PLSbiplot1 (version 0.1)

SPLS.GLM.biplot: The Partial Least Squares (PLS) biplot for Sparse Partial Least Squares-Generalized Linear Model (SPLS-GLM)

Description

Takes in a set of predictor variables and a set of response variables and produces a PLS biplot for the (univariate) SPLS-GLMs.

Usage

SPLS.GLM.biplot(X, y, algorithm = NULL, eps, lambdaY = NULL, lambdaX = NULL, ax.tickvec.X = NULL, ax.tickvec.y = NULL, ax.tickvec.b = NULL, ...)

Arguments

X

A (NxP) predictor matrix

y

A (Nx1) response vector

algorithm

Any of the SPLS-GLM algorithm ("SPLS.GLM", "SPLS.binomial.GLM")

eps

Cut off value for convergence step

lambdaY

A value for the penalty parameters for the soft-thresholding penalization function for Y-weights

lambdaX

A value for the penalty parameters for the soft-thresholding penalization function for X-weights

ax.tickvec.X

tick marker length for each X-variable axis in the biplot

ax.tickvec.y

tick marker length for the y-variable axis in the biplot

ax.tickvec.b

(purple) tick marker length for the y-variable axis in the biplot

...

Other arguments. Currently ignored

Value

The PLS biplot of a SPLS-GLM of D=[X y] with some parameters

Examples

if(require(robustbase))
possum.mat
y = as.matrix(possum.mat[,1], ncol=1)
dimnames(y) = list(paste("S", 1:nrow(possum.mat), seq=""), "Diversity")
X = as.matrix(possum.mat[,2:14], ncol=13)
dimnames(X) = list(paste("S", 1:nrow(possum.mat), seq=""), colnames(possum.mat[,2:14]))
#choosing a value for the penalty parameters lambdaY and lambdaX for this data
main2B = opt.penalty.values(X=scale(X), Y=scale(y), A=2, algorithm=SPLS.GLM, eps=1e-3,
from.value.X=1, to.value.X=4, from.value.Y=0, to.value.Y=0, lambdaY.len=1, lambdaX.len=100)
min.RMSEP.value = main2B$min.RMSEP.value
lambdaY.to.use = main2B$lambdaY.to.use
lambdaX.to.use = main2B$lambdaX.to.use
list(lambdaY.to.use=lambdaY.to.use, lambdaX.to.use=lambdaX.to.use, min.RMSEP.value=min.RMSEP.value)
#SPLS-GLM analysis
main3 = SPLS.GLM(scale(X), scale(y), A=2, lambdaY=lambdaY.to.use, lambdaX=lambdaX.to.use, eps=1e-3)
X.to.use = main3$X.select
X.new = as.matrix(X[,names(X.to.use)])
colnames(X.new)
main3$Y.select #note
SPLS.GLM.biplot(X.new, y, algorithm=SPLS.GLM, eps=1e-3, lambdaY=lambdaY.to.use,
lambdaX=lambdaX.to.use, ax.tickvec.X=c(10,5,5,5,5,5,5,5,5,5,5,5,5), ax.tickvec.y=8,
ax.tickvec.b=12)

#Pima.tr data
if(require(MASS))
data(Pima.tr, package="MASS")
X = as.matrix(cbind(Pima.tr[,1:7]))
dimnames(X) = list(1:nrow(X), colnames(X))
y = as.matrix(as.numeric(Pima.tr$type)-1, ncol=1)
#0=No and 1=Yes
dimnames(y) = list(1:nrow(y), paste("type"))
main2 = opt.penalty.values(X=scale(X), Y=scale(y), A=2, algorithm=SPLS.binomial.GLM,
eps=1e-3, from.value.X=0, to.value.X=95, from.value.Y=0, to.value.Y=0, lambdaY.len=1,
lambdaX.len=100)
min.RMSEP.value = main2$min.RMSEP.value
lambdaY.to.use = main2$lambdaY.to.use
lambdaX.to.use = main2$lambdaX.to.use
#SPLS-GLM analysis
main3 = SPLS.binomial.GLM(scale(X), scale(y), A=2, lambdaY=lambdaY.to.use,
lambdaX=lambdaX.to.use, eps=1e-3)
X.to.use = main3$X.select
X.new = as.matrix(X[,names(X.to.use)])
colnames(X.new)
SPLS.GLM.biplot(X.new, y, algorithm=SPLS.binomial.GLM, eps=1e-3, lambdaY=lambdaY.to.use,
lambdaX=lambdaX.to.use, ax.tickvec.X=c(3,3,3,3,3,3,1), ax.tickvec.y=3, ax.tickvec.b=3)