Decreasing rhat for Black-Box Likelihood Model

Thanks for the suggestions!

I defined PDFs for each observation point. That’s why I use pm.Potential, because I defined an Op that calls the external model (simulator) and calculates a likelihood of its output values using the PDF of each observation point.

I didn’t know about pm.Simulator. Maybe I can investigate that. I’ll have to think about the distance metric I’ll use in order to utilize the PDF information on observations.