What is the purpose of the "size" parameter in sample_posterior_predictive

What is the purpose of the “size” parameter in sample_posterior_predictive?
I’m also wondering why it’s not an available parameter when using weighted sample a la sample_posterior_predictive_w.

I’m asking this because I’ve noticed that when I don’t set the size parameter, my predictions on held-out-data don’t change much in response to changes to the data.

It suppose to return multiple random draws conditioned on one set of posterior parameters (ie. one slice of the trace). To me it never really makes much sense and it is pretty buggy, so I dont recommend using it.

1 Like