Will Tebbutt

> But wait SelectTransfrom((1, 3)) would take a N dimensional vector x and only return x[[1,3]], a 2-dimensional vector right? So the input to SEKernel would be Euclidean(2) ? Sorry,...

The only way I can imagine that it would not be possible to derive a kernel based on the final domain would be something like ```julia k1 = SEKernel() k2...

Overall, I'm of the opinion that parametrising the input type into the kernel would not be the right move. This is for the following reasons: - I can't see any...

> On the other hand, I do think there is value in including additional type parameters in distance-based kernels: specifically, both the distance used to create it, and the dimension...

Hi @simonschoelly . Please do consider making your graph kernels subtype KernelFunctions.jl's `Kernel`! I would love to be able to just plug your kernels into a GP from AbstractGPs. >...

> The idea here is that one that implements a graph kernel then just implements these two functions. Other functions such as kernelmatrix also work with these two functions I...

To be honest, in the absence of a really clear example, I'm not sure what this gets us over just using a `Vector{Tuple}`, so my inclination would be to stick...

I'm opening this as this seems to be part of a larger discussion, and I don't want it to hold up #304 . @devmotion pointed out that this is related...

@nickrobinson251 would you mind weighing in on this? My impression is that you have opinions on the various conventions for `show` with different numbers of arguments (I never quite grasped...

I would also be very keen to make this use-case possible. At the minute, it's unclear to me how one should automatically install JuliaUp in a non-interactive manner, which is...