When I use keep_size=True on pm.fast_sample_size_predictive(), I still
get a posterior dataset with only a single chain rather than two. And now the dataset has 2 draws (instead of 600 it should have, or 1200 it has with keep_size=False), and an extra (unwanted) dimension (of 600) on all my result variables.
When I try pm.summary() on the posterior predictive, it gives me thousands of error messages, instead of the 19 message before:
and so on, for longer than I wanted to wait.
Have I discovered a bug in the keep_size logic of pm.fast_sample_size_predictive()? Or perhaps I am making a different mistake now?