wlmnzf / BN

Spin up with Variational Bayes

What you get:

1. mlp.py - MLP classifier on MNIST (in JAX and Haiku).
2. vae.py - Bernoulli VAE generative model on MNIST.
   See: USING A BERNOULLI VAE ON REAL-VALUED OBSERVATIONS.
3. bayes.py - Variational Bayes NN classifier on MNIST.
   • Bayesian Neural Networks with TensorFlow Probability.
   • CSC421/2516 Lecture 19: Bayesian Neural Nets.
   • Making Your Neural Network Say “I Don’t Know” — Bayesian NNs using Pyro and PyTorch.

How to run:

Install dependencies with pip install -r requirements.txt. It was tested in Python 3.7. As always, use of a virtual environment is recommended.

Each file is an independent implementation that uses HumbleSL library (see below), run with: python <file>. Run python <file> --help to see all the configurable parameters.

Note on HumbleSL - deep supervised learning library.

It's a straightforward supervised learning (SL) Python library. It provides all the boilerplate code needed to do Deep SL: a network definition factory, metrics and losses, a data loader, train loop, etc.

It's backed by the JAX library and the Haiku framework. It uses TensorFlow Datasets for data loading and preprocessing.