Major changes
- The threshold for worrying E-BFMI values was increased from 0.2 to 0.3 (to comply with Betancourt, 2018, arXiv:1701.02434v2).
- The term “pooled” (effects) is replaced by “population-level” (effects). Correspondingly, the term “partially pooled” (effects) is replaced by “group-level” (effects). In v1.6.0, the term “pooled” was introduced as an improvement for continuous predictors (which cannot have partially pooled or group-level effects), but this now turned out to be incorrect for those categorical predictors which do not have partially pooled effects. The “population-level”/“group-level” terminology is used in brms, so it makes sense already in terms of consistency to use it in shinybrms, too.
Minor changes
- In the table showing the default priors, column “Prior” should not have any empty fields anymore. The reason for this change is that empty fields arising from a flat prior could have been confounded too easily with empty fields arising from the vectorization (“inheritance”) of the prior within a given parameter class. This was especially problematic because the former help text underneath the table was worded mistakably.
- A recompilation of the C++ code is now avoided more often. More specifically, affected cases are those where
brms:::update.brmsfit() is used with a formula that was extended on its right-hand side. In those cases, shinybrms now avoids a recompilation by specifying argument newdata of brms:::update.brmsfit().
- The
brms::gen_extreme_value() family has been removed from the drop-down list of advanced distributional families, due to brms’s issue paul-buerkner/brms#1345. (Note that before, the brms::gen_extreme_value() family was misclassified under “Continuous outcome on the positive real line”: In fact, the support of this family depends on its shape parameter.)
Bug fixes
- Avoid a warning about a deprecated icon called
"home" when launching the app. This required increasing the required shiny version to 1.7.0.
Major changes
- Support for more distributional families for the outcome.
Minor changes
- Use a darker gray color for help texts (which constitute the majority of the in-app text). This should make them easier to read and reduce the contrast with regard to the dark red hyperlinks.
- Adapt to changes in brms’s CRAN version 2.17.0.
- Perform some basic sanity checks with respect to bounds of custom priors.
- Enhance and generalize the help text for the link functions.
Bug fixes
- Previously, advanced option “Range of random initial values […]” could have had an effect even if “Zero” was chosen under “Initial values” (namely, for the cmdstanr backend in brms versions <= 2.16.3). Now, advanced option “Range of random initial values […]” only has an effect if “Random” is chosen under “Initial values”.
- Show errors thrown by
brms::brm() to the user.
Major changes
- Allow for
offset() terms without the need to resort to the previous workaround based on a constant() prior (see brms’s GitHub issue #923 solved in brms v2.16.0). This increases the required brms version to at least 2.16.0.
- Allow the upload of a previously created
brmsfit object (page “Posterior”, tab “Run Stan”, panel “Run Stan”). This required increasing the default size limit for file uploads (global option shiny.maxRequestSize; now at 50 MB if not set by the user).
- If a previous
brmsfit exists (i.e., it is currently presented on page “Posterior”), then—if possible—a click on the “Run Stan” button causes this previous brmsfit to get updated using brms:::update.brmsfit(), saving the compilation time. Note that shinybrms automatically takes care of the fact that brms:::update.brmsfit() does not recompute the default priors if the dataset has changed. However, for uploaded brmsfits which were not fitted through shinybrms, this is not possible (and for uploaded brmsfits which were fitted through shinybrms, this is still complicated), so an uploaded brmsfit may not be updated. The updating of a previous brmsfit may be turned off using the global option shinybrms.allow_upd (see ?launch_shinybrms for details).
- Allow the cmdstanr backend for brms. The global option
brms.backend can be used to control the default backend selected in the “Advanced options” in the shinybrms app (see ?launch_shinybrms for details).
- Replace the term “nonpooled” (effects) by “pooled” (effects). This confusion was introduced by the replacement of “nonvarying” (effects) by “nonpooled” (effects) in v1.3.0 (which was obviously incorrect since “pooled” is the opposite of “varying”).
Minor changes
- UI: Minor formatting and wording improvements in help texts and other UI elements.
- UI: In case of a failed file upload, a modal dialog is shown now (instead of a notification in the lower right corner).
- Theme: Some colors have been slightly changed to match the overall theme better.
- Improved documentation of global options used by shinybrms.
- Page “Posterior” (tab “Run Stan”, panel “Run Stan”) now also shows important software versions used for the Stan run.
- The download of the MCMC diagnostics was moved from tab “Run Stan” to tab “MCMC diagnostics” (both on page “Posterior”).
- The “Default summary” from page “Posterior” may now be downloaded (as a text file).
- UI: The width of some UI elements in panel “Run Stan” has been adjusted.
Bug fixes
- Clear the input field for the name of a custom summary expression as soon as new Stan results are obtained.
- Fix a typo in the “Get started” vignette shown on the pkgdown website. (The scale parameter for the Student-t prior is 4.)
- Fix an error for Stan runs with advanced option “Open progress” set to
FALSE.
Minor changes
- Append trailing zeros in the rounding of the E-BFMI values.
- Set a minimum width for the output on tab “Default summary”.
- UI: Minor wording improvements.
Bug fixes
- Fixed a NOTE occurring in some CRAN checks.
- Avoid the deprecation warning for
brms::parnames().
- UI: Fixed some typos and incorrect spacing in the text.
Minor changes
- Renamed tab “Summary” (page “Posterior”) to “Default summary”.
- Improved help texts in the UI.
- Now using
parallel::detectCores(logical = FALSE) instead of parallel::detectCores() when determining the default number of cores for the Stan run.
- Updated some references.
Major changes
- Added the possibility to perform custom posterior summary analyses, e.g., for a transformation of parameters or for a sum of parameters.
- Added support for conditional-effects plots (see
brms::conditional_effects()).
- Use the automatic
rstan::rstan_options("javascript" = FALSE) fix on all platforms (not only Windows) if rstan version 2.21.1 or above is used.
Minor changes
- Minor improvements in the documentation for
launch_shinybrms().
- Added CITATION file.
- Changed package description.
- Included some changes necessary for brms version 2.14.0 (and above).
- Minor cosmetic improvement in the CSS theme.
- UI: Minor improvements (e.g., in label and notification texts).
- Example dataset
kidiq: Converted column mom_hs to factor.
- For partially pooled main effects, numeric (including integer) variables are not allowed anymore.
- Tab “Summary”: Replaced posterior mean and posterior standard deviation by posterior median and posterior median absolute deviation, respectively.
- Tab “Summary”: Increased the number of printed decimal digits from 2 to 4.
Bug fixes
- Fixed a possible security issue for the text input field “Prior distribution”.
Bug fixes
- Corrected a mistake in the DESCRIPTION file: The contributors and copyright holders for external components have to be listed among the authors.
Major changes
- A new theme is used for the shinybrms app. This theme is basically the “United” theme from Bootswatch v3.3.7 with just a few minor modifications. For the font, the “Open Sans” font from Google Fonts is used. See the new file “LICENSE” for details concerning the licenses of these new components.
Minor changes
- UI: Added internal links.
- On Windows with rstan version >= 2.21.1: Automatically run
rstan::rstan_options("javascript" = FALSE) at the launch of the shinybrms app to prevent occasional crashes of the R session when starting the Stan run.
- UI: Minor improvements (e.g., improved button labels and help texts).
- When specifying a custom prior and switching the class or the coefficient, the input field for the group is now only cleared (automatically) if necessary. Before, it was cleared at every change of class or coefficient (provided that the clearing was allowed, meaning that such a combination was present in the table of the default priors).
- When updating the selections for the class or the coefficient of a custom prior, the former value is now selected if it exists among the new choices.
- When deselecting a variable from the list for the nonpooled main effects or deselecting a variable from the list for the partially pooled main effects, the interactions involving this variable are now reduced to the corresponding lower-order interactions (instead of being simply removed).
Bug fixes
- Fixed the bug that when setting the input field “Group” (of a custom prior) to a nonempty value and later switching back to an empty value, the “Group” was empty in the UI, but internally, the former value was kept. This bug was solved by initializing the input field “Group” with a choose prompt (which made it necessary to catch impossible combinations of “Class”, “Coefficient”, and “Group” with a user notification).
Major changes
- Added a description how to include an offset. For this, the required version of brms had to be increased to 2.13.3.
- Removed page “Help” (which was listed under “More”). Its content is now located at other (more appropriate) places in the shinybrms app.
- Added a page “References” (listed under “More”).
- Added a page “Home”.
- Up to now, some general MCMC diagnostics (R-hat, bulk-ESS, tail-ESS) were only included implicitly (as part of the warnings introduced in version 1.2.2), besides being accessible from within the shinystan app. Now, these general MCMC diagnostics are accessible from within the shinybrms app and the warnings from version 1.2.2 are handled differently.
- Restructured page “Posterior”.
- Use the term “nonpooled effects” instead of “nonvarying effects” and the term “partially pooled effects” instead of “varying effects”.
Minor changes
- Added example dataset
roaches from package rstanarm.
- Minor improvements on pages “About” and “Links” (both listed under “More”).
- Minor improvements in the README file.
- Added the possibility to download the MCMC diagnostics (including the newly added general MCMC diagnostics).
- Minor improvements in the UI (e.g., improved and added help texts and notifications, but also added hyperlinks to the online documentation of R functions).
Bug fixes
- Fixed a bug when switching the dataset.
Minor changes
- Show more warnings from the Stan run. Some of these additional warnings overlap with the HMC diagnostics added in v1.2.0, but rather show too many warnings (even if they are partly duplicated) than too few.
- Minor improvements in the UI.
Bug fixes
- Fixed a NOTE occurring in some CRAN checks.
Minor changes
- Updated information about the shinybrms package and app (e.g., added the URL of the shinybrms website).
- Minor improvements in the UI.
Bug fixes
- Fixed a bug causing the sampling progress file not to open up if RStudio’s internal
rs_shinyviewer was used for option "browser".
Major changes
- Show Hamiltonian Monte Carlo (HMC) diagnostics in the “Output” panel on page “Posterior”.
Minor changes
- Require R >= 3.6.0.
- Added a package logo created with
hexSticker.
- Added a package website built with
pkgdown.
Bug fixes
- Fixed a NOTE occurring in some CRAN checks.
Major changes
- Added the possibility to include varying effects.
Minor changes
- Allowed *.dat files in the file upload.
- Changed the way how RStudio’s default setting for the global option
browser is avoided. Most importantly, option shinybrms.RStudio_browser was removed and replaced by the two new options shinybrms.prog_browser and shinybrms.shinystan_browser.
- Removed argument
launch.browser from shinybrms::launch_shinybrms() so that the default from shiny::runApp() is used.
- User interface (UI): Numerous minor improvements, especially concerning button labels, input fields, help texts, and notifications.
- For the “Advanced options” (for brms and Stan): Use explicit default values and show them to the user. Set
adapt_delta per default to 0.95 as done in package rstanarm. Set max_treedepth per default to 15 as done for most models in rstanarm.
- Added the possibility to download the matrix of posterior draws as a CSV file.
- Automatically create the R objects needed for the posterior predictive checks in shinystan.
- Removed some example datasets and added new ones.
- Restricted the choices for the outcome and the predictor selections (e.g., the outcome variable is now automatically removed from the list of possible predictor variables).
- It is now possible to remove interaction terms one by one.
- All predictor terms (main effects as well as interactions) are now shown in a preview table.
Bug fixes
“Under the hood”, there have been many changes which should make the shinybrms app more stable. In particular, the following bugs have been fixed:
- Fixed a bug preventing the progress file from opening up on Linux (when started interactively from the terminal).
- Fixed a bug causing the app to crash when selecting a new dataset after having constructed the model formula and chosen the distributional family for an older dataset.
- Fixed a bug making it possible to add an interaction term and then remove the main effect of an involved predictor variable while keeping the interaction term involving it.
- Fixed a bug making it possible to add an already existing interaction term with a different order of the involved predictor variables.
- Fixed a bug making it possible to clear the selection of an example dataset or the outcome, but to keep showing the default priors and the custom priors for the former selection.
- Fixed a bug making it possible to keep the outcome and the predictor variables selected when clearing the selection of an example dataset.
- Fixed a bug causing the app to crash if the dataset was changed under certain circumstances.
Bug fixes
- Example datasets:
- Added packages required for some example datasets to the “Suggests” list.
- Check for availability of packages required for some example datasets.
- Added links to the online example datasets on page “Links”.
First release. Offers support for Bayesian regression models with a Gaussian, Bernoulli, or negative binomial distribution for the (univariate) outcome. For the predictors, only nonvarying (a.k.a. population-level or “fixed”) effects are supported. Varying (a.k.a. group-level or “random”) effects are not supported yet. Neither supported are most of brms’s other features, like monotonic effects for ordinal predictors or non-linear effects. Interactions are supported, though. For the inspection of the output, only a short summary (from brms::summary.brmsfit()) and the possibility to launch shinystan is offered.