Is the marginal likelihood of GP with white noise the same as having a normal distribution with mean as a latent GP? I ask this because the first case seems to sample faster than the second case. Wondering if this is because the latent variable is marginalized out in the first case.
I think they are the same, as for speed differences maybe @bwengals has some idea.
Yes, your guess is exactly right. If you check what’s in the trace you’ll see the GP there for the Latent case, but not the Marginal case.
Thank you very much.