New stan_glmer_inits() computes data-based starting values for the Bayesian models used by perf_mod(). The results can be passed to rstanarm::stan_glmer() (directly or via perf_mod(..., init = )) and can shorten warmup for models that are slow to converge, such as those with heterogeneous variances.
perf_mod() gains an initialize argument (default FALSE) that computes and applies these starting values automatically, reusing the sampler seed (when one is passed) so that the entire fit is reproducible.
Fixed a bug where perf_mod() duplicated the resampling statistics for workflow sets: when more than one metric had been collected, each statistic was included in the Bayesian model once per metric.
New compare_to_leader() contrasts every model in a perf_mod() object against the best model in the set and returns a data frame of probability statements about how much each candidate gives up relative to that leader. autoplot() on the result draws the posterior metric intervals alongside those probabilities.
New initialize_keys() builds a template of model labels to edit and pass to the key argument of compare_to_leader().
perf_mod() gains a select_best argument for workflow sets. Workflows are always reduced to their own best tuning parameter candidate; select_best = TRUE additionally collapses workflows that fit the same type of model with the same engine down to the best of each group, so that each model type/engine combination competes once. The default, FALSE, keeps the existing behavior of one competitor per wflow_id.
Maintenance release for an upcoming ggplot2 release (#74).
Transition from the magrittr pipe to the base R pipe.
Maintenance release for a broken test in current R-devel.
Maintainer email change.
Transition from gather() to pivot_longer().
Update to testthat 3e.
The ggplot() methods are now removed in factor of autoplot() methods.
perf_mod() methods added for tuning parameter objects from tune, finetune, and workflowsets.
rstanarm version 2.21.1 or higher is now required due to changes to their APIs.
Re-licensed package from GPL-2 to MIT. All copyright holders are RStudio employees and give consent.
contrast_models() now returns a tibble and has an extra column called contrast.
The plot methods are now deprecated and will be removed in the next version. They are not very good and can be replaced with simple ggplot code.
An optional formula argument was added to perf_mod.rset() and perf_mod.data.frame(). When the resampling method has multiple ID columns, a nested data structure is assumed (with a warning). The new formula argument can be used to over-ride the nesting.
dplyr 1.0.0. For the newer dplyr version, if critical columns for posterior or posterior_diff objects are removed, the objects is down-graded to a tibble. For earlier versions of dplyr, the object is not down-cast.A small, maintenance release.
Moved from the broom package to the generics package to get the tidy generic.
ggplot2 was moved to Suggests
The sole tidy method was more explicitly exported so that the generics man files show the method.
The large RData objects containing the examples have been removed from the package and are accessible via a link to the GitHub repo.