PLSV-VAE

Pytorch Implementation of Auto-Encoding Variational Bayes for Inferring Topics and Visualization

Train

Python version: 3.6, Pytorch version: 1.2.0.

To train the plsvVAE model:

python train.py -e 1000 -dist gaussian

Available distance functions: gaussian, multi_quadratic, and inverse_quadratic.

To train your own dataset:

• Preprocess your dataset (use your own preprocessing functions or use our preprocessing function in data_preprocessing.py).
• Vectorize train data (for large dataset, to avoid memory problem, we recommend using gensim instead of CountVectorizer of sklearn).
• Run train.py script.

Example

• train_nb.ipynb: Training example with top words and visualization on REUTERS dataset.
• Example on arxiv dataset with top-10 topic words:

Credit

I started from a fork of this repo.

Citing

If you use the data or the model, please cite,

@inproceedings{pham2020auto,
  title={Auto-Encoding Variational Bayes for Inferring Topics and Visualization},
  author={Pham, Dang and Le, Tuan},
  booktitle={Proceedings of the 28th International Conference on Computational Linguistics},
  year={2020}
}