Yeah, why? I had the feeling that it would be easier to diagnose and extract if all was done within one stan function call. Otherwise I would have to collect all the samples together from the K fits. Obviously this is easily arranged in R, but I thought I would lose the “compact-ness”.
Yes, that is what I am doing. My parameters have an additional dimension which represents to which group they belong. The parameters for group k are then only used once for training (with all data (here y and X) except for the data belonging to group k) and later for creating replicated data (with all the data belonging to group k).
Side note: Would I/you/the Bayesian community call this “replicated” data or “simulated” data (as I do not use the same data points for training and testing), is there a consensus?