Sometimes, it is convenient to put priors into the likelihood function directly (for instance, if scipy distributions don't have the function type needed, or for multidimensional priors). It would be convenient to allow prior to be None and remove computation of priors from the core algorithm.