Python Review Session

CS224N - Winter 25
Stanford University
1
Two entwined snakes, based on Mayan representations.
However, named after Monty Python’s Flying Circus 😅
2
Charting a Course

1 2 3

Why Python? Setting Up Python Basics

4 5 6

Data Structures Numpy Practical Tips

3
Charting a Course

1 2 3

Why Python? Setting Up Python Basics

4 5 6

Data Structures Numpy Practical Tips

4
Why Python?

● Widely used, general purpose

● Easy to learn, read, and write

● Scientific computation functionality

similar to Matlab and Octave

● Used by major deep learning

frameworks (PyTorch, TensorFlow)

● Active open-source, many libraries!

5
The Python Interpreter

Ex. Interactive Mode (line-by-line) Ex. Script Mode (.py file)

Python code → interpreted into bytecode (.pyc) → compiled by a VM implementation

into machine instructions (most commonly using C.)

“Slower”, but can run highly optimized C/C++ subroutines to make operations fast 6
 Language Basics

Strongly Interpreter always “respects” the types of each variable.

Typed Interpreter keeps track of all variable types (strict handling)

Types will not Cases like float and int

1 + ‘1’ → Error! addition are allowed by
be coerced
silently like in explicit implementation
[1, 2] + set([3]) → Error!
JavaScript, Perl (no auto conversion)

https://wiki.python.org/moin/Why%20is%20Python%20a%20dynamic%20language%20and%20also%20a%20strongly%20typed%20language 7
Language Basics

Dynamically A variable is simply a value or object reference bound to a name.

Typed Data types of variables are determined at runtime (flexible!)

def find(required_element, sequence):

Variables can be
for index, element in enumerate(sequence):
if element == required_element:
assigned to values
return index
return -1
of a different type.

print(find(2, [1, 2, 3])) # Outputs: 1

print(find("c", ("a", "b", "c", "d"))) # Outputs: 2 ✅ num = 1 # int
num = "One" # str ✅

https://medium.com/@pavel.loginov.dev/typing-in-python-strong-dynamic-implicit-c3512785b863
8
A Quick Check-In 🥳

🎯 In Python, what will the

following code output? A. 8
B. "53"
x = 5
y = "3" C. TypeError
print(x + y) D. "53.0"

9
A Quick Check-In 🥳

🎯 In Python, what will the

following code output? A. 8
B. "53"
x = 5
y = "3" C. TypeError
print(x + y) D. "53.0"

10
Charting a Course

1 2 3

Why Python? Setting Up Python Basics

4 5 6

Data Structures Numpy Practical Tips

11
Syntax Going Forward

Code is in Courier New.

Command line input is prefixed with ‘$’.

Output is prefixed with ‘>>’.

12
Python Installation

https://www.python.org/downloads/ 🥳

13
Helpful Commands See Installed Libraries pip is Python’s
$python -m pip list package installer
Print out Version -m runs a module (ex. pip) as a script
$python --version
$python -v
Run in Different Modes
$python -vv
$python script.py
-i remains in interactive
$python -i script.py
Print out Location mode after running .py
$which python (mac, linux)
$where python (windows) $python -c “print(‘hello there!’)”

-c runs one-liner code snippet

14
Environment Management

Problem
● Different versions of Python
● Countless Python packages
and their dependencies

● Different projects require

different packages → even
worse, different versions of
the same package!

15
Environment Management

Problem Solution: Virtual Envs

● Different versions of Python ● Keep multiple Python environments
● Countless Python packages that are isolated from each other
and their dependencies
● Each environment
● Different projects require ○ Can use different Python version
different packages → even ○ Keeps its own set of packages
worse, different versions of (can specify package versions)
the same package! ○ Can be easily replicated

16
 Solution 1: venv
● Created on top of existing $python -m venv /path/to/new/virtual/env
installation, known as the
Creates a new directory → can activate (differs based on OS)
virtual env’s “base” Python
● Directory contains a specific
Python interpreter and
libraries, binaries which are
needed to support a project
● Isolated from software in other
virtual envs and interpreters
and libraries installed in OS

https://docs.python.org/3/library/venv.html 17
Solution 2: Anaconda (or Miniconda)
https://www.anaconda.com/download/
Choose specific
Very popular Python Basic Workflow Python version
env/package manager Create a new environment
$ conda create –n <environment_name>
● Supports Windows, $ conda create -n <environment_name> python=3.7
Linux, MacOS $ conda env create -f <environment.yml>
Activate/deactivate environment
● Can create and $ conda activate <environment_name>
<...do stuff...>
manage different Export/create
$ conda deactivate from env files!
isolated envs Export environment
$ conda activate <environment_name>
$ conda env export > environment.yml
18
Installing Packages
pip installs only Python packages, conda installs packages which may contain software written in any language

🚨 Best to first use conda to install as many packages as possible and use pip to install remaining packages after.

conda install -n myenv [package_name][=optional version number]

Install packages using pip in a conda environment (necessary when package not available through conda):

conda install -n myenv pip # Install pip in environment

conda activate myenv # Activate environment

pip install # Install package individually OR

[package_name][==optional version number]

pip install -r <requirements.txt> # Install packages from file

19
IDEs / Text Editors
Write a Python program in your
IDE or text editor of choice 😁
● PyCharm
● Visual Studio Code
● Sublime Text
● Atom
● Vim (for Linux or Mac) IDEs often have useful extensions! (ex. VS Code)

In terminal, just activate virtual

environment and run command:
$ python <filename.py>
20
Python Notebooks https://colab.research.google.com/

Jupyter Notebook Google Colab

● .ipynb → write and execute ● Hosted Jupyter notebooks, run in

Python locally in web browser cloud, requires no setup to use,
provides free access to GPUs
● Interactive, re-execute code, ● Comes with many Python
result storage, can interleave libraries pre-installed
text, equations, and images
● Can integrate with Git (pull/run),
● Can add conda environments Google Drive, local storage
● Read-Eval-Print-Loop (REPL) ● Tools > Settings > Misc > 😉😁

21
🎯 Matching time! A. Python package manager used to
install and manage libraries.

B. Tool for creating isolated Python

1. venv environments for dependency
management.
2. Anaconda C. Distribution that simplifies package
and environment management,
3. Jupyter Ntbk designed for data science.

D. An interactive platform for writing

4. pip and running code alongside
visualizations and notes. 22
🎯 Matching time! A. Python package manager used to
install and manage libraries.

B. Tool for creating isolated Python

1. venv environments for dependency
management.
2. Anaconda C. Distribution that simplifies package
and environment management,
3. Jupyter Ntbk designed for data science.

D. An interactive platform for writing

4. pip and running code alongside
visualizations and notes. 23
Language Basics

1 2 3

Why Python? Setting Up Python Basics

4 5 6

Data Structures Numpy Practical Tips

24
Common Operations
x = 10 # Declaring two integer variables

y = 3 # Comments start with hash

x + y >> 13 # Arithmetic operations

x ** y >> 1000 # Exponentiation

x / y >> 3 # Dividing two integers

x / float(y) >> 3.333… # Type casting for float division

str(x) + “+” >> “10 + 3” # Casting integer as string and

+ str(y) string concatenation
25
Built-in Values
True, False # Usual true/false values

None # Represents the absence of something

x = None # Variables can be assigned None

array = [1, 2, None] # Lists can contain None

# Functions can return None
def func():

return None

26
Built-in Values
and # Boolean operators in Python written
as plain English, as opposed to &&,
or ||, ! in C++
not

if [] != [None]: # Comparison operators == and !=

check for equality/inequality, return
print(“Not equal”) true/false values

27
Spacing: Brackets → Indents

Code blocks are created using indents and newlines, instead of brackets like in C++

● Indents can be 2 or 4 def sign(num):

# Indent level 1: function body
spaces, but should be if num == 0:
consistent throughout # Indent level 2: if statement body
print(“Zero”)
elif num > 0:
● If using Vim, set this # Indent level 2: else if statement body
print(“Positive”)
value to be consistent else:
in your .vimrc # Indent level 2: else statement body
print(“Negative”)

28
 🎯 Debugging Derby

0length = 10
float width = 5.0

print "Beginning work..."

area = 0length * Width Find the errors!

if area > 20
print("Area: " + area)

message = "Completed!'

29
 🎯 Debugging Derby

0length = 10 # can’t start var name with number

float width = 5.0 # no explicit type declaration!

print "Beginning work..." # parentheses around print

area = 0length * Width # capitalization mismatch “Width”

if area > 20 # missing colon after condition

print("Area: " + area) # need to cast area to string type

message = "Completed!' # mismatch in quotation (“ vs ‘)

30
 🎯 Debugging Derby

length = 10
width = 5.0

print("Beginning work...")

area = length * width All fixed!🥳

if area > 20:
print("Area: " + str(area))

message = "Completed!"

31
Language Basics

1 2 3

Why Python? Setting Up Python Basics

4 5 6

Data Structures Numpy Practical Tips

32
Collections: List
Lists are mutable arrays (think std::vector).

names = [‘Zach’, ‘Jay’]

names[0] == ‘Zach’
names.append(‘Richard’)
print(len(names) == 3) >> True
print(names) >> [‘Zach’, ‘Jay’, ‘Richard’]
names += [‘Abi’, ‘Kevin’]
print(names) >> [‘Zach’, ‘Jay’, ‘Richard’, ‘Abi’, ‘Kevin’]
names = [] # Creates an empty list
names = list() # Also creates an empty list
stuff = [1, [‘hi’,’bye’], -0.12, None] # Can mix types

33
List Slicing
List elements can be accessed in convenient ways.
Basic format: some_list[start_index:end_index]

numbers = [0, 1, 2, 3, 4, 5, 6]
numbers[0:3] == numbers[:3] == [0, 1, 2]
numbers[5:] == numbers[5:7] == [5, 6]
numbers[:] == numbers == [0, 1, 2, 3, 4, 5, 6]
numbers[-1] == 6 # Negative index wraps around
numbers[-3:] == [4, 5, 6]
numbers[3:-2] == [3, 4] # Can mix and match

34
Collections: Tuples
Tuples are immutable arrays.

names = (‘Zach’, ‘Jay’) # Note the parentheses

names[0] == ‘Zach’
print(len(names) == 2) >> True
print(names) >> (‘Zach’, ‘Jay’)
names[0] = ‘Richard’ >> TypeError: 'tuple' object does not
support item assignment
empty = tuple() # Empty tuple
single = (10,) # Single-element tuple. Comma matters!

35
Collections: Dictionary
Dictionaries are hash maps.

phonebook = {} # Empty dictionary

phonebook = dict() # Also creates an empty dictionary
phonebook = {‘Zach’: ‘12-37’} # Dictionary with one item
phonebook[‘Jay’] = ‘34-23’ # Add another item
print(‘Zach’ in phonebook) >> True
print(‘Kevin’ in phonebook) >> False
print(phonebook[‘Jay’]) >> ‘34-23’
del phonebook[‘Zach’] # Delete an item
print(phonebook) >> {‘Jay’ : ‘34-23’}

36
Loops
For loop syntax in Python
Instead of for (i=0; i<10; i++) syntax in languages like C++, use range()

for i in range(10):
print(i)
>> 0
1…
8
9

37
Loops
To iterate over a list
names = [‘Zach’, ‘Jay’, ‘Richard’] >> Hi Zach!
for name in names: Hi Jay!
print(‘Hi ‘ + name + ‘!’) Hi Richard!

To iterate over indices and values

# One way >> 1 Zach
for i in range(len(names)): 2 Jay
print(i, names[i]) 3 Richard

# A different way
for i, name in enumerate(names):
print(i, name)
38
Loops
To iterate over a dictionary
phonebook = {‘Zach’: ‘12-37’, ‘Jay’: ‘34-23’}
for name in phonebook: >> Jay
print(name) Zach

for number in phonebook.values(): >> 12-37

print(number) 34-23

for name, number in phonebook.items(): >> Zach 12-37

print(name, number) Jay 34-23

Note: Whether dictionary iteration order is guaranteed depends on the version of Python. 39
Classes
class Animal(object): # Constructor `a =
def __init__(self, species, age): Animal(‘human’, 10)`
# Refer to instance with `self`
self.species = species
# Instance variables are public
self.age = age

def is_person(self):
# Invoked with `a.is_person()`
return self.species

def age_one_year(self):
self.age += 1

class Dog(Animal): # Inherits Animal’s methods

def age_one_year(self): # Override for dog years
self.age += 7
40
Model Classes
In the later assignments, you’ll see and write model classes in PyTorch that inherit from
torch.nn.Module, the base class for all neural network modules.

import torch.nn as nn

class Model(nn.Module):
def __init__():
…

def forward():
…
41
🎯 Inner Interpreter

v1 = ["Eeyore", "Goofy", "Nemo", "Wall-E"]

v2 = {"Eeyore": 12, "Nemo": 2, "Goofy": 42}

m1 = v1[1:-1]

for n in m1:
Output?
print(f"{n} is {v2[n]} years old.")

42
🎯 Inner Interpreter

v1 = ["Eeyore", "Goofy", "Nemo", "Wall-E"]

v2 = {"Eeyore": 12, "Nemo": 2, "Goofy": 42}

m1 = v1[1:-1]

for n in m1:
print(f"{n} is {v2[n]} years old.")

>> Goofy is 42 years old.

>> Nemo is 2 years old.
43
Language Basics

1 2 3

Why Python? Setting Up Python Basics

4 5 6

Data Structures Numpy Practical Tips

44
Prelude: Importing Package Modules
# Import ‘os’ and ‘time’ modules
import os, time

# Import under an alias

import numpy as np
np.dot(x, y) # Access components with pkg.fn

# Import specific submodules/functions

from numpy import linalg as la, dot as matrix_multiply
# Can result in namespace collisions...

45
Now, NumPy!

● NumPy: Optimized library for matrix and vector computation

● Makes use of C/C++ subroutines and memory-efficient data structures
○ Lots of computation can be efficiently represented as vectors

This is the data type that you will use to

Main data type represent matrix/vector computations.
np.ndarray
Note: constructor function is np.array()

On average, a task in Numpy is 5-100X faster than standard list!

46
np.ndarray
x = np.array([1,2,3]) >> [1 2 3]
y = np.array([[3,4,5]]) [[3 4 5]]
z = np.array([[6,7],[8,9]]) [[6 7]
print(x,y,z) [8 9]]

print(x.shape) >> (3,) A 1-D vector!

print(y.shape) >> (1,3) A (row) vector!

print(z.shape) >> (2,2) A matrix!

Note: shape (N,) != (1, N) != (N, 1) 47

np.ndarray Operations
Reductions: np.max, np.min, np.amax, np.sum, np.mean,...

# shape: (3, 2) tl;dr “collapsing”

Always reduces x = np.array([[1,2],[3,4], [5, 6]]) this axis into the
along an axis. func’s output.
# shape: (3,)
Or will reduce
print(np.max(x, axis = 1)) >> [2 4 6]
along all axes if
not specified. # shape: (3, 1)

print(np.max(x, axis = 1, keepdims = True)) >> [[2] [4] [6]]

48
np.ndarray Operations
Infix operators (i.e. +, -, *, **, /) are element-wise.

Element-wise product Matrix product /

(Hadamard product) of A * B multiplication of np.matmul(A, B)
matrix A and B, A ᐤ B, is: matrix A and B is: or A @ B

Dot product is: np.dot(u, v) np.dot()can also be used, but if A and B are both
2-D arrays, np.matmul() is preferred.
Matrix vector
product (1-D np.dot(x, W) Transpose is: x.T
array vectors) is:

Note: SciPy and np.linalg have many, many other advanced functions that are very useful! 🥳 49
Indexing
x = np.random.random((3, 4)) # Random (3,4) matrix

x[:] # Selects everything in x

x[np.array([0, 2]), :] # Selects the 0th and 2nd rows

x[1, 1:3] # Selects 1st row as 1-D vector

# and 1st through 2nd elements

x[x > 0.5] # Boolean indexing

x[:, :, np.newaxis] # 3-D vector of shape (3, 4, 1)

Note: Selecting with an ndarray or range will preserve the dimensions of the selection. 50
Broadcasting
x = np.random.random((3, 4)) # Random (3, 4) matrix

y = np.random.random((3, 1)) # Random (3, 1) vector

z = np.random.random((1, 4)) # Random (1, 4) vector

x + y # Adds y to each column of x

x * z # Multiplies z (element-wise) with each row of x

Note: If you’re getting an error, print the shapes of the matrices and investigate from there.
51
Broadcasting (visually)

1 2 3 4 1 1 1 1 2 3 4 5

5 6 7 8 2 2 2 2 7 8 9 10

9 10 11 12 3 3 3 3 12 13 14 15

x + y

1 2 3 4 1 2 3 4 1 4 9 16

5 6 7 8 1 2 3 4 5 12 21 32

9 10 11 12 1 2 3 4 9 30 33 48

x * z
52
Broadcasting (generalized)
When operating on two arrays, NumPy compares their shapes element-wise. It starts with the trailing (i.e.
rightmost) dimensions and works its way left. Two dimensions are compatible when

1. they are equal, or

2. one of them is 1 (in which case, elements on the axis are repeated along the dimension)

a = np.random.random((3, 4)) # Random (3, 4) matrix

b = np.random.random((3, 1)) # Random (3, 1) vector
c = np.random.random((3, )) # Random (3, ) vector

What do the following operations give us? What are the resulting shapes?
b + b.T
a + c If the arrays have different ranks (number of dimensions), NumPy
b + c implicitly prepends 1s to the shape of the lower-rank array.
53
Broadcasting (generalized)
When operating on two arrays, NumPy compares their shapes element-wise. It starts with the trailing (i.e.
rightmost) dimensions and works its way left. Two dimensions are compatible when

1. they are equal, or

2. one of them is 1 (in which case, elements on the axis are repeated along the dimension)

a = np.random.random((3, 4)) # Random (3, 4) matrix

b = np.random.random((3, 1)) # Random (3, 1) vector
c = np.random.random((3, )) # Random (3, ) vector

What do the following operations give us? What are the resulting shapes?
b + b.T → (3, 3)
a + c → Broadcast Error If the arrays have different ranks (number of dimensions), NumPy
b + c → (3, 3) implicitly prepends 1s to the shape of the lower-rank array.
54
Broadcasting Algorithm
p = max(m, n)
if m < p:
left-pad A's shape with 1s until it also has p dimensions
else if n < p:
left-pad B's shape with 1s until it also has p dimensions

result_dims = new list with p elements

for i in p-1 ... 0:

A_dim_i = A.shape[i]; B_dim_i = B.shape[i]
if A_dim_i != 1 and B_dim_i != 1 and A_dim_i != B_dim_i:
raise ValueError("could not broadcast")
else:
# Pick the Array which is having maximum Dimension
result_dims[i] = max(A_dim_i, B_dim_i)
55
Efficient NumPy Code

Avoid explicit for-loops over indices/axes at all costs. (~10-100x slowdown).

for i in range(x.shape[0]): for i in range(100, 1000):

for j in range(x.shape[1]): for j in range(x.shape[1]):

x[i,j] **= 2 x[i, j] += 5

x **= 2 x[np.arange(100,1000), :] += 5
56
🎯 Numpy Knowhow

How do you create a NumPy array What does np.random.rand(3, 4)

with numbers from 1 to 10? generate?

A. np.arange(1, 10) A. A 3x4 array of random integers

B. np.arange(1, 11)
C. np.array(range(1, 10)) B. A 3x4 array of random values
D. np.linspace(1, 10) between 0 and 1
C. A 3x4 array of random values
between -1 and 1
D. A 3x4 identity matrix

57
🎯 Numpy Knowhow

How do you create a NumPy array What does np.random.rand(3, 4)

with numbers from 1 to 10? generate?

A. np.arange(1, 10) A. A 3x4 array of random integers

B. np.arange(1, 11)
C. np.array(range(1, 10)) B. A 3x4 array of random values
D. np.linspace(1, 10) between 0 and 1
C. A 3x4 array of random values
between -1 and 1
D. A 3x4 identity matrix

58
Language Basics

1 2 3

Why Python? Setting Up Python Basics

4 5 6

Data Structures Numpy Practical Tips

59
List Comprehensions
● Similar to map() from functional programming languages (readability + succinct)
● Format: [func(x) for x in some_list]

=
squares = []
for i in range(10): squares = [i**2 for i
squares.append(i**2) in range(10)]

● Can be conditional:

odds = [i**2 for i in range(10) if i%2 == 1]

60
Convenient Syntax
Multiple assignment / unpacking iterables Join list of strings with delimiter
age, name, pets = 20, ‘Joy’, [‘cat’] “, ”.join([‘1’, ‘2’,
x, y, z = (‘TF’, ‘PyTorch’, ‘JAX’) ‘3’]) == ‘1, 2, 3’

Returning multiple String literals with both Single-line if else

items from a function single and double quotes result = "even"
def some_func(): message = ‘I like if number % 2
return 10, 1 “single” quotes.’ == 0 else "odd"
ten, one = reply = “I prefer
some_func() ‘double’ quotes.”

61
Debugging Tips
Python has an interactive shell where you can execute arbitrary code.
● Great replacement for TI-84 (no integer overflow!)
● Can import any module (even custom ones in the current directory)
● Try out syntax you’re unsure about and small test cases (especially helpful for matrix operations)

$ python
Python 3.9.7 (default, Sep 16 2021, 08:50:36) Helpful Commands
[Clang 10.0.0 ] :: Anaconda, Inc. on darwin
>> import numpy as np Ctrl-d: Exit IPython Session
>> A = np.array([[1, 2], [3, 4]])
>> B = np.array([[3, 3], [3, 3]])
>> A * B
Ctrl-c: Interrupt current command
[[3 6]
[9 12]] Ctrl-l: Clear terminal screen
>> np.matmul(A, B)
[[9 9]
[21 21]]
62
Debugging Tools

Code What it does

array.shape Get shape of NumPy array

array.dtype Check data type of array (for precision, for weird

behavior)

type(stuff) Get type of variable

import pdb; pdb.set_trace() Set a breakpoint [1]

print(f’My name is {name}’) Easy way to construct a string to print

https://docs.python.org/3/library/pdb.html 63
Common Errors
ValueError(s) are often caused by mismatch of dimensions in broadcasting or
matrix multiplication. If you get this type of error, a good first step s to print out the
shape of relevant arrays to see if they match what you expect: array.shape

[Very Active, Open-Source Community] When debugging, check Ed and forums such as
StackOverflow or GitHub Issues → likely that others have encountered the same error!

64
Other Great References
Official Python 3 documentation: https://docs.python.org/3/
Official Anaconda user guide:
https://docs.conda.io/projects/conda/en/latest/user-guide/index.html
Official NumPy documentation: https://numpy.org/doc/stable/
Python tutorial from CS231N: https://cs231n.github.io/python-numpy-tutorial/
Stanford Python course (CS41): https://stanfordpython.com/#/

Several Python and library-specific (ex. NumPy) “Cheat Sheet” guides online as well!

65
 Yayy, we did it! 🥳
Thanks for listening!
66