I am new to Pyro. Hope some one can help me start.
I want to learn Pyro with the Mnist classification problem. I assume I could build a Baysian Net with pyro.random_module as the tutorial Bayesian Regression.
My question is how could I register a complex deep Neural net with lots of latent variable and their respective prior in the pyro.random_module.
Can any one give me an example ?
im not sure exactly what youre trying to do. it would help if you posted (even a non-running) example to illustrate your idea. in general, all the parameters in a pytorch nn module are uniquely named. you can put priors on these parameters by referencing their name (see example in the tutorial). for the prior(s), random_module can either take a pyro distribution as an argument or a dict keyed on parameter names with distributions as values.