camdavidsonpilon/probabilistic-programming-and-bayesian-methods-for-hackers

This project is a free online book called Bayesian Methods for Hackers, also titled Probabilistic Programming and Bayesian Methods for Hackers. It is an introduction to Bayesian inference, which is a way of reasoning about uncertainty by updating beliefs as new data comes in. The repository is written almost entirely as Jupyter notebooks in Python, so you read the chapters and run the code alongside the explanations. The book's pitch is that most introductions to Bayesian inference start with two or three chapters of heavy probability theory and then only get to small, artificial examples because the math becomes intractable. The author argues that modern computers let us skip a lot of that math by using probabilistic programming, where you describe a model in code and let the computer do the hard inference work. The text is therefore "computation and understanding first, mathematics second." It uses the PyMC library as the probabilistic programming language, and the examples rely only on PyMC, NumPy, SciPy, and Matplotlib so the dependencies stay light. Both a PyMC2 and a PyMC3 version of the chapters are included. Chapters work through concrete problems rather than pure theory. Examples mentioned include inferring behavior changes from text message rates, estimating cheating among students, analyzing the Challenger space shuttle disaster, sorting Reddit comments, the multi-armed bandit problem, and pricing decisions from the game show The Price is Right. You would use this if you want to learn applied Bayesian statistics without a heavy math prerequisite, or if you want practical, worked examples in PyMC. There is also a printed edition published by Addison-Wesley that adds a chapter on Bayesian A/B testing, updated examples, and answers to end-of-chapter questions.