trekhleb/homemade-machine-learning ? reverse-engineered prompt

Build me a Python learning project that teaches machine learning from scratch with clear math explanations and interactive notebook demos. I want it to feel like a hands on course, where each algorithm has a short theory section, a plain language explanation of the formulas, and a notebook I can run in the browser to play with training data, tweak settings, and immediately see charts and predictions.

Please include examples for things like linear regression, logistic regression, k means clustering, anomaly detection, and a simple neural network or perceptron style model. Use small sample datasets where it makes sense, like Iris or handwritten digits, and keep the implementations homemade instead of relying on one line library shortcuts. Organize it so the code and the notebooks match each other, and make it easy for someone learning ML for the first time to follow along. If you need to, look up current docs online for notebook support or any small details.