Bayesian model diagnostics: Workflows and software tools

Authors

Affiliations

Noa Kallioinen

University of Helsinki

Teemu Säilynoja

PyMC Labs

Osvaldo A. Martin

Aalto University

Material

The material for this tutorial consists of the slides, exercises, and this book.

Exercises

The exercises are available in both Python and R:

• Download and expand this zip file
• This contains the exercises in different formats
• For RStudio, first open the file stancon2026-workflow-tools.Rproj, then open exercises-r.qmd
• For Python, use the exercises-python.ipynb file
• The first chunk will install the required packages. For R: bayesplot, posterior, and priorsense; for Python: arviz

The exercises can also be viewed online: R or Python

An online environment is available here: binder

Solutions are available here: R, Python

Overview

When working with Bayesian models, a range of related tasks must be addressed beyond inference itself. These include:

• Diagnosing the quality of the inferencial method (MCMC, VI, …). As we usually use numerical methods to perform inference.
• Model checking and criticism, which includes posterior predictive checks, prior predictive checks, prior/likelihood sensitivity analysis.
• Model comparison

To simplify the workflow of performing these tasks, we can use some tools. The ones we are going to cover in this tutorial are:

• R
  • posterior: Conversion, manipulation, and summarization of draws from posterior and prior distributions.
  • bayesplot: Visual checks and summaries.
  • priorsense: Prior diagnostics and sensitivity analysis.
• Python
  • arviz: a meta-package that imports all the packages bellow under a unified namespace.
    • arviz-base: Data related functionality, including converters from different PPLs.
    • arviz-stats: Statistical functions and diagnostics.
    • arviz-plots: Visual checks and summaries built on top of arviz-stats and arviz-base.