Scikit-learn: Minimal numpy and scipy versions for 0.21

Created on 27 Sep 2018 · 3Comments · Source: scikit-learn/scikit-learn

Assuming the next 0.21 release happens around or after April 2019, the oldest Ubuntu LTS at that point will be 16.04 LTS, which includes,

numpy 1.11.0
scipy 0.17.0

(checked in a Docker image, not sure where to find the official information)

so we could probably make these our minimum requirements for 0.21.

Making so would for instance allow to implement https://github.com/scikit-learn/scikit-learn/issues/11638

Even assuming there is agreement on this subject, we should probably wait for 0.20.1, at least, before doing anything about it. Eventually this would involve,

[ ] adjusting versions of dependencies in docs and CI
[ ] removing the utils.fixes that are no longer needed.

Source

rth

👍4

Most helpful comment

I would also be in favor of bumping dependencies for scikit-learn 0.21 to those matching ubuntu 16.04 LTS so as to be able to implement what is described in #11638.

ogrisel on 28 Sep 2018

👍3

All 3 comments

I would also be in favor of bumping dependencies for scikit-learn 0.21 to those matching ubuntu 16.04 LTS so as to be able to implement what is described in #11638.

ogrisel on 28 Sep 2018

👍3

+1 on bumping and +1 on waiting for 0.20.1

amueller on 28 Sep 2018

From #12718
We should also consider the version of matplotlib and scikit-image
It's 1.5.1 for matplotlib (https://packages.ubuntu.com/xenial/python3-matplotlib) and 0.10.1 for scikit-image (https://packages.ubuntu.com/xenial/python3-skimage).