search

rpryzant / delete_retrieve_generate

PyTorch implementation of the Delete, Retrieve Generate style transfer algorithm
MIT License
132 stars 26 forks source link

preds file has many meaningless tokens. #22

Closed 27182812 closed 3 years ago

27182812 commented 3 years ago

I tried my own data which similar to positive and negative data. Though epochs I trained are not many，but the result is extremely bad，with lots of meaningless tokens. Can you help me? Thank you very much! image

27182812 commented 3 years ago

What confuses me most is that the original experiment, such as yelp, is basically complete sentences in pred1.

rpryzant commented 3 years ago

Hmm how large are your data, how many epochs are you training the model for, and at what learning rate? It seems like the model hasn't converged. I would recommend trying out different combinations of hyperparameters to achieve a better fit.

27182812 commented 3 years ago

Thank you very much! I solved it.

wasedaward commented 3 years ago

Thank you very much! I solved it.

Sorry for interrupting you. Can I ask about your combinations of hyperparameters? Because I met the same trouble. Thank you in advance.