Reading this post, and reviewing the documentation of NumPyro/Pyro, I think I'm not following the crucial difference between NumPyro/Pyro. I understand that Pyro uses PyTorch as backend, and NumPyro uses JAX as backend, but other than that I'm not sure about the critical differences. If their frontend is about the same (which seems to be the case here) why is JAX mentioned in this post? Could we simply not replace Pyro with Stan for statistical modelling (whether with PyTorch or JAX backend)?
replies(1):