Repository navigation
Phase out functions that reside in posterior #295
Description
bayesplot is still offering metrics like rhat that now reside in the posterior package. Not do both packages implement them, the implementations differ (afaik posterior is the up-to-date one while bayesplot is lacking behind).
Bayesplot should either just wrap posterior or completely remove those functions to have a single space for them I think.
Activity
jgabry commentedon Dec 2, 2022
Thanks for opening the issue. The situation is a bit confusing because the functions unfortunately have the same name but do different things. bayesplot's rhat function isn't computing rhat itself, it's just provided for convenience to extract the rhat values. That means that for a model fit using CmdStanR the rhat values should be the same as the ones computed by posterior. For RStan models we get the old version of Rhat because RStan is lagging behind and hasn't updated to using posterior yet.
Maybe the
rhat()function in bayesplot should be deprecated in favor of a function namedextract_rhat()? That would avoid confusion about what's going on under the hood. What do you think?sims1253 commentedon Dec 2, 2022
I agree, that renaming the functions could help with clarity. It would also prevent masking when loading posterior and bayesplot together.
utkarshpawade commentedon Apr 14, 2026
hi @jgabry it seems like there are roughly three directions this could go:
rhat()/neff_ratio()in favor ofextract_rhat()/extract_neff_ratio()(as suggested above) to avoid masking with posterior.posterior::rhat()/posterior::ess_*()so bayesplot always reflects the up-to-date implementations.jgabry commentedon Apr 14, 2026
I think option 1 is actually the only option because the versions in posterior are doing something different, they're not replacements.