Hi Mina,
I don’t know your use case and data, and I’m not very experienced with ordered logistic regression, but here are some pointers:
- Have you tried with different priors to see if this affects your prior predictive checks?
- Sampling and ADVI issue: have you standardized your predictors? Predictors that are on a very large scale often mess up MCMC sampling.
- You could find this notebook useful – it’s an adaptation to PyMC3 of McElreath’s Rethinking chapter 11 (first edition).
Sorry I can’t give you a more detailed answer, but I hope this already helps you 