Contributor
GStechschulte

Better tools to interpret complex Bambi regression models


Mentors
Osvaldo, Tomas Capretto
Organization
NumFOCUS
Technologies
python
Topics
visualization, machine learning, Probabilistic Programming, Bayesian Statistics, Interpretable machine learning
Bayesian modeling has increased significantly in academia and industry over the past years thanks to the development of high quality and user friendly open source probabilistic programming languages (PPL) in Python and R. Of these is Bambi, a Python library built on top of the PyMC PPL, that makes it easy to specify complex generalized linear multilevel models using a formula notation similar to those found in R. However, as the model building portion of the Bayesian workflow becomes easier, the interpretation of these models has not. Currently, to aid in model interpretability, Bambi only supports conditional adjusted predictions plots. The objective is to take inspiration from the existing Bambi tooling and R package marginaleffects, to extend upon existing plotting functionality and to provide additional plotting functions such as conditional comparisons and conditional marginal effects to allow Bambi modelers to extract insights and interpret their models in a more automatic and effective manner.