Interactive Visual and Numerical Diagnostics and Posterior
Analysis for Bayesian Models
A graphical user interface for interactive Markov chain Monte
Carlo (MCMC) diagnostics and plots and tables helpful for analyzing a
posterior sample. The interface is powered by RStudio's Shiny web
application framework and works with the output of MCMC programs written
in any programming language (and has extended functionality for Stan models
fit using the rstan and rstanarm packages).
- Better compatibility with latest releases of 'shinyjs' and 'DT' packages
shinystan::launch_shinystan_demo() now works without first having to load
the package with a call to
- Unless running on a server, the app now stops running when browser tab is
- Fix bug in
deploy_shinystan preventing some ShinyStan apps from being
- Add optional argument
pars to the
as.shinystan method for stanfit objects,
allowing a subset of parameters to be selected for inclusion in the resulting
drop_parameters function for removing parameters from a shinystan
object (useful for very large objects when you only want to look at a subset of
- Add rstanarm to Suggests (in the DESCRIPTION file) so
can be called.
- Allow selecting
yrep from global environment for PPcheck
as.shinystan to S4 generic with methods
- Add rsconnect to
- Rename some of the slots in shinystan objects. The
update_sso function can
be used to ensure that old shinystan objects have an internal structure
compatible with this release.
- Improve line coverage of api tests
- Compatibility with recent ggplot2 update
- Select parameters via regular expressions in parameters plot ('Estimate' page)
- Fix error when selecting an entire parameter group in parameters
plot ('Estimate' page)
- Fixes small bugs and typos
- Adds support for stanreg objects
- Moves ggplot2 from Depends to Imports
Version 2.0.0 has a new look, a new(ish) name, and a lot of new functionality.
Many bugs have also been fixed (see GitHub issue tracker).
- The name of the R package is shinystan and the app/GUI is ShinyStan.
deploy_shinystan function lets you easily deploy ShinyStan apps
to RStudio's shinyapps.io for any of your models. Each of your apps
(i.e. each of your models) will have a unique url.
- Rebranding (new look to reflect changes to Stan logo and website)
- HMC/NUTS diagnostic plots ('Diagnose' page, 'HMC/NUTS (plots)' tab)
- Specify transformations (e.g. log, logit, sqrt, etc.) for density,
histogram, bivariate, trivariate plots (on 'Explore' page) and HMC/NUTS diagnostics
- Many plots can now also be saved as pdf
- Bivariate scatterplot plot also shows divergent transitions and max treedepth
saturation (on 'Explore' page)
- More detailed glossary entries
- (Experimental) Introduce basic graphical posterior predictive
checking ('Diagnose' page, 'PPcheck' tab) for limited class of models
- Option to show partial autocorrelations ('Diagnose' page, 'Autocorrelation' tab)
- Better customization of of posterior summary statistics table
- Many improvements to GUI design