Posterior predictive when N is different

I estimate a model over (x, y) with N observations. Once I have my posterior samples, I want to predict y given a hypothetical x (linspace). But this time my N is actually bigger. I get some size errors. Do I need to use plate to make the length of x variable?

def model(x=None, y=None):
  obs = numpyro.sample('obs', ..., obs=y)

Silly me, it was a bug, I had mixed up my xs.

1 Like