installing medspacy in a clean environment leads to a numpy binary compatibility error when importing

medspacy / medspacy

Library for clinical NLP with spaCy.

MIT License

535 stars 91 forks source link

installing medspacy in a clean environment leads to a numpy binary compatibility error when importing #305

Open barney-ellis opened 1 month ago

barney-ellis commented 1 month ago

When I install medspacy in a clean environment (Python 3.11) and try importing it in a script, I get a binary incompatibility error. I believe this is because the spacy dependency is still pinned to <3.6, which is constrained to older version of thinc:

Pinning spacy to the latest version (3.8.2) fixes this issue for me. Has anyone else experienced this issue?

burgersmoke commented 3 weeks ago

We're looking into this, but it could be related to this: https://github.com/explosion/spaCy/issues/13669

burgersmoke commented 3 weeks ago

@barney-ellis can you tell us which version of numpy is installed in your environment?

burgersmoke commented 3 weeks ago

It looks like our continuous does fail with this error. So we might consider allowing newer version of spacy now:

https://github.com/medspacy/medspacy/actions/runs/11505912339/job/32028669661

barney-ellis commented 3 weeks ago

Thanks @burgersmoke, when I install medspacy using pip in a python 3.11.5 environment my numpy version is 2.1.2 and I get the above error. Pinning spacy to 3.8.2 my numpy version is 2.0.2 and I no longer get the error