Skip to content

amspector100/discrete-latent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Discrete-Latent

Tests

To run unit-tests, run 'python -m unittest' in the parent directory (discrete-latent).

If you test a class you're building, please add that test to the test folder (it will save a lot of time)! It shouldn't take any time since we'll have to test the stuff we build either way.

Structure

The big-picture structure is as follows:

  • encoders.py contains encoder classes
  • decoders.py contains decoder classes
  • latents.py contains classes which handle the transfer of information between the encoder/decoder (e.g. the VQVAE class)
  • vqvae.py contains the master class which wraps the encoder, decoder, and latent classes
  • training.py contains functions for training the vae
  • data.py handles loading/processing data

There are also a variety of utilities scattered around (kmeans, weight dropout, gumbel softmax sampling).

To-Do

  • Play with norm parameter ('l2' or 'softmax') in gumbel softmax

About

Kmeans-based methods for training discrete latent variable language models

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages