cvms 2.0.1

• Now extracts model coefficients with a suitable environment containing the training data. This restores extraction of nnet::multinom coefficients and coefficients from mixed models whose parameters/insight methods need access to the original model data.

• Uses reformulas::findbars() for detecting random effects, following the move of findbars() from lme4 to reformulas. Adds reformulas and methods as dependencies.

• Fixes a few minor bugs with no real impact.

cvms 2.0.0

• Breaking: Fixes bug in all functions that takes fold_cols. When passing non-sequential fold indices (or sequential but not starting at 1), or string fold IDs, these were not properly matched with the fold index when extracting training and testing data for a cross-validation iteration. Now checks current iteration index against the fold ID's level index instead of its raw value. Fold column factor levels are reset to avoid levels with no observations. NOTE: Users of groupdata2::fold should not be affected by this, unless you sampled the already folded data and ended up with non-sequential (i.e., not 1:num_folds) fold IDs.

• combine_predictors() now handles when max_effect_frequency is NULL.

• Removes deprecated arguments: link, models, model_verbose. These were deprecated in v/1.0.0.

• Improves code formatting for readability.

cvms 1.8.1

• Removes unnecessary test of the pROC ROC curve contents. Thanks @xrobin.

• Fixes test for compatibility with pROC 1.19.0.1. Thanks @xrobin.

• plot_confusion_matrix(): Fixes deprecation warning when enabling add_sums=TRUE without specifying non-default sums_settings. Thanks @lucasxteixeira.

cvms 1.8.0

plot_confusion_matrix():

• Adds arrow_color argument for choosing between black and white arrow icons.

• Deprecates font_color argument in sum_tile_settings() in favor of font_counts_color and font_normalized_color.

Dynamic font settings:

• Adds dynamic_font_colors argument and the associated dynamic_font_color_settings() helper function for specifying colors below and above a given value threshold. This, for instance, allows changing the color of higher numbers when the tile background color is very dark. It also allows inverting colors of arrow icons below/above the threshold. The argument is also added to sum_tile_settings().

• Most arguments in font() (and the font_*_color arguments in sum_tile_settings()) now also accepts a function that decides the setting based on the values (counts, normalized, etc.).

cvms 1.7.0

plot_confusion_matrix():

• Adds option to set intensities of main tiles by row_percentages or col_percentages. Sum tiles cannot be set by these. Idea from Elliot.

• Fixes error when the input has been filtered (for asymmetrical plotting) and add_sums=TRUE and/or diag_percentages_only=TRUE. Thanks @jwang-lilly for reporting.

cvms 1.6.3

• Fixes extraction of nnet::multinom coefficients after change in parameters. Thanks @strengejacke for reporting the issue.

cvms 1.6.2

• Fixes tests after changes to ggnewscale. Thanks @eliocamp for the PR.

cvms 1.6.1

• plot_confusion_matrix() now shows total count when add_normalized=FALSE. Thanks @JianWang2016 for reporting the issue.

• Makes it more clear in the documentation that the Balanced Accuracy metric in multiclass classification is the macro-averaged metric, not the average recall metric that is sometimes used.

• Removes defunct tests after ggplot2 update.

cvms 1.6.0

plot_confusion_matrix():

• Breaking: Adds slight 3D tile effect to help separate tiles with the same count.
  Not tested in all many-class scenarios.

• Fixes image sizing (arrows and zero-shading) when there are different
  numbers of unique classes in targets and predictions.

• Fixes bug with class_order argument when there are different
  numbers of unique classes in targets and predictions.

cvms 1.5.0

plot_confusion_matrix():

• NEW: We created a Plot Confusion Matrix web application!
  It allows using plot_confusion_matrix() without code. Select from multiple
  design templates or make your own.

• For (palette=, sums_settings(palette=)) arguments, tile color
  palettes can now be a custom gradient. Simply supply a named list with hex
  colors for "low" and "high" (e.g. list("low"="#B1F9E8", "high"="#239895")).

• Adds intensity_by, intensity_lims, and intensity_beyond_lims
  arguments to sum_tile_settings() to allow setting them separately
  for sum tiles.

• Adds intensity_lims argument which allows setting a custom
  range for the tile color intensities. Makes it easier to
  compare plots for different prediction sets.

• Adds intensity_beyond_lims for specifying how to handle counts / percentages
  outside the specified intensity_lims. Default is to truncate the intensities.

• Fixes bug where arrow size was not taking add_sums into account.