New cv.function() method meant to replace cvSelect(), direct use of which is now discouraged.
New selectModelList() to be used with cv.function() (or with cvSelect()). selectModelList() implements recursive cross-validation, where the fit of a model selected by CV is assessed by CV. The same procedure is also available by setting recursive=TRUE in a call to cv.modList().
cv.default() and other cv() methods acquire a details argument, which if TRUE includes information about the folds in the returned object.
New as.data.frame.cv() and related methods for turning the detailed results returned by cv() methods into a data frame, with new print() and summary() methods for the objects produced.
Improvements to code, introducing folds(), fold(), and related functions.
Refactoring of code; cv() methods now all call cvCompute() (which is new), cvMixed(), or cvSelect().
Reorganization of package file structure and of documentation.
Make the cv.default() method more robust, particularly for parallel computations.
Reorganize package vignettes (of which there are now 5).
Other small improvements.
cv() et al. now work properly with “non-casewise average” CV criteria such as the new rmse() and medAbsErr(), not just with “casewise-average” fit criteria such as mse() and BayesRule().
Bias adjustment and confidence intervals (which are new) are computed only for casewise-average CV criteria. Demonstrate that 1 - AUC isn’t a casewise-average criterion.
Generally suppress spurious messages about setting the seed in cv.modList() for LOO CV.
Fix bugs in selectTrans() that caused errors when one of response and predictors arguments not specified.
Fix bug in cvMixed() that prevented parallel computations (reported by Craig See).
Fix small bug in cvSelect(), returning properly named “coefficients” element when save.coef is TRUE.
Fix bug in cv.lm() and cv.glm() with method=“hatvalues” for cost criteria other than mse().
Add selectTransStepAIC() procedure for use with cvSelect().
Add medAbsErr() and rmse() cost criteria.
Add coef.cvSelect() method.
Add cv.rlm() method.
plot.cvModList() can show averages +/- SDs, and averages and CIs, as well as averages and ranges.
Add Pigs data set.
change getResponse() and methods to GetResponse() to avoid name clash with nlme.
Improvements and updates to documentation, and expanded cv.Rmd vignette.
Mixed-models methods no longer flagged as “experimental.”
Mixed-models CV functions no longer limited to nested random effects.