Some pre-trained weird warning about nmt-keras

[yftah89]

Thanks a lot :) you can close this subject

[lvapeab]

Hello @yftah89,
by default, the preprocess_binary_word_vectors.py stores the processed vectors in a file with the .npy extension. You should use this file in config.py. If the provided file in config.pydoesn't have the .npy extension, it warns you.
So please, make sure you are using the output from the preprocess_binary_word_vectors.py script.

Cheers! :)

P.S: I've just updated the preprocess_binary_word_vectors.py and preprocess_text_word_vectors.py for taking arguments. Maybe this eases their usage :)

[lvapeab]

Closing this, as it seems there is no activity.
Feel free to reopen it if you find any trouble.

[yftah89]

I run all the necessary scripts and I'm still getting this warning, I think the line which prints this warning is incorrect: if params['SRC_PRETRAINED_VECTORS'] and params['SRC_PRETRAINED_VECTORS'][:-1] !='.npy'
Why the last letter of the embedding path should be the word '.npy'?

In the config file, I have the following line -
TRG_PRETRAINED_VECTORS = "examples/w2v.npy"
Am I am doing anything wrong?

[lvapeab]

You are right. That was a (stupid) bug in that line (it didn't check the last 4 characters, but the last one). I've just fixed it.

Referred to your question, the file extension of the preprocessed word embeddings doesn't necessarily need to be .npy, as long as the given file had the right format and was readable by numpy. But, if the file is preprocessed by the provided scripts, it will have the .npy extension. Therefore, this is a warning that tries to ensure that you have correctly set the format of the word embeddings file.

Thank you for detecting this ^^