Hilde Weerts

> I'm a little confused as to why this broke in the first place. What changed? I *think* the problem is that we didn't specify a version for scikeras: https://github.com/fairlearn/fairlearn/blob/94229647dda802a59bcbfe7f6d64295d64e96efa/test_othermlpackages/conda-tensorflow.yaml

> TF is always a pain for people who test against it. It's really much better to avoid it. Also, TF is a would where it kinda makes people stay...

I can imagine that the repo can be a bit intimidating for new folks looking to contribute to the project, although right now there might not be enough information on...

For a first non-outdated version I'd be very happy with the layout suggested by @ojeda-e. WDYT @adrinjalali?

Ah yes, I agree. @ojeda-e do you feel comfortable adding these suggestions? As indicated by @adrinjalali this might not be such a good first issue after all, so if you...

Awesome, @ojeda-e! Let's give it a try then :)

> @riedgar-ms evidently people are unsure what to do with these errors so we should at least surface which metric failed to process the input data. Additionally, this is an...

> My only concern is making sure that returning NaNs doesn't mess up `group_min()` etc. Ah yes good point. I think we should be able to figure something out, but...

I really like @LeJit's suggestion to first provide a more intuitive high-level explanation before diving into the specifics!

I would be ok to include this. The only thing I don't like about it is that e.g. a very low FPR could 'make up' for a very high FNR,...