scikit-learn 1.0+ compatible release on PyPI

[auvipy]

I will be happy to contribute

[eddiebergman]

Hello @auvipy

It's on our todo list once we get some other PR's that we are currently working on, if youd like to start, we would be very much appreciative.

Doing so would likely require a decent amount of work, more than just updating the requirements file. It would involve going through the changelog and updating each component as neccessary and updating any test values that would change. At that point, we have external tests that validate that autosklearn does not degrade performance on automlbenchmark which we would perform externally from github.

@mfeurer, do you ideas of what else would be required?

Best,
Eddie

[auvipy]

ok i am also up for coding changes as well

[eddiebergman]

Thanks @auvipy,

I'll help guide you along the way and answer questions so feel free to open a PR and continue discussion there!

Best,
Eddie

[mfeurer]

Hey @eddiebergman we could actually also try a more gradual path in which we first allow both scikit-learn version in parallel and update the individual components over time?

[Rattko]

@mfeurer, @eddiebergman is there any timeline for this issue?

[auvipy]

I'm willing to restart the effort

[eddiebergman]

Hi @auvipy,

Please see #1677, this update will already include the updated scikit-learn :)

Best,
Eddie

[auvipy]

the latest PyPI release still using old scikit-learn

[eddiebergman]

Yes, it's not released yet, just saying that the refactored auto-sklearn will already use the latest scikit learn (once it's released)

[auvipy]

are they work in progress? shouldn't we consider any external contribution to the the project in the time being?

[eddiebergman]

We can, but we would hesitate actually encouraging it, as updating sklearn has already been done in the refactor and so has dropping Python 3.7.

Apologies for having it private as there are reasons outside our control for doing so until it is released. I will update the main issue once there is an update.