automultinomial (version 1.0.0)

MPLEinternal: Multinomial regression

Description

Fits an automultinomial model (by pseudolikelihood). Fits the uncentered auto-model by default, #when an adjacency matrix is given. Performs an optional variable selection step, treating each #variable as a group for all response categories, with an option to group related variables.

Usage

MPLEinternal(X, z, A = NULL, groups = 2:dim(X)[2], nLambda = 101,
  centered = TRUE, BIC = TRUE, select = TRUE, standardize = TRUE,
  constraint, polish, innerIndices)

Arguments

X

nxp design matrix, where the first column must contain an intercept

z

nxk response indicator matrix, or a length n vector with the response type of each observation

A

an adjacency matrix or list of adjacency matrices defining neighborhood structures

groups

the grouping of the (non-intercept) coefficients. A vector of length (p-1)

nLambda

the length of the lasso path, defaults to 101

centered

logical, TRUE for centered and FALSE for uncentered, default is TRUE

BIC

logical, default TRUE uses BIC and FALSE uses AIC

select

logical, TRUE for variable selection step, default is TRUE

standardize

logical, standarize non-intercept columns? Default is TRUE, FALSE is almost certainly a bad idea

constraint

"symmetric" or "diagonal"

polish

use full Newton-Raphson to sharpen convergence from BFGS

innerIndices

indices of internal grid points

Value

a list containing the (group) lasso path,

Examples

# NOT RUN {
#
# }