The rest of failures are related with #17707 (comment).

The rest of failures are related with #17707 (comment).

There is also the following deprecation warning in test_pairwise_parallel[float64-array-pairwise_distances-wminkowski-kwds1]:

DeprecationWarning: scipy.distance.wminkowski is deprecated and will be removed in SciPy 1.8.0, use scipy.distance.minkowski instead.

The rest of failures are related with #17707 (comment).

There is also the following deprecation warning in test_pairwise_parallel[float64-array-pairwise_distances-wminkowski-kwds1]:

DeprecationWarning: scipy.distance.wminkowski is deprecated and will be removed in SciPy 1.8.0, use scipy.distance.minkowski instead.

Good catch @ogrisel 😄.

I will have a look to the rest of failures to see if there is any DeprecationWarning missing.

Still one missing failure in test_exactly_zero_info_score:

E               AssertionError: assert 2.2207031250001146 == 0.0 ± 1.0e-12
E                +  where 2.2207031250001146 = adjusted_mutual_info_score(array([1, 1, 1, ..., 1, 1, 1]), array([   0,    1,    2, ..., 9997, 9998, 9999]), 'min')
E                +  and   0.0 ± 1.0e-12 = <function approx at 0x7f579ae87680>(0.0)
E                +    where <function approx at 0x7f579ae87680> = pytest.approx

Another two in test_not_symmetric_metric[precision_recall_curve] and test_binary_sample_weight_invariance[precision_recall_curve]:

E       ValueError: cannot convert float NaN to integer

@thomasjpfan I think this is already a net improvement. Shall we merge and open new PRs for fixing remaining failures?

@alfaro96: Do you know if the test_not_symmetric_metric[precision_recall_curve] and test_binary_sample_weight_invariance[precision_recall_curve] failures you noted above have been fixed on master? I'm seeing the same failures with 0.24.1 when running with numpy 1.20.

Sorry for the noise. It looks like #19101 is the patch I was looking for.