cs591_deeplearning_project
BU Deep Learning Final Project
This directory is a modification and extension of the work 《Style Transformer: Unpaired Text Style Transfer without Disentangled Latent Representation》. I wish to thank them for their contribution and generousness.
Environment:
I implemented it by creating a virtual environment on the scc and pip install each of them except pytorch. You need to download the pytorch=0.4.0 wheels to pip install it to the virtual environment. However, the virtual environment folder and the pytorch wheel are too large to be uploaded.
pytorch >= 0.4.0
torchtext >= 0.4.0
nltk
fasttext == 0.8.3
kenlm
Run:
Simply run python main.py
Parameters:
parameters could be found in the Config() in main.py.
Some important parameters for evaluation are:
discriminator_method : the type of discriminator ('Multi' or 'Cond')
data_path = './data/yelp/' (or amazon_new)
Results:
My evaluation results are saved in the folder results.