CP-VAE

On Variational Learning of Controllable Representations for Text without Supervision

Paper published in ICML 2020: arXiv

Preprocess

Download the GloVe embeddings from this link to data/. Then run:

python preprocess.py --data_name <data_name>

Data name can be either yelp or amazon.

Train

Vanilla VAE:

python run_baseline.py --data_name <data_name>

Hyper-parameters can be set in baseline_config.py.

CP-VAE:

python run.py --data_name <data_name>

Hyper-parameters can be set in config.py.

Evaluation

Train the sentiment classifier:

python classifier.py --data_name <data_name>

Perform unsupervised style transfer:

Vanilla VAE:

python transfer_baseline.py --data_name <data_name> --load_path <path_to_checkpoint> --type <magnitude_type>

Magnitude type can be:

0, no change
1, move by one std
2, move by two std
3, move to extremum

CP-VAE:

python transfer.py --data_name <data_name> --load_path <path_to_checkpoint>

Calculate sentiment classification accuracy and BLEU score against the source sentence:

python evaluate.py --data_name <data_name> --target_path <path_to_target>

Analysis

Visualization of NLL discrepency

Copy the nll statistics from the checkpoint folders to plot/ and then run:

python plot_nll.py

Topological Data Analysis

python tda.py --data_name <data_name> --load_path <path_to_checkpoint> --resolution <n>

python tda_baseline.py --data_name <data_name> --load_path <path_to_checkpoint> --resolution <n>

Cite

If you found this codebase or our work useful, please cite:

@InProceedings{xu2020variational,
    author={Xu, Peng and Cheung, Jackie Chi Kit and Cao, Yanshuai},
    title={On Variational Learning of Controllable Representations for Text without Supervision},
    booktitle={The 37th International Conference on Machine Learning (ICML 2020)},
    month={July},
    year={2020},
    publisher={PMLR},
}

BorealisAI/CP-VAE