Learning outcomes: Introduction

You’ll be able to:

• explain what programming is and why it matters for AI/ML
• explain why Python is the language of choice
• describe the course structure: units and sessions
• find help when stuck: documentation, help(), the internet, AI tutors

Learning outcomes: Development environment

You’ll be able to:

• setup a development environment with Jupyter
• explain what a virtual environment is and activate it
• install a Python package
• explain what the Python interpreter is and use it interactively and to run scripts
• create and use a Jupyter Notebook: code and Markdown cells, kernel

Learning outcomes: Variables, I/O and data types

You’ll be able to:

• store and load data of different types in variables
• retrieve and display information
• convert data between types
• perform operations with numeric types
• format a string

Learning outcomes: Making decisions

You’ll be able to:

• make decisions based on conditions
• import a package and use its functions
• load data from an Excel file
• basic data filtering using boolean logic

Why learn to program?

• AI and ML are applied through code : every step from loading data, to training models and evaluating results is written in a programming language
• Libraries are the real syllabus : scikit-learn, PyTorch, pandas… are what you’ll use; they have no graphical interface, only an API
• Debugging is thinking : when a model underperforms, you need to inspect data, trace transformations, test assumptions; that process is also programming

What is programming?

The art of programming is the art of organizing complexity. — Edsger Dijkstra.

What about no-code, low-code and AI generated code?

• No-code tools are built by programmers : when they break, hit a limit, or do something unexpected, you’re stuck
• AI assistants write code, not decisions : AI can generate code in seconds, but you should still understand it, judge it, and know when it’s wrong
• The interesting problems don’t have buttons : custom data, novel models, real deployments; at some point serious projects touch code
• Go beyond just using tools : a postgrad in Applied AI/ML means understanding what’s happening under the hood, not just clicking “train model”

Why Python?

• The world runs AI and ML in Python : scikit-learn, PyTorch, TensorFlow, Hugging Face are all Python-first
• Syntax similar to natural language : if value > threshold: reject reads like a decision, not a program; focus on ideas, not punctuation
• Open source, free
• One language, everything included — Python is not the best language for anything, but its a close second best for everything

Course Overview

Sessions

Each session opens with the learning outcomes of its units and closes with their recaps.

Learning Resources

Trying and failing , and trying again. Nothing beats experience.

Recommended bibliography

• Sweigart, A. (2019). Automate the Boring Stuff with Python: Practical Programming for Total Beginners (2nd ed.). No Starch Press.
• McKinney, W. (2022). Python for Data Analysis: Data Wrangling with Pandas, NumPy, and IPython (3nd ed.). O’Reilly Media. Free Access
• Guttag, J. V. (2016). Introduction to Computation and Programming Using Python: With Application to Understanding Data (2nd ed.). MIT Press. Free Access

Development environment

• Python distribution
• package manager
• Jupyter Notebooks are very popular. We’ll install that to get you started.

uv

• uv is increasingly becoming the standard way of installing Python and managing environments and packages
• it works through the terminal (so go ahead and open a terminal, PowerShell, whatever you have in your system)
• one tool replaces several: Python installer, pip, venv, …
• fast: installs packages in seconds

uv install

macOS / Linux (Terminal):

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Close and reopen the terminal, then check it worked:

uv --version

uv project

mkdir aaiml-python
cd aaiml-python
uv init

What is a virtual environment?

Installing JupyterLab

uv add jupyterlab

• downloads Python if needed, creates .venv, installs the package
• the package is recorded in pyproject.toml

Start JupyterLab:

uv run jupyter lab

• uv run runs a command inside the project environment
• JupyterLab opens in your browser

Activating the virtual environment

source .venv/bin/activate

.venv\Scripts\activate

uv cheat sheet

The Python interpreter

uv run python

>>> 2 + 3
5
>>> print("hello")
hello
>>> exit()

uv run python main.py

Jupyter Notebook

uv vs Anaconda

You only need one of them.

• uv: lightweight and fast, terminal-based, one environment per project; install only what you need
• Anaconda: graphical installer, comes with Python, Jupyter and hundreds of data science packages pre-installed; much larger download

Prefer a graphical installer with everything bundled? Anaconda is the alternative.

Anaconda Distribution

Anaconda is a popular open-source Python distribution that makes it easy to install Python and many data science libraries.

This link will take you to the Anaconda download page. Download Anaconda for your operating system.

Anaconda Download Verification

Amazon SageMaker

Amazon SageMaker provides an extensive set of tools to develop the complete lifecycle of a ML solution.

It provides a managed Jupyter Notebook service, i.e. running on remote servers (AWS).

Beware of running processes after no longer in use, because of associated costs. Small mistakes can cost a lot of money.

Google Colab

Another managed Jupyter Notebook provider is Google Colab.

Google Colab is a free Jupyter Notebook provider that will allow you to run code on Google’s servers, using only CPUs. Execution with GPUs is paid.

Getting Started

From now on, it’s assumed you have a functional Python environment.

Go on and open a Jupyter Notebook to follow along.

What is a Variable?

Creating Variables

Let’s create some variables.

accuracy = 0.87
num_samples = 1000
model_name = "LogisticRegression"
previous_accuracy = accuracy
accuracy = 0.91
print("Model trained!")
print(accuracy)
print(model_name)
print(accuracy, num_samples, previous_accuracy)

accuracy, num_samples and model_name are all variables.

Assignment Operator

accuracy = 0.87
num_samples = 1000
model_name = "LogisticRegression"
previous_accuracy = accuracy
accuracy = 0.91
print("Model trained!")
print(accuracy)
print(model_name)
print(accuracy, num_samples, previous_accuracy)

The = operator assigns a value (any value) to the variable.

Variable Assignment

accuracy = 0.87
num_samples = 1000
model_name = "LogisticRegression"
previous_accuracy = accuracy
accuracy = 0.91
print("Model trained!")
print(accuracy)
print(model_name)
print(accuracy, num_samples, previous_accuracy)

We can also assign the value of a variable to another variable.

previous_accuracy now has the value of accuracy, which is 0.87

Overwriting Variables

accuracy = 0.87
num_samples = 1000
model_name = "LogisticRegression"
previous_accuracy = accuracy
accuracy = 0.91
print("Model trained!")
print(accuracy)
print(model_name)
print(accuracy, num_samples, previous_accuracy)

And we can overwrite the value of an existing variable.

Retrieving Values

We can retrieve the value of a variable, simply by typing its name.

accuracy = 0.87
num_samples = 1000
model_name = "LogisticRegression"
previous_accuracy = accuracy
accuracy = 0.91
print("Model trained!")
print(accuracy)
print(model_name)
print(accuracy, num_samples, previous_accuracy)

We’re retrieving the value of accuracy and then assigning it to another variable.

I/O Methods

• Terminal (print)
• Files (later)
• Networking (not covered)
• Graphical interfaces (not covered)
• Inter-process communication (not covered)

Interacting through the terminal

• print - send values to user
• input - receive values from user

The print Function

print will write some text to the terminal/screen.

accuracy = 0.87
num_samples = 1000
model_name = "LogisticRegression"
previous_accuracy = accuracy
accuracy = 0.91
print("Model trained!")
print(accuracy)
print(model_name)
print(accuracy, num_samples, previous_accuracy)

The print function will comes in handy to check the values of variables.

Printing Multiple Values

accuracy = 0.87
num_samples = 1000
model_name = "LogisticRegression"
previous_accuracy = accuracy
accuracy = 0.91
print("Model trained!")
print(accuracy)
print(model_name)
print(accuracy, num_samples, previous_accuracy)

We can also print several values at once, if we separate them with commas.

The input Function

print("Enter patient glucose level (mg/dL):")
glucose = input()
age = input("Enter patient age: ")

After executing the input function, the terminal will wait for user input.

The user can write whatever she wishes, ending with an enter.

Input with a prompt

print("Enter patient glucose level (mg/dL):")
glucose = input()
age = input("Enter patient age: ")

The input function can receive an optional string that is printed to the terminal before receiving input from the user.

Numeric Types

train_split = 0.8
num_samples = 5000
print(type(train_split))   # float
print(type(num_samples))   # int
dataset_name = "diabetes"
print(type(dataset_name))  # str
is_balanced = False
print(type(is_balanced))   # bool

Both train_split and num_samples are numeric types, but they’re not the same.

The type Function

train_split = 0.8
num_samples = 5000
print(type(train_split))   # float
print(type(num_samples))   # int
dataset_name = "diabetes"
print(type(dataset_name))  # str
is_balanced = False
print(type(is_balanced))   # bool

The type function will tell us the type of any value we give it.

train_split is a float (as in floating point number), the representation of a real number.

num_samples is an int (as in integer), the representation a natural number including negative numbers.

String Type

str (short for string) is how we represent text.

train_split = 0.8
num_samples = 5000
print(type(train_split))   # float
print(type(num_samples))   # int
dataset_name = "diabetes"
print(type(dataset_name))  # str
is_balanced = False
print(type(is_balanced))   # bool

Boolean Type

bool (short for boolean) encodes a True or False value.

train_split = 0.8
num_samples = 5000
print(type(train_split))   # float
print(type(num_samples))   # int
dataset_name = "diabetes"
print(type(dataset_name))  # str
is_balanced = False
print(type(is_balanced))   # bool

Many operations, like comparisons between two values, will return a True or False value.

Numeric types

int, float

Operations: + - * / // % **

Conversion

💻 Let’s see in practice! 💻

String

How can we deal with text? Strings!

String methods

Formatted strings

Conversion

💻 Back to the code! 💻

Indentation

In Python, indentation defines structure — there are no { } (like in C, for example)

if True:
    print("this is inside the block")   # indented = belongs to the if
    print("so is this")
print("this is outside")                # not indented = runs regardless

Use 4 spaces, or a consistent tab, per level. Mixing them causes errors.

Modern editors allow configuration and usually fix mixing.

Making decisions

A program has very limited functionality if it’s not able to make decisions.

Conditionals Example

Conditionals enable us to execute a portion of code based on some decision.

new_accuracy = 0.92
decision = None
if new_accuracy >= 0.9:
    decision = "deploy"
else:
    decision = "keep training"

What is happening here?

The if Statement

new_accuracy = 0.92
decision = None
if new_accuracy >= 0.9:
    decision = "deploy"
else:
    decision = "keep training"

In Python, we state conditionals with the if statement.

We’re checking if new_accuracy meets our deployment threshold of 0.9.

If it is, the comparison operator (>=) will return True, otherwise False.

Executing Conditional Code

new_accuracy = 0.92
decision = None
if new_accuracy >= 0.9:
    decision = "deploy"
else:
    decision = "keep training"

When the condition after if is True, the program will execute the code that is indented forward.

Here, that code is in line 4.

The else Statement

new_accuracy = 0.92
decision = None
if new_accuracy >= 0.9:
    decision = "deploy"
else:
    decision = "keep training"

The else statement allows us to define code to be executed when the condition fails.

Here, that code is in line 6.

Comparison Operators

Conditions are usually built by comparing values. The result is always a bool.

Chaining Conditions

new_accuracy = 0.88
champion_accuracy = 0.87
decision = None
if new_accuracy >= champion_accuracy + 0.02:
    decision = "deploy new model"
elif new_accuracy >= champion_accuracy - 0.01:
    decision = "retrain and monitor"
elif new_accuracy >= 0.80:
    decision = "collect more data"
else:
    # below minimum threshold
    decision = "discard model"

We can also chain several conditions with elif.

elif works like else, but it allows us to define more conditions.

Condition Order

new_accuracy = 0.88
champion_accuracy = 0.87
decision = None
if new_accuracy >= champion_accuracy + 0.02:
    decision = "deploy new model"
elif new_accuracy >= champion_accuracy - 0.01:
    decision = "retrain and monitor"
elif new_accuracy >= 0.80:
    decision = "collect more data"
else:
    # below minimum threshold
    decision = "discard model"

When chaining conditions in this way, a given condition is only processed if all the previous conditions fail.

Condition Processing Order

new_accuracy = 0.88
champion_accuracy = 0.87
decision = None
if new_accuracy >= champion_accuracy + 0.02:
    decision = "deploy new model"
elif new_accuracy >= champion_accuracy - 0.01:
    decision = "retrain and monitor"
elif new_accuracy >= 0.80:
    decision = "collect more data"
else:
    # below minimum threshold
    decision = "discard model"

The condition on line 6 will only be processed if the condition on line 5 fails.

Multiple elif Statements

new_accuracy = 0.88
champion_accuracy = 0.87
decision = None
if new_accuracy >= champion_accuracy + 0.02:
    decision = "deploy new model"
elif new_accuracy >= champion_accuracy - 0.01:
    decision = "retrain and monitor"
elif new_accuracy >= 0.80:
    decision = "collect more data"
else:
    # below minimum threshold
    decision = "discard model"

The condition on line 8 will only be processed if the conditions on lines 5 and 7 fail.

Final else Statement

new_accuracy = 0.88
champion_accuracy = 0.87
decision = None
if new_accuracy >= champion_accuracy + 0.02:
    decision = "deploy new model"
elif new_accuracy >= champion_accuracy - 0.01:
    decision = "retrain and monitor"
elif new_accuracy >= 0.80:
    decision = "collect more data"
else:
    # below minimum threshold
    decision = "discard model"

The final else statement will only be processed if the conditions on lines 5, 7 and 8 fail.

Comments

new_accuracy = 0.88
champion_accuracy = 0.87
decision = None
if new_accuracy >= champion_accuracy + 0.02:
    decision = "deploy new model"
elif new_accuracy >= champion_accuracy - 0.01:
    decision = "retrain and monitor"
elif new_accuracy >= 0.80:
    decision = "collect more data"
else:
    # below minimum threshold
    decision = "discard model"

# is the start of a comment in Python.

About Comments

Comments are not code, they’re just extra information the programmer adds to the program so it’s easier to understand the code.

Comments are never executed, even if they contain valid code.

Boolean logic

Combine conditions with:

a and b → result is True, if both arguments are True

a or b → result is True, if any of the arguments is True

not a → if a was True, it is now False, and vice versa.

Boolean logic example

patience = 15 # confit param

# these two variables would come from the
# current iteration of a training run
accuracy = 0.88
epoch = 30

# these two variables would come from the
# best past iteration of a training run
best_accuracy = 0.87
best_epoch = 24

decision = None
if accuracy < best_accuracy and epoch - best_epoch >= patience:
    decision = "stop training"
elif accuracy >= best_accuracy:
    # continue training
    best_accuracy = accuracy
    best_epoch = epoch
# there is no else in this example

Demo

• Installing and importing packages
• Quick teaser: filtering data in an Excel file with boolean logic
  • See animations for bitwise boolean logic in the following slides