2
$\begingroup$

I have a model which includes a Beta distribution and I am looking for guidance on how to parameterize a hyper-prior for it. For example, this post uses a Beta parameterized with a mean and concentration and takes the concentration kappa ~ Pareto(1, 1.5).

What are some recommended distributions that I should consider for my kappa? And what principles might I use to select the right one for my setting?

$\endgroup$
1

1 Answer 1

0
$\begingroup$

I ended up using a uniform distribution with a lower bound selected to be not-too-low through trial and error:

kappa_s = numpyro.sample("kappa_s", dist.Uniform(5, 500))  # Model converges to unacceptable values when kappa_s is smaller (e.g. 1)
$\endgroup$
1
  • $\begingroup$ I won't accept this, but I though it might inspire someone to offer a more acceptable answer. $\endgroup$ Dec 22, 2021 at 0:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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