Divergences in examples/generalized_linear_models/GLM-ordinal-regression.ipynb

Regression Models with Ordered Categorical Outcomes:
https://www.pymc.io/projects/examples/en/latest/generalized_linear_models/GLM-ordinal-regression.html:

Issue description

After make_model() is defined, five different models are created and sampled from. Unfortunately, models 1-2 have hundreds of divergences. Models 3-4 sample fine. Model 5 is identical to model 4, presumably the argument logit=False is missing. If I add this argument, 2 of 4 chains are completely stuck when sampling.

These divergences were not automatically reported because the numpyro sampler was used. However, since the most recent release we get warnings for divergences also for this sampler. 👍

Proposed solution

Fix the models so that they can be sampled from. Or if that turns out to be difficult, consider whether they can be removed. Or add a discussion of the problem with these models.

Open examples/generalized_linear_models/GLM-ordinal-regression.ipynb and inspect make_model() and models 1–5. Re-run the notebook with the numpyro sampler, compare divergences and chain behavior, and determine whether the models can sample reliably. Done means corrected models with no reported divergences, or a documented rationale for removing or retaining problematic models.