Machine Learning

A 15-module course on classical machine learning — regression, classification, ensemble methods, and how to choose between them. Each module is a self-contained Jupyter notebook that walks through one algorithm from problem framing to working code, with a worked example you can read by hand and a "when it breaks" section that names the failure modes you'll actually hit.

The course is designed to be read in order, but every module also stands on its own as a reference.

What's inside

Modules

#	Topic
00	Introduction — what machine learning is, the workflow, and how to use this course
01	Linear Regression
02	Polynomial & Regularized Regression
03	Logistic Regression
04	Support Vector Machines
05	Decision Trees
06	Bagging
07	Random Forests
08	AdaBoost
09	Gradient Boosting
10	Stacking & Voting
11	k-Nearest Neighbors
12	Naive Bayes
13	Algorithm Selection — given a problem, which algorithm should you reach for?
14	Model Comparison & Evaluation — capstone on metrics, validation, and head-to-head comparison

Reference notebooks (references/)

Concept-focused notebooks the modules cross-link to. Read them when a module assumes a concept you want to revisit.

• correlation.ipynb
• feature-encoding.ipynb
• gradient-descent.ipynb
• learning-rate.ipynb
• loss-functions.ipynb
• n-grams.ipynb
• regularization.ipynb
• standardization.ipynb
• statistical-inference.ipynb
• tf-idf.ipynb

Datasets (data/)

File	Used in	Description
california-housing.csv	Modules 00–02 (regression)	20,640 California Census block groups, 8 features, median house price target
breast-cancer.csv	Modules 03–12 (classification)	569 samples, 30 features, malignant/benign target (Wisconsin Diagnostic)
breast-cancer-explore.csv	Exploration	Same data with a small subset for hand-readable walkthroughs

See data/README.md for full dataset documentation, sources, and the pattern for adding new ones.

Running the notebooks

You'll need Python 3.10+ and the standard scientific stack. From the repo root:

pip install jupyter numpy pandas scikit-learn matplotlib seaborn
jupyter notebook

Then open any .ipynb. Notebooks load data from the local data/ directory, so they work offline.

How the modules are structured

Every algorithm module follows the same arc:

1. Problem — the situation where this algorithm earns its keep
2. Model — what it predicts and what shape its decisions take
3. How it learns — the training procedure, formulas, and intuition
4. Superpower — what this algorithm does better than its peers
5. When it breaks — the failure modes, visually and quantitatively
6. Worked example — a 10-row dataset you can compute by hand
7. Code — full scikit-learn implementation with sensible defaults
8. Comparisons — head-to-head tables against algorithms covered earlier

Formulas are followed by a "Reading this formula" block that names every symbol. Tunable numbers (learning rates, depths, regularization strengths) are named variables, not magic literals.

License

All rights reserved.