register LKJPrior if it's set.

[Jimbo994]

Motivation

closes #745

now the LKJPrior is registered if it is specified as a task prior.

Have you read the Contributing Guidelines on pull requests?

Yes

Have you updated CHANGELOG.md?

Yes

Test Plan

Added tests.

[Jimbo994]

@jpfolch can you have a doublecheck if this correctly sets the LKJPrior? If yes @jduerholt can you review this?

[jduerholt]

Looks good to me in general, but do you know if the original issue is fixed? The one with the LKJ Prior mentioned in the comment?

[Jimbo994]

Yes, so I now test that the LKJPrior is set, and test if the model is successfully fitted using model.fit, which uses fit_gpytorch_mll, and this works. So with that I assume it is fixed.

[jpfolch]

Here's the original issue which remains open, with some code snippets that claim to fix it here, but not sure if it was merged. Maybe it was fixed in some other PR but I can't find where exactly.

[Jimbo994]

Ah, thanks for the context @jpfolch, I just checked, and it is indeed not solved/merged! If the model fits correctly on the first try, no error is thrown. If it doesn't, then sample_all_priors is called, and this still throws an error. When I ran the test I added 20 times, it sometimes fails the first fit and then indeed throws this error.

Let's maybe leave this PR open and keep an eye out for fixes in botorch.