Tweedie distribution / Tweedie GLM

Domains Statistics
, ,
#1

I’m interested in fitting a GLM via GLM.jl with a Tweedie distribution (https://en.wikipedia.org/wiki/Tweedie_distribution), particularly when the domain of the index parameter is not a special case that’s already available. Something like glm(...,Tweedie(1.25), LogLink()) for instance.

It looks like GLM.jl uses Distributions.jl for the families, and I don’t see a clear way in the latter to get to a Tweedie distribution.

Before I get into the weeds trying to implement this myself, am I missing any obvious solution?

1 Like
#2

I started TweedieDistributions.jl, but I only needed to simulate from the compound Poisson-Gamma case (e.g. I haven’t implemented any density functions) and I haven’t integrated it into the Distributions.jl type hierarchy. I’m not sure what would need to be done for it to be compatible with GLM.jl.

A search on JuliaHub Repositories doesn’t turn up much, but might be another place to start.