Escaping Octave: Andrew Ng’s ML course in Python

What it does This repo holds the eight programming exercises from Andrew Ng’s introductory Machine Learning Coursera course, rewritten in Python. You work through linear regression, neural networks, SVMs, and recommender systems using numpy and matplotlib instead of Octave or Matlab, checking your output visually at each step just like the original assignments.

The interesting bit The author deliberately flattens Octave’s column vectors into one-dimensional ndarrays and bans numpy.matrix entirely. It is a quietly opinionated choice that keeps the code feeling like Python rather than a literal syntax translation.

Key highlights

• Covers the full course arc: linear and logistic regression, neural networks, bias/variance, SVMs, K-means, PCA, anomaly detection, and recommender systems.
• Mirrors the original assignment structure so you can verify your work step-by-step.
• Uses plain numpy.ndarray instead of numpy.matrix, avoiding a common Matlab-to-Python impedance mismatch.
• Points to a separate solutions repo where the author vectorized the implementations.
• Targets Python 3.6 and depends on the usual scientific Python suspects (numpy, scipy, matplotlib, plus scikit-learn, scikit-image, and nltk for specific exercises).

Caveats

• The README specifies Python 3.6 but does not mention compatibility with newer Python releases.
• This is strictly coursework; do not expect a maintained library or API.

Verdict A solid companion if you are taking (or re-taking) Andrew Ng’s course and would rather write Python than Octave. Look elsewhere if you need a modern, production-ready machine learning framework.