I'm interested in calculating the prior variance of a two dimensional latent model I've built in GPflow using the folllowing notebook 'Heteroskedastic Likelihood and Multi-Latent GP'.
Currently I'm doing this in a pretty hacky way by asking the model to predict far away from the training data. Looking at the source code it's not obvious I can do this without to having to calculate the quadratures. Is there a simpler way of getting the priors that I've overlooked?
Edit: I'm also interested in generating samples!