Yale-LILY / SummerTime

An open-source text summarization toolkit for non-experts. EMNLP'2021 Demo
https://arxiv.org/abs/2108.12738
Apache License 2.0
268 stars 30 forks source link

troyfeng116/requirements #16

Closed troyfeng116 closed 3 years ago

troyfeng116 commented 3 years ago
niansong1996 commented 3 years ago

@troyfeng116 have you tested this out in a new conda environment? use pip install -r requirements.txt and see if the code runs?

Also, I am not sure about what it means for @ file://, is it installing from local builds?

troyfeng116 commented 3 years ago

@troyfeng116 have you tested this out in a new conda environment? use pip install -r requirements.txt and see if the code runs?

Also, I am not sure about what it means for @ file://, is it installing from local builds?

@niansong1996 Just pushed a commit fixing pip install -r, everything works in a new environment

troyfeng116 commented 3 years ago
niansong1996 commented 3 years ago

@troyfeng116 Do pytextrank and summeval depend on different versions of spacy? can we find something that works for both?

troyfeng116 commented 3 years ago

@troyfeng116 Do pytextrank and summeval depend on different versions of spacy? can we find something that works for both?

No, they're completely incompatible. I think summeval requires <2.3.0 and pytextrank (specifically en-core-web-sm model) requires >3.0.0. Still works and runs if you manually install spacy >3.0.0 as the final pip installation (manual pip installation of en-core-web-sm), but a regular pip install -r can't resolve the conflict

niansong1996 commented 3 years ago

hmm, I see. The good news is that I now have admin access to summEval as well so we can update it on that end as well. That <2.3.0 constraint seems rather strange to me, maybe that is also something imposed by one of their own dependencies.

That being said, is this problem basically blocking the automatic installation or if you can find a way around it for now?

troyfeng116 commented 3 years ago

hmm, I see. The good news is that I now have admin access to summEval as well so we can update it on that end as well. That <2.3.0 constraint seems rather strange to me, maybe that is also something imposed by one of their own dependencies.

That being said, is this problem basically blocking the automatic installation or if you can find a way around it for now?

Yeah, we won't be to run a single pip install -r with that conflict. I've been looking into whether 1) we can use a different spacy model for pytextrank (i.e. not en-core-web-sm) or 2) if there's another good TextRank implementation that doesn't have conflicting dependencies

niansong1996 commented 3 years ago

okay, sounds good to me. In the meantime, I will see what I can do with SummEval

niansong1996 commented 3 years ago

Merging this PR as a temporary solution.