tips.txt

before you use:
    -please cite this work if you found it useful. details for citation can be found in README.md
    -please provide citations for any datasets you use for your work. E.g. if you are using StylePTB's (https://github.com/lvyiwei1/StylePTB) voice, tenses and ppr dataset from our ./data,
    please provide a citation for the original author's work.
    -please use these models responsibly and comply with guidelines for ethical use of AI. (https://www.acm.org/code-of-ethics)

tips:
    -the outputs that were used to compute statistics reported in the paper along with preprocessed datasets used are available at:
    https://drive.google.com/drive/folders/1ZvAPAfd_pGFb37vA9gOyTBUiEOgPmHBJ?usp=sharing
    -if you are using any of the pre-processed datasets from this work, please cite the original authors.
    - to compute content preservation metrics, use: https://github.com/Maluuba/nlg-eval
    - to compute naturalness or fluency metric, use: https://github.com/passeul/style-transfer-model-evaluation
    -Use the following batch sizes while training
        yelp: batch_size 256
        snli: batch_size 256
        dnli: batch_size 256
        qqp: batch_size 256
        scitail: batch_size 32
        voices: batch_size 32
        tenses: batch_size 32
        ppr: batch_size 32
    - for all other hyper parameters, use those mentioned in paper or the default args in every script.
    - train for epochs = 30 and test on last 10 checkpoints i.e. model.pt20-30 to determine best model.
    - if you want to train a brand new model with a newly created vocab, delete the original dir so that the old vocab.txt is not reused.
    - if you want to run EPAAE for your own dataset, preprocess your dataset to follow the same format as any one of the already available pre
    processed datasets from this work. Place it in ./data and call it according to examples mentioned in README.md.

contact:
    -please contact me for any help or questions: sharan.n21@gmail.com
    -report any bugs or questions on the "Issues" page and we will respond as soon as possible.