David Widmann

I'll close this PR for now. Please reopen it if you intend to address https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/pull/329#issuecomment-1191856948 :slightly_smiling_face:

Thanks for opening this PR! > I would consider this a bugfix, the only thing that would break is already-broken usage. I disagree: It breaks working code that operates with...

Was anything changed? It seems on the master branch it is around (or slightly below) 20 minutes currently: https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/actions/runs/3134043689/jobs/5088108942

I guess a reason for why there is no API for destructuring a `PosteriorGP` is that [it is not part of the public API](https://juliagaussianprocesses.github.io/AbstractGPs.jl/dev/api/#Internal-AbstractGPs-API): > This functionality is not intended...

It's not necessary to compute it if you don't call `rand` or `logpdf` :shrug: One approach that doesn't touch the structure of `FiniteGP` would be to add a method `MvNormal(::FiniteGP)`...

E.g., it is not required if you compute the marginals. Or more generally just the mean and covariance matrix. > Maybe we have a @require in MeasureTheory that adds the...

Haha, my suggestion was `Distributions.MvNormal` since `FiniteGP` is already a subtype of `Distributions.AbstractMvNormal` :smile:

> BTW, is there a way to avoid using PDMats? I'd rather allow positive semidefinite coavariance. You can use `PSDMat` in https://github.com/invenia/PDMatsExtras.jl.

Oh, I completely forgot that I had made a draft locally during the discussion last week. I'll open a PR.

I got the impression that it does not really fit into this package, even though it is intended to be a replacement of the Iris dataset since there are multiple...