Gamma(.., ..)^2 -> generalized_gamma(?, ?, ?)

[Bob_Carpenter]

I don’t know who was resisting. If it’s a mainstream distribution with a Wikipedia page and lots of publications, there’s no harm in including it even if it’s just a matter of adding error testing to the automatically generated Stan code. The bottleneck now is testing.

[sakrejda]

Ok, that makes sense.

[simkovic]

@sakrejda

The hand-coded generalized gamma in Stan …

Two versions were discussed in this thread. Which one do you recommend? S1 or S2 (Lawless)?

[sakrejda]

Pretty sure 1S worked better. If you’re willing to read some code it’s all in here :

Gamma(.., ..)^2 -> generalized_gamma(?, ?, ?)

I don’t recall writing one, I would just use a slice sampler, just make sure you work with the log density… sorry, I had a simple R implememtation kicking around but I can’t find it right now. Really a more sensible thing is to work with the RNG from R’s flexsurv package. If you want to see how the parameterizations are related check out this repo: https://github.com/sakrejda/survival-densities Specifically, here are tests relating flexsurv’s parameterizations to mine, for 1S: https://github.…

Those Stan implementations are tested against flexsurv (in the linked repo) although I haven’t made it any easier to build…