jimth001/formality_style_transfer_with_shared_latent_space

The repo of the paper "Formality Style Transfer with Shared Latent Space"

formality_style_transfer_with_shared_latent_space's Issues

how to prepare the data ?

hello! can I ack you some questions about the data? In sls_settings_v2_FR.py file,

how to make the './new_exp_fr/embedding/embedding.bpe.big.txt' file? And is './new_exp_fr/informal.train.rule' a txt format file? Thank you very much!

Returning "nan" in running train stage

@jimth001

I trained (sls_rnn.py) this model on GYAFC dataset and pre-trained embedding 300d but it returned "nan" in running train stage which is shown in log.txt below.
furthermore, returned Zero Blue score in running test stage. the output of the model was all "UNK".

log.txt

output:
informal.semd.result.txt

why will this question happen? I get the same result

hello, thank you for answering me the question about "how to prepare the data" again. Now I want to ask you another question when I train and evaluate the data. The dataset is made by myself. I want to transfer one style to another style. I have 6000 for train, 2000 for validation and 2000 for test parallel data. But when I print the inf2fm data in the file sls_gpt.py in 559 line in batch_num: 100

I get the result:

I don't know why. When I run another experiment "Style Transfer Through Back-Translation. ACL 2018" . I also get the same result.

So I turn to you for help~

best wishes, thank you very much!!!!