Skip to content

Issue sampling from model using Beta distribution with small alpha values #2159

Open
@yahrMason

Description

@yahrMason

Hello, I have a model that I am trying to sample with a BetaBinomial likelihood. I describe the model in detail with the code in this discourse post. The error I am receiving is as follows:

BetaBinomial: the condition α > zero(α) is not satisfied.

I have inverse logit constraints in my model, so it is not possible for the alpha value to reach zero (though can be arbitrarily small). From my investigations it looks like this is an issue related to this issue with Distributions.jl. The developers of Gen.jl have identified this issue as well outlined here and here. It looks like their solution was rather straightforward. I am not sure if it is worth exploring something similar for Turing.jl (if it doesn't exist already).

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions