cvms 1.0.0 released

After 6 months of work, cvms version 1.0.0 has finally been released!

This version is a major refactoring of the package and includes tons of new features and changes.

The most important additions are:

  • Hyperparameter tuning of custom model functions
  • Within-cv preprocessing
  • Multiple new metrics
  • Identification of observations that are difficult to predict
  • Four new vignettes (tutorials)

Importantly, cvms no longer depends on caret, as the creation of confusion matrices and calculation of related metrics are now implemented in cvms. This should make installation easier.

cross_validate_fn() has been improved and should allow cross-validation of most model functions.

Breaking changes:

There’s a list of breaking changes here .

  • A big one is that the models argument in cross_validate() and validate() has been renamed to formulas to be consistent with cross_validate_fn() and validate_fn().
  • Another big one is that the family/type argument no longer has a default value.

New functions:

  • validate_fn()
  • confusion_matrix()
  • evaluate_residuals()
  • summarize_metrics()
  • most_challenging()
  • select_definitions()
  • model_functions()
  • predict_functions()
  • preprocess_functions()
  • update_hyperparameters()
  • simplify_formula()
  • gaussian_metrics(), binomial_metrics(), multinomial_metrics()
  • baseline_gaussian(), baseline_binomial(), baseline_multinomial()
  • plot_confusion_matrix(), plot_metric_density(), font()

Twitter thread:

One of my new favorite functions is plot_confusion_matrix():

select_definitions() makes it faster to extract relevant columns from the output when comparing the models: