Hi David,
I don’t think that’s a problem: that’s probably because sample_posterior_predictive spits out a Multitrace object, which stacks draws and chains, whereas summary, which is actually the ArviZ function, expects an InferenceData object with seperate dimensions for chains and draws.
I’m guessing that if you inject your posterior predictive samples in the idata object where you already have the trace, the warnings should go away – speaking under the control of @OriolAbril here
2 Likes