Beta Variational Autoencoder in Tensorflow 2

Demo of a Beta-VAE with eager execution in TF2.

Usage

Begin training the model with train.py

--learning_rate    n   (optional) Float: learning rate
--epochs           n   (optional) Integer: number of passes over the dataset
--batch_size       n   (optional) Integer: mini-batch size during training
UNSUPPORTED --logdir          dir  (optional) String: log file directory
UNSUPPORTED --keep_training        (optional) loads the most recently saved weights and continues training
UNSUPPORTED --keep_best            (optional) save model only if it has the best training loss so far
--help

Track training by starting Tensorboard and then navigate to localhost:6006 in browser

tensorboard --logdir ./tmp/log/

References

Understanding disentangling in β-VAE (Burgess et al. 2018)
https://arxiv.org/abs/1804.03599

From Autoencoder to Beta-VAE (Lilian Weng)
https://lilianweng.github.io/lil-log/2018/08/12/from-autoencoder-to-beta-vae.html

Auto-Encoding Variational Bayes (Kingma & Welling 2013)
https://arxiv.org/abs/1312.6114