In built method to sample params with a constraint?

[5183nischal]

Is there a way that I can sample parameters with prior.sample with a constraint? For example, if I have two parameters and I want to systematically disregard pairs of parameter when the ratio is less than 1, It can be done post-hoc once the sampling is complete but want to see if there is a simple way to do it with inbuilt approaches?

Thanks!

[michaeldeistler]

You can probably achieve that with the RestrictedPrior.

from sbi.utils import RestrictedPrior

prior = BoxUniform(-torch.ones((3,)), torch.ones((3,)))
def accept_reject_fn(params):
   return True is torch.sum(params) > 1.5 else False
restricted_prior = RestrictedPrior(prior, accept_reject_fn)

Let me know if this works!
Michael

[jorobledo]

Looking at this answer, I notice that RestrictedPrior escaped my attention. Would we need to add a to(device) to this type of prior too @michaeldeistler ?