NEWS - Feb. 20, 2023

Check out the Pytorch version of the Dirichlet Variational Autoencoder, implemented with Recjection Sampling Variational Inference, available at https://github.com/mayanknagda/neural-topic-models.

Another Pytorch version of the DVAE using pathwise gradients is available here: https://github.com/ahoho/topics.

Dirichlet Variational Autoencoders

Implementation of different Dirichlet Variational Autoencoders. Accepted in JMLR 2019

Implements the following methods

• Dirichlet-Autoencoder with "implicit gradients"
• Dirichlet-Autoencoder with RSVI
• Dirichlet-Autoencoder with inverse CDF
• Dirichlet-Autoencoder with Weibull distribution

Running the Code

RSVI

python3 nvdm_dirichlet_rsvi.py 2

Implicit Gradients

python3 nvdm_dirichlet_implicitGradients.py 4

Only with current Tensorflow version. Need to install tensorflow-probability

Inverse CDF

python3 nvdm_dirichlet_invCDF.py 4

Weibull

python3 nvdm_dirichlet_weibull.py 4

Dataset format

See example dataset in data folder