Roman Lutz

I'm a bit confused. I just checked the website and I don't see the N/A there.... Still shows the old string:  The [website repo](https://github.com/fairlearn/fairlearn.github.io) looks like there have been...

Update: CircleCI's deploy step failed. Rerunning right now to see if that fixes it: https://app.circleci.com/pipelines/github/fairlearn/fairlearn/2658/workflows/b148e4aa-653b-4a6c-b6d8-29fc7a5f7888/jobs/5080

It started with #1046 @riedgar-ms Did you try running it with the changes you made? It seems like it never succeeded after this PR.

I'm guessing something is missing in that image that was there in the previous one resulting in the failure of npm install (?) We could copy some of the steps...

According to Miro this may be because it's at the edge of the grid where one of the classes is ignored, and that may lead to imbalance resulting in multiple...

Just to add to this: the "fix windows notebook" change that I just merged does not solve the overall problem, it just adds the special case handling back. Somehow it...

To my knowledge this occurs in py3.6 as well

Of course they are, because we put a special case in the test :-) ``` # TODO: investigate where the different outcomes for the first grid point are from, likely...

https://github.com/fairlearn/fairlearn/blob/b07209058322d36e0962e25f82cd720ada650cde/test/unit/reductions/grid_search/test_grid_search_regression.py#L82

When deprecating Python 3.6 in #966 we should consider removing this special test case assertion. Hopefully, it works deterministically on all Python versions that are more recent than 3.6