plsgenomics (version 1.5-3)

rpls.cv: Determination of the ridge regularization parameter and the number of PLS components to be used for classification with RPLS for binary data

Description

The function rpls.cv determines the best ridge regularization parameter and the best number of PLS components to be used for classification for Fort and Lambert-Lacroix (2005) RPLS algorithm.

Usage

rpls.cv(Ytrain, Xtrain, LambdaRange, ncompMax, NbIterMax=50, ncores=1)

Value

A list with the following components:

Lambda

the optimal regularization parameter.

ncomp

the optimal number of PLS components.

Arguments

Xtrain

a (ntrain x p) data matrix of predictors. Xtrain must be a matrix. Each row corresponds to an observation and each column to a predictor variable.

Ytrain

a ntrain vector of responses. Ytrain must be a vector. Ytrain is a {0,1}-valued vector and contains the response variable for each observation.

LambdaRange

the vector of positive real value from which the best ridge regularization parameter has to be chosen by cross-validation.

ncompMax

a positive integer. the best number of components is chosen from 1,...,ncompMax. If ncompMax=0,then the Ridge regression is performed without reduction dimension.

NbIterMax

a positive integer. NbIterMax is the maximal number of iterations in the Newton-Rapson parts.

ncores

a positive integer. The number of cores to be used for parallel computing (if different from 1)

Author

Sophie Lambert-Lacroix (http://membres-timc.imag.fr/Sophie.Lambert/).

Details

A cross-validation procedure is used to determine the best ridge regularization parameter and number of PLS components to be used for classification with RPLS for binary data (for categorical data see mrpls and mrpls.cv). At each cross-validation run, Xtrain is split into a pseudo training set (ntrain-1 samples) and a pseudo test set (1 sample) and the classification error rate is determined for each value of ridge regularization parameter and number of components. Finally, the function mrpls.cv returns the values of the ridge regularization parameter and bandwidth for which the mean classification error rate is minimal.

References

G. Fort and S. Lambert-Lacroix (2005). Classification using Partial Least Squares with Penalized Logistic Regression, Bioinformatics, vol 21, n 8, 1104-1111.

See Also

rpls, mrpls, mrpls.cv.

Examples

if (FALSE) {
## between 5~15 seconds
# load plsgenomics library
library(plsgenomics)

# load Colon data
data(Colon)
IndexLearn <- c(sample(which(Colon$Y==2),12),sample(which(Colon$Y==1),8))

# preprocess data
res <- preprocess(Xtrain= Colon$X[IndexLearn,], Xtest=Colon$X[-IndexLearn,],
                    Threshold = c(100,16000),Filtering=c(5,500),
                    log10.scale=TRUE,row.stand=TRUE)
# the results are given in res$pXtrain and res$pXtest

# Determine optimum ncomp and lambda
nl <- rpls.cv(Ytrain=Colon$Y[IndexLearn]-1,Xtrain=res$pXtrain,LambdaRange=c(0.1,1),ncompMax=3)

# perform prediction by RPLS
resrpls <- rpls(Ytrain=Colon$Y[IndexLearn]-1,Xtrain=res$pXtrain,Lambda=nl$Lambda,
			ncomp=nl$ncomp,Xtest=res$pXtest)
sum(resrpls$Ytest!=Colon$Y[-IndexLearn]-1)

}