Ross Viljoen

I think I came across it when writing tests, so I don't have a practical use case in mind. The major benefit would be that BLRs could be more easily...

Whether or not we end up adding these specific parameterisations, I think it would still be good to add the `AbstractSparseVariationalApproximation` type anyway to allow people to add parameterisations like...

`mean(fx)` is a very easy fix - just need to replace `fill` with `FillArrays.Fill` etc. for mean functions. The problem with `cov(fx)` is that `broadcast(+, ::CuMatrix, ::Diagonal{

It looks like the reason it doesn't work is that GPUArrays can only broadcast over wrapped array types if the wrapped array is also an `AbstractGPUArray` - i.e. `Diagonal{

Another thing to do along with this: the naming used for covariance matrices and inducing points is a bit inconsistent (also in AbstractGPs). i.e. sometimes it's `Kxu`, sometimes `Cxu`, sometimes...

I've implemented a basic version of this [here](https://github.com/rossviljoen/PathwiseSampling.jl). There are essentially two parts to pathwise sampling: 1. An approximate sample from the prior GP (usually by using random fourier features)...

Test failures are unrelated to this PR. Mysteriously, they seem to be caused by https://github.com/JuliaStats/PDMats.jl/pull/149 - pinning PDMats to `0.11.7` solves them

Docs build will fail because of #105 EDIT: I think I will just remove the example from this PR and add it in a separate one once this is merged...

~@devmotion is this alright to be merged then?~ EDIT: my reading comprehension is apparently limited

Strange, in a fresh emacs 29.4 with the following `init.el` it works fine (installing Eat 0.9.4 from NonGNU ELPA): ```elisp (require 'package) (when (not package-archive-contents) (package-refresh-contents)) (package-initialize) (use-package eat :ensure...