Bayesian ML you can execute, not just read

What it does This repository is a curated stack of Jupyter notebooks covering Bayesian machine learning from linear regression to variational autoencoders. Each notebook introduces a concept—Gaussian processes, Bayesian optimization, Bayesian neural networks—and implements it using anything from plain NumPy to Keras, JAX, and TensorFlow Probability. It functions as a structured set of labs for turning probabilistic theory into working code.

The interesting bit Many topics are covered twice: once with raw NumPy or SciPy, and again using libraries like scikit-learn, GPy, or PyMC3, sometimes in companion notebooks. That repetition lets you see what the library is hiding under the hood without sacrificing mathematical notation, which is rendered properly via nbviewer.

Key highlights

• Covers Gaussian processes (regression, classification, and sparse), Bayesian optimization, Bayesian neural networks, and variational autoencoders.
• Many topics include both from-scratch NumPy/SciPy code and library versions (scikit-learn, PyMC3, GPy, TensorFlow Probability), sometimes across companion notebooks.
• Sparse Gaussian processes and some examples use JAX.
• Practical applications include hyper-parameter tuning and conditional generation in latent space.
• Several notebooks include direct Colab links for immediate execution, and formulas are rendered via nbviewer.

Caveats

• This is a collection of standalone tutorials, not a unified library; expect notebooks, not importable modules.
• Dependencies are split across subdirectory requirements.txt files, so local execution requires checking each folder individually.
• Notebooks target specific framework versions such as TensorFlow 2.x, Keras, and PyMC3; compatibility with newer releases is not addressed in the README.

Verdict A solid reference for students and practitioners who learn best by implementing Bayesian concepts. Look elsewhere if you need a maintained, high-level package to drop into production.