Hello

I am trying to compare a hierarchical model in pyMC3 to one in numpyro and have had trouble making them consistent.

Numpyro uses plates and an argument called “dim” that can be used to control conditional independence. I recently realized that the hierarchical model I made in pyMC3 which was supposed to be a random slope and intercept model is not the model I am after and that it is actually equivalent to a different numpyro model I made with non nested plate statements which I made to try out a random intercept only model. Any ideas how I can get my pymc3 model to be equivalent to my correct numpyro model with nested plate statements? Also note that the pymc3 model still has (92,2) coefficients even though it looks identical to my numpyro model where slope coefficients aren’t level dependent only (92,1). Thank you for any advice!

pymc3

correct numpyro

incorrect numpyro same as pymc3