Skip to content
forked from BorealisAI/CP-VAE

On Variational Learning of Controllable Representations for Text without Supervision https://arxiv.org/abs/1905.11975

License

Notifications You must be signed in to change notification settings

larrylawl/CP-VAE

 
 

Repository files navigation

RPP4 CP-VAE

Setup

Create environment

conda env create -f env.yml

Set up the gyafc data directory. Data path follows this convention:

data_pth = os.path.join(args.hard_disk_dir, "data", args.data_name, "processed")

where args.hard_disk_dir and args.data_name are specified in the script.

Training

Training with CP-VAE. Current model is CP-VAE with shared BERT encoder, GPT2 decoder, and style loss.

data_name=gyafc
save=checkpoint/exp_name
hard_disk_dir=/hdd2/lannliat/CP-VAE  # change to your data directory

python run.py --hard_disk_dir $hard_disk_dir
            --data_name $data_name \
            --save $save \
            --subset \  # trains with only subset of data for faster experiments
            --to_plot  # plots simplex p and umap of z1.

Style Transfer

Apply style transfer on trained model. Currently z1 is taken as the mean of the z1 of the input validation examples.

data_name=gyafc
path_to_checkpoint="/hdd2/lannliat/CP-VAE/checkpoint/subset-cpvae-styleloss-gyafc-fm/20221003-091150/"

python transfer.py --data_name $data_name \
                   --load_path $path_to_checkpoint

About

On Variational Learning of Controllable Representations for Text without Supervision https://arxiv.org/abs/1905.11975

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 70.1%
  • Jupyter Notebook 29.9%