vae

Categorical Variational Auto-Encoder

Code for Categorical Variational Auto-Encoder adapted from ARSM. Runs on Python 3.6 and TensorFlow 1.12. It includes implementations of our unordered set estimator, as well as REINFORCE without replacement (using a built-in baseline) and a number of other baselines.

• See all_experiments.sh for the commands that were used to run experiments.
• Run jupyter notebook from this (/vae) directory and use plot_vae_results.ipynb to plot the results.

Acknowledgements

Thanks to the original implementation of the Categorical VAE in https://github.com/ARM-gradient/ARSM.