Apologies if this is a dumb question: but when we define a logp function for a DensityDist, for a continuous RV, does it have to be normalized, or can we use an un-normalized representation?
Specifically, I’m interested in using Miller’s prior for a gamma RV, and its potential is easy to compute, but the normalizing constant is expensive to compute. So if the sampler could get away with just using the un-normalized potential, that would save both me and the sampler a ton of work!
Thank you