Ok, thanks. I did use approx=“FITC” in my code so I assume that was an oversight when you said FITC might have a proper likelihood.
Fundamentally, I don’t understand much of this theory so I’m not equipped to implement something like calculating the likelihood from scratch.
Let’s say I just want to compare these two models and decide which is better. I think the long-term answer is: “go learn the theory.” Check…understood. But what is my short-term answer? Are there any easy one-liners I can add to get some kind of model comparison? My only idea for a short-term fix would be implement a k-fold (with small k!!!) manual cross validation. Obviously that would blow up the run time because MCMC is slow.