Bob Carpenter wrote and posted this article a couple years ago. I think it’s still worth reading, so I’m posting it again here.
Bob’s article covers much of the material in our Bayesian Workflow article and our Bayesian Workflow book in progress, but in just 6 pages, so it could be a good thing to read to get started on the topic.
Here’s how it goes:
What do we need from a probabilistic programming language to support Bayesian workflow?
BOB CARPENTER, Flatiron Institute, New York City
1 INTRODUCTION
2 THE NITTY GRITTY OF BAYESIAN WORKFLOW
3 INFERENCE REQUIRED FOR BAYESIAN WORKFLOW
3.1 Prior predictive checks
3.2 Simulation-based calibration
3.3 Posterior predictive checks
3.4 Cross-validation
3.5 Sensitivity analysis
3.6 Model comparison by calibration and sharpness
3.7 Summary of inference and modeling required for workflow
4 REMEDIATING PROBLEMS DURING WORKFLOW
5 WORKFLOW SUPPORT IN EXISTING PPLS
BUGS
Stan
PyMC3
Pyro
Edward2
Turing.jl
Oryx6 CONCLUSION