22. Regularization and Shrinkage
In today’s era of high dimensional (many variables) data, methods are needed to prevent overfitting. Traditionally, this has been done with variable selection, as described in Chapter 21, although with a large number of variables that can become computationally prohibitive. These methods can take a number of forms; we focus on regularization and shrinkage. For these we will use glmnet from the glmnet package and bayesglm from the arm package.
22.1 Elastic Net
One of the most exciting algorithms to be developed in the past five years is the Elastic Net, which is a dynamic blending of lasso and ridge regression. The lasso uses an L1 penalty to perform variable selection and dimension reduction, while the ridge ...