[MRG] Support for list of lists or list of arrays multilabel indicator (continuation)

[leonardbinet]

Reference Issues/PRs

Rebase of #9158 to solve #7931

What does this implement/fix? Explain your changes.

Any other comments?

[jnothman]

Yes, let's do this!

[jnothman]

Test failures...

[leonardbinet]

@jnothman, fixed it

[jnothman]

Thanks, it's looking pretty good.

Please add an Enhancement entry to the change log at doc/whats_new/v0.22.rst. Like the other entries there, please reference this pull request with :pr: and credit yourself (and other contributors if applicable) with :user:

[jnothman]

Please append commits rather than force pushing. Much easier to see changes between reviews that way

[jnothman]

Might be more helpful to advertise the changes in terms of how they affect metrics and other users of type_of_target

[leonardbinet]

how would you put it? one entry per impacted function or a more generic explanation on what is considered as multiclass-multioutput/multilabel-indicator?

[jnothman]

I've not looked at the code or codecov's complaint this time

[jnothman]

I think this mostly affects sklearn.metrics, so commenting there that "multilabel metrics now accept ..." would do; or add a "multiple modules" section to the change log.

[leonardbinet]

thanks for your help and patience @jnothman, you're the victim of my first contribution to scikit-learn 😅

[jnothman]

I'm very happy you're finishing this off. It's something we should have done years ago, having removed sequence of sequence support around 2015

[jnothman]

sklearn/preprocessing/tests/test_label.py:27:1: F401 'sklearn.utils.testing.assert_raises' imported but unused
from sklearn.utils.testing import assert_raises, assert_raise_message

[jnothman]

Otherwise lgtm

[jnothman]

It looks like you've introduced a syntax error. And it's worth looking at the codecov error

[jnothman]

Maybe merging in the latest master will fix the codecov issue

[NicolasHug]

Thanks @leonardbinet , only super nitpicks from me

The PR has merge conflicts but apart from that LGTM

[NicolasHug]

question: is this a bugfix?

[jnothman]

It is more-or-less a bug fix. Either way, type_of_target is expected to return "the most specific type that can be inferred". Multilabel is a specific subtype of multiclass-multioutput. So this is correct

[jnothman]

@leonardbinet please resolve conflicts with master and address @NicolasHug's teeny weeny comments so that we can merge this for release. If that's looking unlikely in the next week or so, please let us know.

[leonardbinet]

@jnothman @NicolasHug sure, thanks for your feedbacks I'll fix this this week-end

[jnothman]

@leonardbinet, force pushing makes it much harder for reviewers to track what changes have occurred since last review.

[jnothman]

Thank you @leonardbinet, @herilalaina and @srivatsan-ramesh!!