plot.tune: Plot model performance
In mixOmicsTeam/mixOmics: Omics Data Integration Project

plot.tune

R Documentation

Plot model performance

Description

Function to plot performance criteria, such as classification error rate or correlation of cross-validated components for different models.

Function to plot performance criteria, such as classification error rate or balanced error rate on a tune.splsda result.

Usage

## S3 method for class 'tune.spls'
plot(
  x,
  measure = NULL,
  comp = c(1, 2),
  pch = 16,
  cex = 1.2,
  title = NULL,
  size.range = c(3, 10),
  sd = NULL,
  ...
)

## S3 method for class 'tune.block.splsda'
plot(x, sd = NULL, col, ...)

## S3 method for class 'tune.spca'
plot(x, optimal = TRUE, sd = NULL, col = NULL, ...)

## S3 method for class 'tune.spls1'
plot(x, optimal = TRUE, sd = NULL, col, ...)

## S3 method for class 'tune.splsda'
plot(x, optimal = TRUE, sd = NULL, col, ...)

Arguments

`x`	an `tune.splsda` object.
`measure`	Character. Measure used for plotting a `tune.spls` object. One of c('cor', 'RSS').
`comp`	Integer of length 2 denoting the components to plot.
`pch`	plot character. A character string or a vector of single characters or integers. See `points` for all alternatives.
`cex`	numeric character (or symbol) expansion, possibly vector.
`title`	Plot title.
`size.range`	Numeric vector of length 2. Range of sizes used in plot.
`sd`	If 'nrepeat' was used in the call to 'tune.splsda', error bar shows the standard deviation if sd=TRUE
`...`	Not currently used.
`col`	character (or symbol) color to be used, possibly vector. One colour per component.
`optimal`	If TRUE, highlights the optimal keepX per component

Details

plot.tune.splsda plots the classification error rate or the balanced error rate from x$error.rate, for each component of the model. A lozenge highlights the optimal number of variables on each component.

plot.tune.block.splsda plots the classification error rate or the balanced error rate from x$error.rate, for each component of the model. The error rate is ordered by increasing value, the yaxis shows the optimal combination of keepX at the top (e.g. ‘keepX on block 1’_'keepX on block 2'_‘keepX on block 3’)

plot.tune.spls plots either the correlation of cross-validated components or the Residual Sum of Square (RSS) values for these components against those from the full model for both t (X components) and u (Y components). The optimal number of features chosen are indicated by squares.

If neither of the object$test.keepX or object$test.keepY are fixed, a dot plot is produced where a larger size indicates the strength of the measure (higher correlation or lower RSS). Otherwise, the measures are plotted against the number of features selected. In both cases, the colour shows the dispersion of the values across repeated cross validations.

plot.tune.spca plots the correlation of cross-validated components from the tune.spca function with respect to the full model.

Value

none

plot arguments for pls2 tuning

For tune.spls objects where tuning is performed on both X and Y, arguments 'col.low.sd' and 'col.high.sd' can be used to indicate a low and high sd, respectively. Default to 'blue' & 'red'.

Author(s)

Kim-Anh Lê Cao, Florian Rohart, Francois Bartolo, Al J Abadi

Kim-Anh Lê Cao, Florian Rohart, Francois Bartolo, AL J Abadi

Examples


## Not run: 
## validation for objects of class 'splsda'

data(breast.tumors)
X = breast.tumors$gene.exp
Y = as.factor(breast.tumors$sample$treatment)
out = tune.splsda(X, Y, ncomp = 3, nrepeat = 5, logratio = "none",
test.keepX = c(5, 10, 15), folds = 10, dist = "max.dist",
progressBar = TRUE)


plot(out, sd=TRUE)

## End(Not run)
## Not run: 
## validation for objects of class 'mint.splsda'

data(stemcells)
data = stemcells$gene
type.id = stemcells$celltype
exp = stemcells$study

out = tune(method="mint.splsda", X=data,Y=type.id, ncomp=2, study=exp, test.keepX=seq(1,10,1))
out$choice.keepX

plot(out)

## validation for objects of class 'mint.splsda'

data("breast.TCGA")
# this is the X data as a list of mRNA and miRNA; the Y data set is a single data set of proteins
data = with(breast.TCGA$data.train, list(mrna = mrna, 
            mirna = mirna,
            protein = protein,
            Y = subtype))
# set number of component per data set
ncomp = 5

# Tuning the first two components
# -------------

# definition of the keepX value to be tested for each block mRNA miRNA and protein
# names of test.keepX must match the names of 'data'
test.keepX = list(mrna = seq(10,40,20), mirna = seq(10,30,10), protein = seq(1,10,5))

# the following may take some time to run, note that for through tuning
# nrepeat should be > 1
tune = tune.block.splsda(X = data, indY = 4,
ncomp = ncomp, test.keepX = test.keepX, design = 'full', nrepeat = 3)

tune$choice.ncomp
tune$choice.keepX

plot(tune)
## --- spls model
data(nutrimouse)
X <- nutrimouse$gene  
Y <- nutrimouse$lipid
list.keepX <- c(2:10, 15, 20)
# tuning based on correlations
set.seed(30)
## tune X only
tune.spls.cor.X <- tune.spls(X, Y, ncomp = 3,
                           test.keepX = list.keepX,
                           validation = "Mfold", folds = 5,
                           nrepeat = 3, progressBar = FALSE,
                           measure = 'cor')

plot(tune.spls.cor.X)
plot(tune.spls.cor.X, measure = 'RSS')

## tune Y only
tune.spls.cor.Y <- tune.spls(X, Y, ncomp = 3,
                           test.keepY = list.keepX,
                           validation = "Mfold", folds = 5,
                           nrepeat = 3, progressBar = FALSE,
                           measure = 'cor')

plot(tune.spls.cor.Y)
plot(tune.spls.cor.Y, sd = FALSE)
plot(tune.spls.cor.Y, measure = 'RSS')

## tune Y and X
tune.spls.cor.XY <- tune.spls(X, Y, ncomp = 3,
                             test.keepY = c(8, 15, 20),
                             test.keepX = c(8, 15, 20),
                             validation = "Mfold", folds = 5,
                             nrepeat = 3, progressBar = FALSE,
                             measure = 'cor')

plot(tune.spls.cor.XY)
## show RSS
plot(tune.spls.cor.XY, measure = 'RSS')
## customise point sizes
plot(tune.spls.cor.XY, size.range = c(6,12))

## End(Not run)

mixOmicsTeam/mixOmics documentation built on Oct. 26, 2023, 6:48 a.m.