Skip to content

Modernize MCMC convergence diagnostics #528

Description

@BJMCox

BrooksGelmanConvergence uses the older variance-based Brooks-Gelman statistic and is the default for TransformedMCMC.

Modern practice recommends rank-normalized split R-hat, folded R-hat, and bulk/tail effective sample sizes. These better detect non-stationarity and tail problems.

Consider adding these diagnostics and then reassessing the default.

Reference: Vehtari et al. (2021), https://doi.org/10.1214/20-BA1221

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions