rxlgq / t_vae

pytorch implementation for "Student-t Variational Autoencoder for Robust Density Estimation".

Student-t Variational Autoencoder for Robust Density Estimation

This is a pytorch implementation of the following paper [URL]:

@inproceedings{takahashi2018student,
  title={Student-t Variational Autoencoder for Robust Density Estimation.},
  author={Takahashi, Hiroshi and Iwata, Tomoharu and Yamanaka, Yuki and Yamada, Masanori and Yagi, Satoshi},
  booktitle={IJCAI},
  pages={2696--2702},
  year={2018}
}

Please read license.txt before reading or using the files.

Prerequisites

Please install python>=3.6, torch, numpy and scikit_learn.

Usage

usage: main.py [-h] [--dataset DATASET] [--decoder DECODER]
               [--learning_rate LEARNING_RATE] [--seed SEED]

• You can choose the dataset from following datasets: SMTP.
  • We are preparing other datasets.
• You can choose the decoder of the VAE from normal or student-t.
• You can also change the random seed of the training and learning_rate of the optimizer (Adam).

Example

SMTP with Gaussian decoder:

python main.py --dataset SMTP --decoder normal

SMTP with Student-t decoder:

python main.py --dataset SMTP --decoder student-t

Output

• After the training, the mean of log-likelihood for test dataset will be displayed.
• The detailed information of the training and test will be saved in npy directory.