Movatterモバイル変換

We provide extremely efficient procedures for fitting the entirelasso or elastic-net regularization path for linear regression(gaussian), multi-task gaussian, logistic and multinomial regressionmodels (grouped or not), Poisson regression and the Cox model. Thealgorithm uses cyclical coordinate descent in a path-wise fashion.Details may be found in Friedman, Hastie, and Tibshirani (2010), Simon et al. (2011), Tibshirani et al. (2012), Simon, Friedman, and Hastie (2013).

Version 3.0 is a major release with several new features,including:

• Relaxed fitting to allow models in the path to be refit withoutregularization. CV will select from these, or from specified mixtures ofthe relaxed fit and the regular fit;
• Progress bar to monitor computation;
• Assessment functions for displaying performance of models on testdata. These include all the measures available viacv.glmnet, as well as confusion matrices and ROC plots forclassification models;
• print methods for CV output;
• Functions for building thex input matrix forglmnet that allow forone-hot-encoding of factorvariables, appropriate treatment of missing values, and an option tocreate a sparse matrix if appropriate.
• A function for fitting unpenalized a single version of any of theGLMs ofglmnet.

Version 4.0 is a major release that allows for any GLM family,besides the built-in families.

Version 4.1 is a major release that expands the scope for survivalmodeling, allowing for (start, stop) data, strata, and sparse X inputs.It also provides a much-requested method forsurvival:survfit.

References