Build a model

We have worked on the Liver Toxicity dataset, see Bushel, Wolfinger, and Gibson (2007). This data set contains the expression measure of 3116 genes and 10 clinical measurements for 64 subjects (rats) that were exposed to non-toxic, moderately toxic or severely toxic doses of acetaminophen in a controlled experiment. Therefore the structure is : \[\mathbf{X}\in\mathbb{R}^{64\times3116},\mathbf{Y}\in\mathbb{R}^{64\times10}\]

library(ddsPLS)
library(doParallel)
library(RColorBrewer)
data("liver.toxicity")
X <- scale(liver.toxicity$gene)
Y <- scale(liver.toxicity$clinic)
mddsPLS_model_reg <- mddsPLS(Xs = X,Y = Y,lambda=0.9,R = 1,
                             mode = "reg",verbose = TRUE)

## At most 2 variable(s) can be selected
##     For each block of X, are selected in order of component:
##         @ (2) variable(s)
##     For the Y block, are selected in order of component:
##         @ (2) variable(s)

Start cross-validation

res_cv_reg <- perf_mddsPLS(Xs = X,Y = Y,
                           R = 1,lambda_min=0.4,n_lambda=2,
                           mode = "reg",NCORES = 1,
                           kfolds = 2)

## Warning: executing %dopar% sequentially: no parallel backend registered

plot(res_cv_reg,legend_names=colnames(Y))

Build a model

The data set penicilliumYES has 36 rows and 3754 columns, see Clemmensen et al. (2007) The variables are 1st order statistics from multi-spectral images of three species of Penicillium fungi: Melanoconidium, Polonicum, and Venetum. These are the data used in the Clemmemsen et al “Sparse Discriminant Analysis” paper. Therefore the structure is, where \(\mathbf{Y}\) is the dummy matrix of the \(3\) classes : \[\mathbf{X}\in\mathbb{R}^{36\times3754},\mathbf{Y}\in\mathbb{R}^{36\times3}\]

data("penicilliumYES")
X <- penicilliumYES$X
X <- scale(X[,which(apply(X,2,sd)>0)])
Y <- as.factor(unlist(lapply(c("Melanoconidiu","Polonicum","Venetum"),
                             function(tt){rep(tt,12)})))
mddsPLS_model_class <- mddsPLS(Xs = X,Y = Y,lambda = 0.958,R = 2,
                               mode = "clas",verbose = TRUE)

## At most 3 variable(s) can be selected
##     For each block of X, are selected in order of component:
##         @ (2,1) variable(s)
##     For the Y block, are selected in order of component:
##         @ (1,1) variable(s)

Start cross-validation

res_cv_class <- perf_mddsPLS(X,Y,R = 2,lambda_min=0.94,n_lambda=2,
                             mode = "clas",NCORES = 1,
                             fold_fixed = rep(1:12,3))
plot(res_cv_class,legend_names = levels(Y))