Should we review this?

Maybe we should add some comments to the line of code because to me it is not clear why it causes so many test failures :-/

I think the dare resulted from #462 (comment) (I am a bit slow today :)

It is more complicated than that, as we should also deal with the normalize_y logic I suppose, as in https://github.com/scikit-learn/scikit-learn/blob/ec0dbf0399c4187a1d17e66de17265c67588373f/sklearn/gaussian_process/gpr.py#L183

Right, thanks. I'd not checked the tests.

You mean you want it to store zeros if not normalize_y?

This change isn't passing tests locally. I might leave it to you folks to fix.