plsgenomics (version 1.5-3)

pls.regression.cv: Determination of the number of latent components to be used in PLS regression

Description

The function pls.regression.cv determines the best number of latent components to be used for PLS regression using the cross-validation approach described in Boulesteix and Strimmer (2005).

Usage

pls.regression.cv(Xtrain, Ytrain, ncomp, nruncv=20, alpha=2/3)

Value

The number of latent components to be used in PLS regression, as determined by cross-validation.

Arguments

Xtrain

a (ntrain x p) data matrix containing the predictors for the training data set. Xtrain may be a matrix or a data frame. Each row is an observation and each column is a predictor variable.

Ytrain

a (ntrain x m) data matrix of responses. Ytrain may be a vector (if m=1), a matrix or a data frame. If Ytrain is a matrix or a data frame, each row is an observation and each column is a response variable. If Ytrain is a vector, it contains the unique response variable for each observation.

ncomp

the vector of integers from which the best number of latent components has to be chosen by cross-validation. If ncomp is of length 1, the best number of components is chosen from 1,...,ncomp.

nruncv

the number of cross-validation iterations to be performed for the choice of the number of latent components.

alpha

the proportion of observations to be included in the training set at each cross-validation iteration.

Author

Anne-Laure Boulesteix (https://www.ibe.med.uni-muenchen.de/mitarbeiter/professoren/boulesteix/index.html) and Korbinian Strimmer (https://strimmerlab.github.io/korbinian.html).

Details

The cross-validation procedure described in Boulesteix and Strimmer (2005) is used to determine the best number of latent components to be used for classification. At each cross-validation run, Xtrain is split into a pseudo training set and a pseudo test set and the squared error is determined for each number of latent components. Finally, the function pls.regression.cv returns the number of latent components for which the mean squared error over the nrun partitions is minimal.

References

A. L. Boulesteix and K. Strimmer (2005). Predicting Transcription Factor Activities from Combined Analysis of Microarray and ChIP Data: A Partial Least Squares Approach.

A. L. Boulesteix, K. Strimmer (2007). Partial least squares: a versatile tool for the analysis of high-dimensional genomic data. Briefings in Bioinformatics 7:32-44.

S. de Jong (1993). SIMPLS: an alternative approach to partial least squares regression, Chemometrics Intell. Lab. Syst. 18, 251--263.

See Also

pls.regression, TFA.estimate, pls.lda.cv.

Examples

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

# load Ecoli data
data(Ecoli)

# determine the best number of components for PLS regression using the cross-validation approach
# choose the best number from 1,2,3,4
pls.regression.cv(Xtrain=Ecoli$CONNECdata,Ytrain=Ecoli$GEdata,ncomp=4,nruncv=20)
# choose the best number from 2,3
pls.regression.cv(Xtrain=Ecoli$CONNECdata,Ytrain=Ecoli$GEdata,ncomp=c(2,3),nruncv=20)

}