Gamma distribution predictive sometime outputs 0

1

Hello devs. I’m using Gamma distribution for the observation distribution and sometimes (rarely) it outputs a 0 when I’m doing prior predictive check. Is this normal? Does anyone know about this or have any idea what could be going wrong?

If I were to fit my model which contains a 0 observation, the model would simply finish in 1 second and return garbage values (with r_hats very large ~1000).

2

are you using 32 or 64 bit precision? generally a good idea to use the latter when using distributions like gamma that rely on special functions or some more or less complex rejection sampling scheme or similar

3

@martinjankowiak I’m using 64 bit precision