Multigroup Partial Least Squares Regression
Multigroup PLS regression
mgPLS( DataX, DataY, Group, ncomp = NULL, Scale = FALSE, Gcenter = FALSE, Gscale = FALSE )
DataX |
a numeric matrix or data frame associated with independent dataset |
DataY |
a numeric matrix or data frame associated with dependent dataset |
Group |
a vector of factors associated with group structure |
ncomp |
number of components, if NULL number of components is equal to 2 |
Scale |
scaling variables, by defalt is FALSE. By default data are centered within groups |
Gcenter |
global variables centering, by defalt is FALSE. |
Gscale |
global variables scaling, by defalt is FALSE. |
list with the following results:
DataXm |
Group X data |
DataYm |
Group Y data |
Concat.X |
Concatenated X data |
Concat.Y |
Concatenated Y data |
coefficients |
Coefficients associated with X data |
coefficients.Y |
Coefficients associated with regressing Y on Global components X |
Components.Global |
Conctenated Components for X and Y |
Components.Group |
Components associated with groups in X and Y |
loadings.common |
Common vector of loadings for X and Y |
loadings.Group |
Group vector of loadings for X and Y |
expvar |
Explained variance associated with global components X |
cum.expvar.Group |
Cumulative explained varaince in groups of X and Y |
Similarity.Common.Group.load |
Cumulative similarity between group and common loadings |
Similarity.noncum.Common.Group.load |
NonCumulative similarity between group and common loadings |
A. Eslami, E. M. Qannari, A. Kohler and S. Bougeard (2013). Multi-group PLS regressMathematics and Statistics, Springer Proceedings (ed), New Perspectives in Partial Least Squares and Related Methods, 56, 243-255.
A. Eslami, E. M. Qannari, A. Kohler and S. Bougeard (2014). Algorithms for multi-group PLS. Journal of Chemometrics, 28(3), 192-201.
data(oliveoil)
DataX = oliveoil[,2:6]
DataY = oliveoil[,7:12]
Group = as.factor(oliveoil[,1])
res.mgPLS = mgPLS (DataX, DataY, Group)
barplot(res.mgPLS$noncumper.inertiglobal)
#----- Regression coefficients
#res.mgPLS$coefficients[[2]]
#----- Similarity index: group loadings are compared to the common structure (in X and Y spaces)
XX1= res.mgPLS$Similarity.noncum.Common.Group.load$X[[1]][-1, 1, drop=FALSE]
XX2=res.mgPLS$Similarity.noncum.Common.Group.load$X[[2]][-1, 1, drop=FALSE]
simX <- cbind(XX1, XX2)
YY1=res.mgPLS$Similarity.noncum.Common.Group.load$Y[[1]][-1, 1, drop=FALSE]
YY2=res.mgPLS$Similarity.noncum.Common.Group.load$Y[[2]][-1, 1, drop=FALSE]
simY <- cbind(YY1,YY2)
XLAB = paste("Dim1, %",res.mgPLS$noncumper.inertiglobal[1])
YLAB = paste("Dim1, %",res.mgPLS$noncumper.inertiglobal[2])
plot(simX[, 1], simX[, 2], pch=15, xlim=c(0, 1), ylim=c(0, 1),
main="Similarity indices in X space",
xlab=XLAB, ylab=YLAB)
abline(h=seq(0, 1, by=0.2), col="black", lty=3)
text(simX[, 1], simX[, 2], labels=rownames(simX), pos=2)
plot(simY[, 1], simY[, 2], pch=15, xlim=c(0, 1), ylim=c(0, 1),
main="Similarity indices in Y space",
xlab=XLAB, ylab=YLAB)
abline(h=seq(0, 1, by=0.2), col="black", lty=3)
text(simY[, 1], simY[, 2], labels=rownames(simY), pos=2)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.