That notebook is where I started. Austin has a real valued prediction (normal mixture of real values). I have a discrete integer valued prediction - that’s why I took a Mixture of poissons and then the mode().
The challenges I ran into along the way were:
- discrete mixtures (for x1)
- bounded discrete mixtures (for x1)
- dependent bounded discrete mixtures (for x2)
- non-convergence (with Metropolis)
- ppc errors
I am waiting for a test run with 50 samples