Extension 'has_coref' already exists on Doc. about neuralcoref

[humehta]

If I restart the or reload the program it works fine for the first time and gives this error the second time I execute it!!

[thomwolf]

Indeed, that arise when you do several successive en_coref_sm.load() (see my longer discussion in #52 ). You can try to load the model only once and pass around the nlp object to fix it for now.
This case seems to be quite common so I will add something to fix it. I keep you updated.

[thomwolf]

Ok, I have updated the models, it should be fixed now. Please try the new model and tell me if it doesn't for you.

[humehta]

Okay i'll try it!!!

[humehta]

It still gives me the same error when I try executing the second time.

[thomwolf]

Hi @humehta, can you execute the following script and tell me what it displays:

import en_coref_sm
with open(en_coref_md.__file__.replace('__init__.py', 'neuralcoref/neuralcoref.pyx')) as f:
    l = list(f.readlines())
print(l[525])

If you have the latest version of neuralcoref installed, it should print something like:
if not Doc.has_extension('huggingface_neuralcoref'):

[humehta]

Yes it gives me the following line:
if not Doc.has_extension('huggingface_neuralcoref'):

ALso you have imported the wrong model

[humehta]

Hey @thomwolf any updates?

[thomwolf]

Well I must say I have no idea what happening on your system. I can't seems to reproduce this bug with the latest version that you seems to have too (the one that says if not Doc.has_extension('huggingface_neuralcoref'):).

Can you give me a detailed walk-thought of all the steps to reproduce the bug?

[humehta]

I try running it on my jupyter notebook the first time I run and it runs successfully but when I edit the sentence or in fact run it the second time it gives the above error.

[humehta]

@thomwolf

[thomwolf]

can you show me a picture or a detailed script of what command you are running?

[thomwolf]

Oh another idea, maybe try reinstalling a clean version of NeuralCoref without using the cached version by using this command:

pip install --upgrade --force-reinstall --no-cache-dir MODEL_URL

[thomwolf]

This should also be fixed in the latest release (4.0).