Unpickled models may fail to load if moved to different location about webstruct HOT 9 CLOSED

I'm using the following hack for loading NER classes:

def load_webstruct_model(path):
    model = joblib.load(path)

    # fix model filename path
    crf = model.steps[-1][1]
    folder = os.path.dirname(path)
    crf.model_filename = os.path.join(folder, crf.model_filename)

    if not os.path.exists(crf.model_filename):
        raise Exception("%s must exist" % crf.model_filename)
    return model

from webstruct.

why move the wapiti model to another place? maybe we should raise a exception earlier if the wapiti model is not there or check the checksum in _load_model?

from webstruct.

If I train a model locally and then upload it to the server it doesn't work - this is the issue.

from webstruct.

+1 got hit by it a couple of days ago, I have a workaround but it will be nice to see this merged. 👍

from webstruct.

Hi @shreyu86 - do you use latest webstruct? The referenced commit is already merged. Models based on CRFsuite don't have this issue, and it should also work fine now with Wapiti-based models if I'm not mistaken. I wonder why haven't I closed this ticket..

from webstruct.

To make it work don't pass model_filename argument explicitly.

from webstruct.

Ah I see, I am using latest webstruct I use wapiti with model_filename I will check and let you guys know. Thanks!

from webstruct.

It works! Thanks @kmike

from webstruct.

Thanks for poking :) Let's close this ticket.

from webstruct.