Sampling with full conditionals


I am a new user so sorry if my question is stupid.
I have a model which can be written in two ways:
in one (A) I know the full conditionals in closed-form whereas in the other one (B) I don’t.
Is there a way to write the model (A) in Stan, and save all the generated samples ?

I know that could simply write my own generator for (A) but the point of having both in Stan to make all sorts of comparisons.


None of the algorithms in Stan utilize full-conditional distributions. But if you can write out all the full-conditional distributions, then you should be able to write the kernel of the (log) density function, which is what is needed for Stan.

But then if I write my density function, Stan’s sampler will not have the full acceptance rate of my Gibbs sampler based on full conditionals right?

Stan’s sampler doesn’t work in a way a Gibbs sampler works. For more details you may want to look at Michael Betancourt’s introduction to hamiltionian monta-carlo.