Mastering Machine Learning with R - Second Edition

Cross-validation with glmnet

We have used LOOCV with the caret package; now we will try k-fold cross-validation. The glmnet package defaults to ten folds when estimating lambda in cv.glmnet(). In k-fold CV, the data is partitioned into an equal number of subsets (folds) and a separate model is built on each k-1 set and then tested on the corresponding holdout set with the results combined (averaged) to determine the final parameters.

In this method, each fold is used as a test set only once. The glmnet package makes it very easy to try this and will provide you with an output of the lambda values and the corresponding MSE. It defaults to alpha = 1, so if you want to try ridge regression or an elastic net mix, you will need to specify it. ...