I ran the tests in test/test_utils.py, and everything seems to be working good. Actually, the current version of metric-learn was throwing errors in these tests even before it broke, because of scikit-learn deprecation warnings.

For reference, this branch passes all tests with scikit-learn 1.8, while main fails a lot of tests.

Interestingly, when using scikit-learn 1.7, both this branch and main throw a bunch of errors saying e.g. FAILED test/test_utils.py::test_check_tuples_valid_tuple_size[2] - assert 4 == 0. This is because in both cases, the pattern below

  with warnings.catch_warnings(record=True) as record:
    check_input(tuples, type_of_inputs='tuples', preprocessor=None)
  assert len(record) == 0

I looked into one of these tests, and as I supposed, it was a FutureWarning from scikit-learn, warning about ensure_all_finite:

is catching `scikit-learn` warnings:

sklearn/utils/deprecation.py:132: FutureWarning: 'force_all_finite' was renamed to 'ensure_all_finite' in 1.6 and will be removed in 1.8.
warnings.warn(

In sum, it the current PR makes `metric-learn` work with `scikit-learn` 1.8, at least in the tests I ran and the use cases I tested for my own work.

Thanks! I had to manually enable the CI to run on this PR, but it started now.

There are plenty of errors in the CI, but at least at a first approximation they don't seem to be cause by this PR.

For example, when running test/test_sklearn_compat.py locally with scikit-learn 1.8 and the current branch, I get the same errors as the CI. But I get even more errors with master, because of the issue this PR fixes. When using scikit-learn 1.7, I continue to get the exact same errors with both branches.

It is difficult to parse this CI output to try to identify issues with this PR when it's not clear (at least to me) what the CI looks like for master. But there do seem to be plenty of errors that require further investigation, not directly related to this PR. Maybe a larger effort to bring metric-learn up to date with scikit-learn is needed.

What do you think @perimosocordiae?

Also, I ran test/test_sklearn_compat.py with scikit-learn 1.5 and both branches pass all the tests. I didn't try 1.6, nor other testing suites yet because they take some time locally.

Great, thanks for the contribution! (And sorry for the delay)