GRAPHS AND TREES

self.nodes = []
stack = [self.bst]
while len(stack) > 0:
curr = stack.pop()
if curr is not None:
self.nodes.append(curr.value)
stack.append(curr.right)
stack.append(curr.left)
return self.nodes

def preorder2(self):
self.nodes = []
current = self.bst
stack = []
while len(stack) > 0 or current is not None:
if current is not None:
self.nodes.append(current.value)
stack.append(current)
current = current.left
else:
current = stack.pop()
current = current.right
return self.nodes

def main():
"""
10
5 15
1 6 11 50
"""
t = TranversalBST()
t.insert(10)
t.insert(5)
t.insert(15)
t.insert(1)
t.insert(6)
t.insert(11)
t.insert(50)
print(t.preorder())
print(t.preorder2())
print(t.inorder())

if __name__ == __main__:
13.4. ADDITIONAL EXERCISES 213

main()

Balance and Depth in a BST

In the following example we use the class in the previous example with some
methods to find the (maximum and minimum) depths, to check whether the
tree is balanced and to find a key in preorder and inorder traversals:
[trees/traversals/BST_extra_methods.py]

from BST_traversal import TranversalBST

from BST import BST

class BSTwithExtra(TranversalBST):
def __init__(self):
self.bst = BST(None)
self.nodes = []

def get_inorder(self, k):

for i, value in enumerate(self.inorder()):
if value == k:
return i

def get_preorder(self, k):

for i, value in enumerate(self.preorder()):
if value == k:
return i

def is_balanced(self, threshold=1):

maxd = self.get_max_depth()
mind = self.get_min_depth()
print(Max depth: + str(maxd))
print(Min depth: + str(mind))
return maxd -mind

def get_min_depth(self, node=None, initial=1):

if node is None and initial == 1:
node = self.bst
if node.left and node.right:
return 1 + min(self.get_min_depth(node.left, 0),
self.get_min_depth(node.right, 0))
else:
214CHAPTER 13. TRAVERSALS AND PROBLEMS ON GRAPHS AND TREES

if node.left:
return 1 + self.get_max_depth(node.left, 0)
elif node.right:
return 1 + self.get_max_depth(node.right, 0)
else:
return 0

def get_max_depth(self, node=None, initial=1):

if node is None and initial == 1:
node = self.bst
if node.left and node.right:
return 1 + max(self.get_max_depth(node.left, 0),
self.get_max_depth(node.right, 0))
else:
if node.left:
return 1 + self.get_max_depth(node.left, 0)
elif node.right:
return 1 + self.get_max_depth(node.right, 0)
else:
return 0

def main():
"""
10
5 15
1 6 11 50
60
70
80
"""
t = BSTwithExtra()
l1 = [10, 5, 15, 1, 6, 11, 50, 60, 70, 80]
for i in l1: t.insert(i)
print(t.inorder())
print(t.preorder())
assert(t.get_max_depth() == 5)
assert(t.get_min_depth() == 2)
assert(t.is_balanced() == 3)
assert(t.get_inorder(10) == 3)
assert(t.get_preorder(10) == 0)

"""
1
2 3
4 5 6 7
13.4. ADDITIONAL EXERCISES 215

"""
t2 = BSTwithExtra()
l2 = [1, 2, 3, 4, 5, 6, 7, 8]
for i in l2: t2.insert(i)
print(t2.inorder())
print(t2.preorder())
assert(t2.is_balanced() == 0)

print("Tests Passed!")

if __name__ == __main__:
main()

Ancestor in a BST
The example bellow finds the lowest level common ancestor of two nodes in
a binary search tree:
[trees/traversals/BST_ancestor.py]

from BST_traversal import TranversalBST

def find_ancestor(path, low_value, high_value):

find the lowest ancestor in a BST
while path:
current_value = path[0]
if current_value < low_value:
try:
path = path[2:]
except:
return current_value
elif current_value > high_value:
try:
path = path[1:]
except:
return current_value
elif low_value <= current_value <= high_value:
return current_value
return None

def test_find_ancestor():
"""
216CHAPTER 13. TRAVERSALS AND PROBLEMS ON GRAPHS AND TREES

10
5 15
1 6 11 50
"""
t = TranversalBST()
l1 = [10, 5, 15, 1, 6, 11, 50]
for i in l1: t.insert(i)
path = t.preorder()
assert(find_ancestor(path, 1, 6) == 5)
assert(find_ancestor(path, 1, 11) == 10)
assert(find_ancestor(path, 11, 50) == 15)
assert(find_ancestor(path, 5, 15) == 10)

print("Tests passsed!")

if __name__ == __main__:
test_find_ancestor()
Bibliography

Websites:
[Interactive Python] http://interactivepython.org

[Pythons Official Documentation] http://docs.python.org/3/tutorial/

[Google Style Guide] http://google-styleguide.code.com/svn/trunk/pyguide.html

[My free GIT Repository] https://bitbucket.org/steinkirch/exercises-in-python:

All the examples in this book are there, plus many exercises that I have
not included here.

[XKCD Comics] https://xkcd.com/

[Big-O Sheet] http://bigocheatsheet.com/

Books:

[A nice Book for Software Eng. Interviews] Cracking the Coding In-
terview, Gayle Laakmann McDowell, 2013

217
218 BIBLIOGRAPHY

[A nice Python 3 Book] Programming in Python 3: A Complete Intro-

duction to the Python 3.1 Language, Mark Summerfield, 2011

[A nice Python Book] Learn Python The Hard Way, Zed A. Shaw, 2010

[A nice Algorithms Book] Mastering Basic Algorithms in the Python

Language, Magnus Lie Hetland, 2010

[Another nice Algorithms Book] The Algorithm Design Manual, S.S.

Skiena, 2008

[Another nice Python Book] Python 3 Object Oriented Programming,

Dusty Phillips, 2010

[Another nice guide for Software Eng. Interviews] Programming

Pearls, Jon Bentley, 1986