Specifically I’m looking for minimal working examples of models that rely (internally, anywhere) on either the gamma_cdf or the gamma_p function and don’t work all that well. I tried a really simple one and I can’t demonstrate an effect of the fix I did, and if I run it on my wonky model the fix makes a HUGE difference. I’d love to have something in between. The symptoms could be divergences or just low acceptance rate and poor mixing.
I seem to remember Ben mentioning some problems with rstanarm Gamma models but can’t find specifics.