search

hipster-philology / pandora

A Tagger-Lemmatizer for Natural Languages
MIT License
9 stars 4 forks source link

Opt-in epoch step results evaluation with CLI #20

Closed PonteIneptique closed 7 years ago

PonteIneptique commented 7 years ago

Make the printing of tests during training optional. Might be related to #8

Jean-Baptiste-Camps commented 7 years ago

Would it not be good too to be able to test only every nth epoch ? Training takes some time, and maybe we don't need to do it each time.

PonteIneptique commented 7 years ago

I have not looked at the repercussion of this. Maybe it can be another issue, once this one is solved ? ;)

PonteIneptique commented 7 years ago

@mikekestemont I am not sure, looking at the code, that dev_prev and train_pred are used by the NN. Could you confirm this ?

Jean-Baptiste-Camps commented 7 years ago

Ok, the current code causes an issue, or, at least, CLI and documentation need to be modified. If I do

python3 main.py config_test.txt --dev data/test/dev --train data/test/train --test data/test/test

I get the error

raise ValueError('Please do not call .test() if no test data is available.')
ValueError: Please do not call .test() if no test data is available.

Though 'test' is not empty. It is, I think, because of:

def main(config, train, dev, test=None, load=False, verbose=True, **kwargs):
PonteIneptique commented 7 years ago

I do not see how this would have changed this. Can you show the config_test.txt content ?

PonteIneptique commented 7 years ago

I'd say it's based on the loading of a precedent config file where include_test could be none. Bue I'd need to see your config_test.txt

Jean-Baptiste-Camps commented 7 years ago

Sorry, meanwhile I saw this was because of my config.txt.

mikekestemont commented 7 years ago

Are you sure dev_prev and train_predare the exact variable names used? In which file?

PonteIneptique commented 7 years ago

See my answer in the correct issue https://github.com/hipster-philology/pandora/issues/26#issuecomment-331795120