PLR: A function which performs penalised logistic regression...

In MCRestimate: Misclassification error estimation with cross-validation

Description

A function which performs penalised logistic regression.

Usage

PLR(trainmatrix, resultvector, kappa=0, eps=1e-4)
       ## S3 method for class 'PLR'
predict(object,...)

Arguments

resultvector	a vector which contains the labeling of the samples
trainmatrix	a matrix which includes the data. The rows corresponds to the observations and the columns to the variables.
kappa	value range for penalty parameter. If more that one parameter is specified the one with the lowest AIC will be used.
eps	precision of convergence
object	a fitted PLR model
...	here a data matrix from samples that should be predicted

Value

a list with three arguments

a	Intercept estimate of the linear predictor
b	vector of estimated regression coefficients
factorlevel	levels of grouping variable
aics	vector of AIC values with respect to penalty parameter kappa
trs	vector of effective degrees of freedom with respect to penalty parameter kappa

Author(s)

Axel Benner, Ulrich Mansmann, based on MathLab code by Paul Eilers

Examples

library(golubEsets)
data(Golub_Merge)
eSet<-Golub_Merge
X0 <- t(exprs(eSet))
m <- nrow(X0); n <- ncol(X0)
y <- pData(eSet)$ALL.AML
f <- PLR(X0, y,kappa=10^seq(0, 7, 0.5))
if (interactive()) {
  x11(width=9, height=4)
  par(mfrow=c(1,2))
plot(log10(f$kappas), f$aics, type="l",main="Akaike's Information Criterion", xlab="log kappa", ylab="AIC")
plot(log10(f$kappas), f$trs, type="l",xlab="log kappa",
ylab="Dim",main="Effective dimension")
}

MCRestimate documentation built on Oct. 31, 2019, 10:29 a.m.