|
Abstract:
|
Regularized regression and classification methods fit a linear model to data, based on some loss criterion, subject to a constraint on the coefficient values. Different forms of the constraint produce different families of solutions whose members are indexed by the constraining value. For example, ridge-regression, subset selection and the lasso use different constraint forms with squared-error loss. For the lasso, a simple boosting strategy has been developed that rapidly computes close approximations to its complete family of solutions. A simple extension to this strategy is presented that can be similarly used with a wide variety of loss criteria and/or constraint forms, including many that are not convex
|
