Reproducing diagnostics after fit?

Interfaces
1

Given a stanfit object returned by rstan::samplng() is there a straightforward way to reproduce the set of diagnostics that were printed when it was fit (divergent transitions, tree depth, bulk ESS, tail ESS, etc.)?

1 Like
2

Hey @potash
I think this might help you: http://mc-stan.org/bayesplot/articles/visual-mcmc-diagnostics.html

2 Likes
3

In addition to @scholz’s suggestion, check out these functions

and these

We’re also starting to move this stuff to our new posterior package, which might interest you:

4

Thanks, those are the functions I was looking for. What threshold does stan currently use for the bulk and tail ESS warnings?

5

I think it’s currently 100 * number_of_chains

but that’s just a heuristic and is the sort of thing that will probably change based on additional experiments/research. Eventually we want to move all of that out of rstan (and other packages) and into the posterior package and allow for the user to set different thresholds for warnings:

1 Like