Following on from his talk at StanCon, Michael Betancourt just wrote three Stan case studies, all of which are must reads:
- Diagnosing Biased Inference with Divergences: This case study discusses the subtleties of accurate Markov chain Monte Carlo estimation and how divergences can be used to identify biased estimation in practice.
- Identifying Bayesian Mixture Models: This case study discusses the common pathologies of Bayesian mixture models as well as some strategies for identifying and overcoming them.
- How the Shape of a Weakly Informative Prior Affects Inferences: This case study reviews the basics of weakly-informative priors and how the choice of a specific shape of such a prior affects the resulting posterior distribution.
Reproducible R scripts
They all come with fully reproducible knitr scripts to run in RStan. The same lessons hold for the other interfaces, so don’t let the R put you off.
A spectator sport
It was particularly fun sitting in the office the day Michael went and figured out all the mixture modeling properties. It followed on from one of our Stan meetings and some of my own failed experiments.
Publish or perish
It’s really a shame that this kind of methodological study is so hard to publish, because all three of these deserve to be widely cited. Maybe Andrew has some ideas of how to turn these into “regular” papers. The main thing journal articles give us is a way to argue that we got research results from our research grants. Not a small thing!
Other case studies
We have a bunch of case studies up and are always looking for more. The full list and instructions for submitting are on the Stan case studies page.
Getting information out of the posterior fit object in R
And in case you didn’t see it, Jonah wrote up a guide for how to extract the kind of information you need for extracting information from a Stan fit object in R.