Basic compute_metrics function fails on Google Colab

[seungguini]

Hey,

I've wanted to test the code on Google colab, but the code throws an error

Here's the code I'm trying to run:

from nlgeval import compute_metrics
metrics_dict = compute_metrics(hypothesis='nlg-eval/examples/hyp.txt', references=['nlg-eval/examples/ref1.txt', 'nlg-eval/examples/ref2.txt'])

Here's the Error log:

I've also made the Google Colab link publicly available - please check it out.

https://colab.research.google.com/drive/1iLiy8bQcR65ll6HufAvxzXExk7kn5Ls2?usp=sharing

I'd appreciate any help / fixes! Thanks.

[juharris]

Thanks for trying nlg-eval! I was able to reproduce this with your notebook, so thanks for sharing that. I didn't see anything immediately crazy wrong. I was looking to make sure that Theano and gensim did get installed and their versions. The Theano version was 1.0.5 but I tried !pip install 'Theano~=0.8.1' and I still got the same error.

[AbhilashPal]

Hi, so I was also facing the same error. For me, the pytest run is also throwing a few errors.

Here's the code with the issue :
https://colab.research.google.com/drive/1nGfXh3Ae4vOxjQW039RsjAqgOF_fjsrZ?usp=sharing

[seungguini]

@juharris Thanks for looking into the issue. nlg-eval seems to work fine on my local computer. Perhaps it's a compatibility issue with Google Colab?

[juharris]

Yeah I think it's using the wrong version of one of the dependencies somehow.