cvms 1.3.5

• Fixes tests for CRAN.

cvms 1.3.4

• Fixes tests for CRAN.
• Adds merDeriv as suggested package.

cvms 1.3.3

• Prepares for parameters 0.15.0. Thanks to @strengejacke.

cvms 1.3.2

• Prepares package for checkmate 2.1.0.

cvms 1.3.1

• Replaces deprecated uses of ggplot2 functions. Now compatible with ggplot2 3.3.4.

cvms 1.3.0

• In order to reduce dependencies, model coefficients are now tidied with the parameters package instead of broom and broom.mixed. Thanks to @IndrajeetPatil for the contributions.

• In cross_validate() and cross_validate_fn(), fold columns can now have a varying number of folds in repeated cross-validation. Struggling to choose a number of folds? Average over multiple settings.

• In the Class Level Results in multinomial evaluations, the nested Confusion Matrix and Results tibbles are now named with their class to ease extraction and further work with these tibbles. The Results tibble further gets a Class column. This information might be redundant, but could make life easier.

• Adds vignette: Multiple-k: Picking the number of folds for cross-validation.

cvms 1.2.1

• Fixes bug in plot_confusion_matrix(), where tiles with a count > 0 but a rounded percentage of 0 did not have the percentage text. Only tiles with a count of 0 should now be without text.

cvms 1.2.0

Breaking

• Breaking change: In plot_confusion_matrix(), the targets_col and predictions_col arguments have been renamed to target_col and prediction_col to be consistent with evaluate().

• Breaking change: In evaluate_residuals(), the targets_col and predictions_col arguments have been renamed to target_col and prediction_col to be consistent with evaluate().

• Breaking change: In process_info_gaussian/binomial/multinomial(), the targets_col argument have been renamed to target_col to be consistent with evaluate().

• In binomial most_challenging(), the probabilities are now properly of the second class alphabetically.

Additions

• In plot_confusion_matrix(), adds argument class_order for manually setting the order of the classes
  in the facets.

• In plot_confusion_matrix(), tiles with a count of 0 no longer has text in the tile by default.
  This adds the rm_zero_percentages (for column/row percentage) and rm_zero_text (for counts and normalized) arguments.

• In plot_confusion_matrix(), adds optional sum tiles. Enabling this (add_sums = TRUE) adds an extra column and
  an extra row with the sums. The corner tile contains the total count. This adds the add_sums and sums_settings arguments. A sum_tile_settings() function has been added to control the appearance of these tiles. Thanks to @MaraAlexeev for the idea.

• In plot_confusion_matrix(), adds option (intensity_by) to set the color intensity of the tiles to the overall percentages (normalized).

cvms 1.1.0

• In plot_confusion_matrix(), adds option to only have row and column percentages in the diagonal tiles. Thanks to @xgirouxb for the idea.

• Adds Process information to output with the settings used. Adds transparency. It has a custom print method, making it easy to read. Underneath it is a list, why all information is available using $ or similar. In most cases, the Family information has been moved into the Process object. Thanks to @daviddalpiaz for notifying me of the need for more transparency.

• In outputs, the Family information is (in most cases) moved into the new Process object.

• In binomial evaluate() and baseline(), Accuracy is now enabled by default. It is still disabled in cross_validate*() functions to guide users away from using it as the main criterion for model selection (as it is well known to many but can be quite bad in cases with imbalanced datasets.)

• Fixes: In binomial evaluation, the probabilities are now properly of the second class alphabetically.
  When the target column was a factor where the levels were not in alphabetical order, the second level in that order was used. The levels are now sorted before extraction. Thanks to @daviddalpiaz for finding the bug.

• Fixes: In grouped multinomial evaluation, when predictions are classes and there are different sets of classes per group, only the classes in the subset are used.

• Fixes: Bug in ROC direction parameter being set wrong when positive is numeric. In regression tests, the AUC scores were not impacted.

• Fixes: 2-class multinomial evaluation returns all expected metrics.

• In multinomial evaluation, the Class Level Results are sorted by the Class.

• Imports broom.mixed to allow tidying of coefficients from lme4::lmer models.

• Exports process_info_binomial(), process_info_multinomial(), process_info_gaussian() constructors to ensure the various methods are available. They are not necessarily intended for external use.

cvms 1.0.2

• Compatibility with dplyr version 1.0.0. NOTE: this version of dplyr slows down some functions in cvms significantly, why it might be beneficial not to update before version 1.1.0, which is supposed to tackle this problem.