Need for a review of my GP tutorial

My package is just using Pymc3 as the only backend (for now, hope to expand it to GPFlow), so it can’t really do anything that Pymc3 can’t. I just implemented the bwengels example there (linear model of coregionalization, I believe?), but you’re right that it’s not immediately obvious how/whether you could use structurally distinct but correlated kernels for the two outputs.