| grpreg-package {grpreg} | R Documentation |
Regularization paths for regression models with grouped covariates
Description
This package fits regularization paths for linear, logistic, and Cox regression models with grouped penalties, such as the group lasso, group MCP, group SCAD, group exponential lasso, and group bridge. The algorithms are based on the idea of either locally approximated coordinate descent or group descent, depending on the penalty. All of the algorithms (with the exception of group bridge) are stable and fast.
Details
Given a design matrix X in which the features consist of
non-overlapping groups and vector of responses y, grpreg
solves the regularization path for a variety of penalties. The
package also provides methods for plotting and cross-validation.
See the "Getting started" vignette for a brief overview of how the package works.
The following penalties are available:
-
grLasso: Group lasso (Yuan and Lin, 2006) -
grMCP: Group MCP; like the group lasso, but with an MCP penalty on the norm of each group -
grSCAD: Group SCAD; like the group lasso, but with a SCAD penalty on the norm of each group -
cMCP: A hierarchical penalty which places an outer MCP penalty on a sum of inner MCP penalties for each group (Breheny & Huang, 2009) -
gel: Group exponential lasso (Breheny, 2015) -
gBridge: A penalty which places a bridge penalty on the L1-norm of each group (Huang et al., 2009)
The cMCP, gel, and gBridge penalties carry out
bi-level selection, meaning that they carry out variable selection at
the group level and at the level of individual covariates (i.e., they
select important groups as well as important members of those groups).
The grLasso, grMCP, and grSCAD penalties carry
out group selection, meaning that within a group, coefficients will
either all be zero or all nonzero. A variety of supporting methods
for selecting lambda and plotting the paths are provided also.
Author(s)
Patrick Breheny
References
Yuan M and Lin Y. (2006) Model selection and estimation in regression with grouped variables. Journal of the Royal Statistical Society Series B, 68: 49-67. doi: 10.1111/j.1467-9868.2005.00532.x
Huang J, Ma S, Xie H, and Zhang C. (2009) A group bridge approach for variable selection. Biometrika, 96: 339-355. doi: 10.1093/biomet/asp020
Breheny P and Huang J. (2009) Penalized methods for bi-level variable selection. Statistics and its interface, 2: 369-380. doi: 10.4310/sii.2009.v2.n3.a10
Huang J, Breheny P, and Ma S. (2012). A selective review of group selection in high dimensional models. Statistical Science, 27: 481-499. doi: 10.1214/12-sts392
Breheny P and Huang J. (2015) Group descent algorithms for nonconvex penalized linear and logistic regression models with grouped predictors. Statistics and Computing, 25: 173-187. doi: 10.1007/s11222-013-9424-2
Breheny P. (2015) The group exponential lasso for bi-level variable selection. Biometrics, 71: 731-740. doi: 10.1111/biom.12300
Examples
## Not run:
vignette("getting-started", "grpreg")
## End(Not run)