Thanks for taking your time to reply.
I dont’ think that the problem is what I am trying to model here. It’s a toy example where I try to recover the parameters of some Beta distribution (0.5 and 0.5) that generates the p`s for a geometric distribution. There might be no practical reason for this, but what happens is that the sape of Beta(.5,.5) places high probability on values near zero and values near one
And given that I sample 2000 points from there, you can assume some will have very low values.
The geometric distribution will then spit out very big numbers for very low values of P and that’s what I believe to be the issue here across different OSes (but I don’t know how/why this happens). To confirm this, with higher parameters for the original Beta, the Geometric will output much smaller numbers and I am able to fit the model on both OSes.