FIX ‘sparse’ kwarg was not used by fowlkes_mallows_score

[cynddl]

Reference Issues/PRs

No issue.

What does this implement/fix? Explain your changes.

The function sklearn.metrics.fowlkes_mallows_score was defined with:

def fowlkes_mallows_score(labels_true, labels_pred, *, sparse=False):

but the code never uses sparse. The parameter sparse should have been passed to contingency_matrix, but instead we currently see a few lines later:

    contingency_matrix(labels_true, labels_pred, sparse=True)

This commit fixes the inconsistency by:

• changing the default to sparse=True so that it doesn't break any test or downstream use
• making sure that contingency_matrix uses the given sparse parameter.

Any other comments?

The function sklearn.metrics.fowlkes_mallows_score has never been tested with sparse=False, it might be good to have a few tests for that use case too?

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 9b0be6c. Link to the linter CI: here}

[prabashbala]

would it be possible to add a test for this please.

[adrinjalali]

Thanks @cynddl . Could you please add a test as well?

[adrinjalali]

I wouldn't change the default here though.

[dschult]

Setting sparse=False after this PR would break the algorithm. It is set up for computing with a sparse contingency matrix.

Instead of "making this keyword argument work" I think it makes sense to "remove his keyword argument". It is not used, has not been used and provides no advantages.

[adrinjalali]

That's true, the code actually relies on the output (c) being sparse. So we can deprecate the parameter here.

[jeremiedbb]

I made the requested changes to deprecate the param instead.

LGTM.