Are conjugate priors required when performing Gibbs sampling?

  • 1
    $\begingroup$ No. Only conditionally conjugate priors are required. $\endgroup$ – mef Jun 9 '16 at 1:08
  • $\begingroup$ By conditionally conjugate priors, do you mean that we can mathematically determine the unnormalized conditional posterior distribution and draw from it? $\endgroup$ – Ijies Jun 9 '16 at 1:16
  • $\begingroup$ No. Only that the prior be conjugate for the conditional distributions (of which the Gibbs sampler is composed). $\endgroup$ – mef Jun 9 '16 at 1:20

You don't require conjugate priors for Gibbs sampling.

What you need to be able to do is produce samples from the full conditional distributions.

Conjugate priors generally make that easier, but easier is not the same as necessary.

In particular situations there may be any of a variety of ways of producing samples from the full conditionals without a conjugate prior.

For example, one approach that is sometimes usable is via accept-reject; if you can bound the ratio of an approximation you can sample from to prior x likelihood, that may be doable (e.g. if you can bound the ratio of the density of some mixture of conjugate priors to the actual prior).

Of course, possible doesn't necessarily imply its always the best available solution to the issue.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.