Segmentation fault (core dumped) about neuralcoref HOT 13 CLOSED

I can verify that the entire SQuAD dataset passes through without problems.

Super fast too. 😄

Thanks for your work @thomwolf

from neuralcoref.

I have the same issue on OS X 10.13.4, using python3.6.4, neuralcoref-3.0 and en_coref_md.
After removing the first token ("Although") of the given error sentence it works normally.

From my side I first noticed the error message when using "As" as a first token:

import spacy
import en_coref_md
nlp = en_coref_md.load()
nlp("As")
Segmentation fault: 11

from neuralcoref.

I am working on it

from neuralcoref.

Thanks to everyone for helping here and giving examples (in particular @petermartigny and @theSage21).

I think the segfaults bugs are now fixed as well as the install scripts of the models. I am uploading new models (sm and md are already up, lg is still uploading). You can try them (just follow the install instructions of the readme).

There should be no need to install NeuralCoref (or even spaCy) separately, everything is bundled in the models. The quickest way to install them is pip install MODEL_URL in a clean environment and then use them with

import en_coref_sm # or en_coref_md or en_coref_lg
nlp = en_coref_sm.load()
doc = nlp("I have a dog, he is very smart")

Please tell me if there is an issue.

from neuralcoref.

I can confirm the issue too.

from neuralcoref.

Can you give me some additional information:
What system are you using? Windows/Mac/linux
Does it work on other examples?

from neuralcoref.

Same exact versions of Python, neuralcoref, and model as @theSage21. In addition, I'm on Ubuntu 16.04. It works seemingly intermittently. It works on many examples, but then not on others. I can get more failure examples with some digging (I'm currently parsing a large file, sentence by sentence).

from neuralcoref.

Can you share the file (or a part of it that doesn't work) with me? For example at thomas[at]huggingface[dot]co?

from neuralcoref.

My ubuntu version is at 18.04 LTS.

In addition I'm parsing SQuAD 2.0 as the dataset. Download link is here

A sample script to generate text which causes segfault

import json
import spacy

nlp = spacy.load('en_coref_sm')
SKIP = 1039

with open('../data/train-v2.0.json', 'r') as f:
    df = json.load(f)


texts = []
for wiki in df['data']:
    for para in wiki['paragraphs']:
        for qas in para['qas']:
                text = para['context'] + ' ' + qas['question']
                texts.append(text)

for index, text in enumerate(texts):
    if index <= SKIP:
        continue
    print(index, text)
    nlp(text)

This will print the last text which segfaulted. you'll have to run it again to find others though.

from neuralcoref.

I experience the same problem. Hope to see this solved.

from neuralcoref.

Mine works too. Thanks!

from neuralcoref.

Hello @thomwolf ,

Environment details: Ubuntu 18 and python 3.6
neural coref sample code from website causing segmentation fault. Could you please help!

from neuralcoref.

Hi @akshayparakh25, it's more convenient if you open a new Issue as this one is 1,5 year old and closed. From the log above it looks like this was an issue in neuralcoref-3.0 and got fixed. You haven't mentioned the version you're using but if it's 3.0, could you retry with 4.0?

If you keep running into issues - please open a new issue and provide more details (setup, versions, example data, example code).

from neuralcoref.