Classic ML algorithms, rewritten in plain Python

What it does A collection of Jupyter notebooks that implements standard machine learning algorithms—linear and logistic regression, k-NN, k-Means, SVMs, a simple neural network, and others—using only Python 3.6+ and no external ML libraries. The author explicitly warns that the code prioritizes clarity over speed or production readiness.

The interesting bit The project treats minimal dependencies as a feature. By refusing to reach for scikit-learn or similar libraries, the notebooks force you to confront the underlying mechanics that are usually buried under abstraction layers.

Key highlights

• Covers a broad undergraduate syllabus: decision trees, PCA, perceptron, softmax regression, Bayesian linear regression, and more.
• Includes starter notebooks on image and numerical data preprocessing, though the README notes these require downloading datasets first.
• Offers a live Binder demo that runs fully in-browser, though the preprocessing notebooks are excluded because they need local datasets.
• MIT licensed and open to algorithm requests via issues.

Caveats

• The author is upfront that these are not efficient implementations; if you need performance, look elsewhere.
• The preprocessing track is incomplete: the README promises more dataset types “within the next months,” and the existing preprocessing notebooks do not work in Binder because they require downloading datasets first.

Verdict Ideal for students or practitioners who have called .fit() a hundred times but never wrote a gradient update by hand. Skip it if you are hunting for optimized, production-ready code.