vae-pytorch: A Jupyter Notebook repository from darleybarreto

Vae-Pytorch

This repository has some of my works on VAEs in Pytorch. At the moment I am doing experiments on usual non-hierarchical VAEs. ConvVAE architecture is based on this repo, and MLPVAE on this.

Currently implemented VAEs:

Standard Gaussian based VAE
Gamma reparameterized rejection sampling by Naesseth et al.. This implementation is based on the work of Mr. Altosaar.

How to run

Example 1:

$ python3 main.py

Example 2:

$ python3 main --model normal --epochs 5000

Example 3:

$ python3 main --model gamma --b_size 256

Example 4:

$ python3 main --model gamma --dataset mnist --z_dim 5

Usage

usage: main.py [-h] [--model M] [--epochs N] [--dataset D]
               [--b_size  B] [--z_dim   Z]

optional arguments:
  -h, --help        show this help message and exit
  --model   M       vae model to use: gamma | normal, default is normal
  --epochs N        number of total epochs to run, default is 10000
  --dataset D       dataset to run experiments (cifar-10 or mnist)
  --b_size  B       batch size
  --z_dim   Z       size of the latent space