Arbitrary dimensions of input and output

[aseembits93]

Hi Vlad, Thanks again for sharing this repo. Being a noob at GPR myself, I wanted to get your opinion on how to use this library for arbitrary dimensions of inputs and outputs. I have 34 data points with 5 dimensional inputs and 11 dimensional outputs. I seem to get an error which suggests that dim(input)==dim(output). Any guidance would be appreciated. If you don't mind, can I email you regarding this?

[vlad17]

Someone actually reached out to me about this via e-mail. Not sure if it was you. Here was my response:

I've only implemented interpolation for up to two input dimensions. It's possible to extend this to three dimensions, but it's complicated. Here's the file where the interpolation occurs. For 5 inputs, I'd still expect interpolation to be reasonably efficient. But you'd need to write a recursive version of my 2d routine.

One suggestion is just to use dimensionality reduction on the input (via a learned projection, learned neural encoding, by which I mean deep kernel learning with a 2 dimensional neural output layer, or a fixed routine like PCA).

I'll leave this issue open for now. Both of these approaches would be welcome improvements!