Do I understand the source code of projpred::cv_varsel() correctly that the element “vind” in the output object of projpred::cv_varsel() contains the ranking of the variables when performing the feature selection on the reference model? So basically the element “vind” from the output object of projpred::varsel() applied to the reference model? The documentation of projpred::cv_varsel() doesn’t contain any information about that.
Yes, $vind returns the ranking of the features for both varsel() and cv_varsel().
The difference between the two functions is that the latter uses cross-validation to asses the performance of a submodel of a given size. Nonetheless, the final ranking of the features is determined using all the data.
1 Like