A basic VAE implementation to reproduce the results in Kingma and Welling, 2014.
conda env create -f environment.yml
conda activate torch
python vae.py
| Arg | Value |
|---|---|
| seed | random seed |
| batch_size | batch size |
| learning_rate | learning rate |
| n_epochs | # training epochs |
| no_cuda | true means don't use cuda |
| hidden_size | dim of encoder/decoder hidden state |
| latent_size | dim of latent encoding |
| test_output | how test samples after each epoch are generated |
| test_output_size | square dimension of test sample plots |
Save decodings of 20x20 uniformly spaced latent codes in the latent space after each epoch as a .png.
python vae.py --n_epochs 5 --latent_size 2 --test_output uniform --tn 20
Save reconstructions of 20x20 random test samples after each epoch as a .png.
python vae.py --test_output random --tn 20