My use case is just gaining practice playing with Pymc. I’ve previously in the course of this project modeled my outcomes as both Poisson and NegativeBinomials even modeling counts of the different types of scoring events. This just presented an opportunity to play with an outcome distribution I didn’t get to work with very much and a unique approach I wanted to try.
In terms of the linear approach to the cutpoints, I really was just doing what I could to get the model to sample to see if I could gain some understanding about how the model was working. I wanted to circle back to using a different distribution, I just couldn’t get either of them to sample or the debug function to shine light on what was going wrong.