search

pymc-devs / pymc-examples

Examples of PyMC models, including a library of Jupyter notebooks.
https://www.pymc.io/projects/examples/en/latest/
MIT License
282 stars 244 forks source link

Problems in (high impact) GLM Hierarchical notebook #421

Open cluhmann opened 2 years ago

cluhmann commented 2 years ago

Notebook title:GLM: Hierarchical Linear Regression Notebook url: source link, website link

Issue description

There are many, many divergences in the sample of the main, hierarchical model. There also appear to be images missing under "The Models" section. Not sure what these are supposed to be.

Proposed solution

I suspect the sampling could be fixed by re-running and/or setting the random seed for stability. Not sure what the images were (at one time).

OriolAbril commented 2 years ago

Maybe we can remove/deprecate it in favour of https://www.pymc.io/projects/examples/en/latest/case_studies/multilevel_modeling.html? Or probably better combine/restructure the two to link between them to make them complementary and avoid duplication.

cc @fonnesbeck who I know was thinking about improving the multilevel modeling primer one.

fonnesbeck commented 2 years ago

Yes, there are a few things about the current version of the multilevel modeling primer I don't like (notably the expression of floor and basement as being independent). There is significant overlap in content with this notebook, so I'd be in favor of somehow merging them. If I'm not mistaken, this notebook originally used the now-extinct glm submodule, which made it stand out a little more. As far as the sampling issues go, I'd bet they are related to the centered random effect parameterization.

fonnesbeck commented 1 year ago

UPDATE: Will be aiming to complete this in the next few days.