search

dssg / aequitas

Bias Auditing & Fair ML Toolkit
http://www.datasciencepublicpolicy.org/aequitas/
MIT License
677 stars 113 forks source link

Weaken numpy requirement #195

Closed diehlbw closed 3 weeks ago

diehlbw commented 3 months ago

Addresses #183 by weakening the pinning on numpy.

While I agree in spirit with the comment on not having strict pinnings I tried not to increase scope of changes more than necessary to hopefully reduce back-and-forth

diehlbw commented 2 months ago

@sgpjesus - what is the process to get feedback on this? is it something that can be considered?

sgpjesus commented 1 month ago

Hi @diehlbw, sorry for the delay. The process is still not very defined yet, for now I wanted to finish the merge request on using poetry before advancing with any dependency changes!

sgpjesus commented 1 month ago

Hi @diehlbw, Thank you for your PR. We are working on extending the unit tests; it would be easier to see if weakening the numpy requirement would affect anything after that. Still, I ask you to pull the changes in Master, and update the pyproject.toml file.

diehlbw commented 3 weeks ago

Closing since this is effectively fixed with the latest commit.

Also want to convey some considerations that non-trivial updates to requirements files should often be considered breaking changes (ie correspond to major version bump, or at least a minor one depending on your update patterns). Environments that had been constructed to work with the old v1.0.0, may not work if other packages are pinned/constrained; much less code itself.