Broken test due to a change in scikit-learn's AUC

[janezd]

Hard-working maintainers of Python scientific stack decided to change the result of AUC when only one class is present in test data: scikit-learn/scikit-learn#27412, merged two days ago (here: https://github.com/scikit-learn/scikit-learn/pull/27412/files#diff-4eb3c023f8a3f088d62208f6adbd02b6df5196de2257ccd228dffc972c964634R383).

Yes, changing None to 0 indeed does not make a huge difference, does it? Well, except for breaking our tests and taking me half an hour to discover who's responsible (scikit? sklearn? pandas?) and what changed and why. Now I only need to make our tests version-dependent and we're fine again. Thank you, thank you, thank you.

I do not have a habit of swearing, but I might start if I keep trying to keep the pace with the ever-changing scipy, sklearn and pandas.