Python for AI (AI-101)
You can already write a loop and a function, and you have watched a tutorial fit
a model in three lines. Then you open a real project: the data arrives as a pile
of raw strings, someone asks whether the model is actually any good, and you have
no honest answer because you never split the data or wrote down a baseline. The
gap between "I ran model.fit" and "I can defend this prediction" is not a
library gap — it is a way of thinking about data, and this course closes it.
AI-101 is the foundation course for the School of AI Engineering. It is not a machine-learning survey; it is the numerical-Python and data-handling course the rest of the school assumes. You will learn to see any dataset as a table of feature vectors paired with labels, load and summarize that table, split it so your evaluation stays honest, and build a first from-scratch predictor whose accuracy you can state out loud. The libraries the school leans on later — NumPy, pandas, and scikit-learn — are named here so you know where you are heading, but every idea is first built by hand in plain Python so you understand what those libraries do for you rather than around you.
- Represent a raw dataset as feature vectors paired with labels using Python's lists, tuples, dicts, comprehensions, and zip - Load a tabular dataset from CSV with the standard library and compute per-feature summary statistics - Split a dataset into train and test sets deterministically, and explain why the split must come before any decision that touches the data - Reason with the baseline-first discipline: state a majority-class baseline before building any model - Build a from-scratch nearest-neighbor predictor and measure it with a held-out accuracy metric
Who this course is for
AI-101 gates the School of AI Engineering; a placement assessment can waive it for learners who already hold DE-103 or equivalent Python experience. It is written for:
- The analyst going deeper who is comfortable with data and basic statistics and wants the Python fluency to move from describing data to modeling it.
- The software engineer pivoting into machine learning, who has solid engineering habits but has never framed a prediction problem or measured a model.
- The data engineer extending up who builds the pipelines these systems consume and now wants to own the modeling leg — this course is the fastest way to fill the numerical-Python and evaluation basics before AI-102.
Prerequisites
You should be comfortable writing basic Python: variables, loops, functions, lists, and dictionaries. If you have taken DE-103 (Python for Data Engineering) or DA-201, you are more than ready. You do not need a cloud account, a GPU, a paid service, or any third-party package: every example and the lab run on Python 3.10 or newer using only the standard library, in keeping with the platform's $0 posture. Bring curiosity about how a prediction is actually made, not just how to call one.
Modules
AI-101 is roughly twelve hours of effort across six modules. This slice delivers three core lessons and your first hands-on lab, covering the data-handling and first-model skills that every later module builds on.
| # | Module | What you leave with |
|---|---|---|
| 1 | The scientific Python stack | Environments, notebooks, and reproducible runs |
| 2 | NumPy: arrays and vectorized thinking | Broadcasting and array operations over loops |
| 3 | pandas for ML | Feature tables, joins, and leakage-safe splits |
| 4 | Visualization for model work | Distributions, relationships, and residual plots |
| 5 | scikit-learn mechanics | Estimators, transformers, and pipelines |
| 6 | Math you actually need | Vectors, matrices, and gradients in code |
The three lessons and the lab below build the intuition beneath Modules 1, 3, and 5 — the shape of ML data, honest dataset handling, and a first predictor — using plain Python so the library abstractions land on solid ground.
Outcomes
By the end of AI-101 you can:
- Turn a pile of raw records into an
(X, y)structure a model can consume. - Load a CSV dataset and describe it with counts, ranges, and per-feature means.
- Split data into train and test sets and say why the order of operations protects you from lying to yourself about accuracy.
- Write down a majority-class baseline and beat it with a real model.
- Implement a nearest-neighbor classifier from scratch and report its held-out accuracy.
Where this leads: the capstone
Everything in this school converges on AI-350: End-to-End ML Application, where you frame a problem, track experiments across model iterations, and ship a trained model behind a documented service. The habits you start here — pairing features with labels, splitting before you decide anything, and never quoting an accuracy without a held-out set — are the ones AI-350 grades you on most strictly. AI-102 (Machine Learning Fundamentals) picks up directly from the baseline-first predictor you build in this course's lab.
Do the lessons in order. Each one feeds the next: Lesson 1 gives you the feature-and-label structure, Lesson 2 shows you how to load and split real data into it, and Lesson 3 turns that split into a measured prediction — which is exactly the pipeline the lab asks you to build end to end.