Beginner, here, wondering what’s idiomatic pyro for a plate for an array variable, when some elements of the array are observed, while others aren’t. Offhand, I can think of (a) using a for loop and an “if observed” test, but that will be slow, and (b) breaking the data into two blocks, for observed and unobserved, and having one plate for each. Is there some clever way to use a single vectorized plate for this?
for those that aren’t observed do you want to introduce them as latent random variables?
if so see the obs_mask arg to sample
Thanks! Having a look at that…