automultinomial (version 1.0.0)

multinomialBCD: Block coordinate descent

Description

Computes, with group lasso regularization penalty, a lasso path for a model with user specified loglikelihood gradient and hessian.

Usage

multinomialBCD(X, z, groups, penaltyFactor, nLambda, H = NULL, theta = NULL)

Arguments

X

design matrix

z

response matrix

groups

a grouping of coefficients

penaltyFactor

a vector specifying the relative penalty level for each coefficient group

nLambda

number of penalty parameters on the lasso path

H

optional Hessian at loglikelihood maximum, used for the quadratic approximation in the centered model

theta

optional coefficients at loglikelihood maximum, used for the quadratic approximation in the centered model

Value

list containing the lambdas on the coordinate descent path and the coefficients for each lambda

Examples

# NOT RUN {
#
# }