David Widmann

Did you do compare performance with Zygote?

That's a quite noticeable regression. Do we know what exactly causes it?

ReverseDiff does not use `rrule`s automatically. You have to "import" them with `@grad_from_chainrules`: https://juliadiff.org/ReverseDiff.jl/dev/api/#ReverseDiff.@grad_from_chainrules

> if e.g. Optimisation.jl is loaded, these helper functions become automatically available. Unfortunately, that's not possible with extensions - you can't use them to define new functions that are available...

> What are the requirements for a fully functional kernel that can be used in AbstractGPs? I guess you can mainly copy `KernelTensorProduct` and replace multiplication with addition. > would...

Maybe this could even be part of AbstractPPL and be defined on `AbstractPPL.AbstractProbabilisticProgram`: `condition` is part of its API, only `rand` is not clearly specified there yet (probably should be...

> Or provide an API for specifying the return type, like rand does (but supporting two optional positional parameters rng and T complicates the interface) Adding `T` to `predict` (with...

One problem is the following: ```julia julia> struct Matern{T} end julia> Matern{1.5}() isa Matern{3/2} true julia> Matern{1.5}() isa Matern{3//2} false julia> Matern{1.5f0}() isa Matern{3/2} false ``` Floating point numbers of...

I did these calculations a few weeks ago and wrote about it in my thesis 😅😅😅 It's quite simple and there exist multiple references (eg., the GPML book). But I...

Can't we just add a branch in the general kernel, in case we want to exploit this formula (I'm still not convinced that it's better than just directly evaluating the...