Interpretation of posterior predictive checks for a Gaussian Process

Hi Bill,

Thank you so much for your help!

Definitely, I will try to look at gp.Latent (I think I’ve seen it).

As a separate check, I’ve tried the following:

Y_ = pm.MvNormal (“Y_”, mu=mu, cov=cov, observed=Y)

which is essentially the same as gp.Marginal - I got a very similar result.

The number of covariates is 2 times (i.e., ~300) the measurement. This means my state (latent) vector is also the size of ~300. My understanding is that in the Bayesian setting, this is not unusual. When I checked the convergence for each latent variable, all looked good, but do you think this could be another problem?

Again, thanks so much!

  • Seongeun