New dask-ml release to resolve upstream breaking changes?

[charlesbluca]

With Dask 2022.01.0 out, the breaking changes discussed in #895 are now out in a stable Dask release. These were resolved by @ayushdg in #898, but have not yet been published in a stable release.

Would it be possible to do another dask-ml release soon so that these breakages are resolved for downstream libraries? For context, this is impacting dask-sql's CI, which depends on the latest stable dask and dask-ml (example of a failing run):

• https://github.com/dask-contrib/dask-sql/runs/4827398398?check_suite_focus=true

cc @TomAugspurger @jrbourbeau

[TomAugspurger]

Yeah, I can do a release now if there's no blockers.

For reference, the release process just requires creating a new GitHub release with the desired tag.

[charlesbluca]

One thing to consider is the upstream failures coming from scikit-learn/scikit-learn#21880 recorded in #894 - it looks like the latest stable release of scikit-learn (1.0.2) doesn't yet have the breaking changes, but its nightlies do. We might want to address this now by either:

• making 1.0.2 the max version for scikit-learn
• adding a special case for scikit-learn>1.0.2 so that we don't import _check_param_grid

Or we could address it later on when this change is in a stable release of sklearn since dask-ml's release process seems pretty simple - currently checking to see if it's possible to make a fix for this now

[charlesbluca]

Think we should be able to resolve the breakage with a pretty small change - opened up #901 attempting to do this

[TomAugspurger]

#901 is merged. Anything else?

[jrbourbeau]

All known upstream failures have been resolved (xref #902 (comment)), so +1 from me

[charlesbluca]

Nothing else needed on dask-sql's end - +1 on releasing

[charlesbluca]

Ah I see that the release is out - thanks for the help @TomAugspurger! Will reopen if something else comes up for any reason.