cvplogistic: Penalized Logistic Regression Model using Majorization
Minimization by Coordinate Descent (MMCD) Algorithm
The package uses majorization minimization by coordinate
descent (MMCD) algorithm to compute the solution surface for
concave penalized logistic regression model. The SCAD and MCP
(default) are two concave penalties considered in this
implementation. For the MCP penalty, the package also provides
the local linear approximation by coordinate descant (LLA-CD)
and adaptive rescaling algorithms for computing the solutions.
The package also provides a Lasso-concave hybrid penalty for
fast variable selection. The hybrid penalty applies the concave
penalty only to the variables selected by the Lasso. For all
the implemented methods, the solution surface is computed along
kappa, which is a more smooth fit for the logistic model.
Tuning parameter selection method by k-fold cross-validated
area under ROC curve (CV-AUC) is implemented as well.