ABSTRACT
An attractive feature of `1-regularized procedures is their ability to combine variable selection with parameter fitting. We often select a model based on cross-validation-as an estimate for prediction or generalization error-and then do further validation on a held-out test set.
