bootsPLS (version 1.1.2)

plot.predictCI: Plot confidence Intervals

Description

Display the confidence intervals for the prediction values of a specific level of the outcome

Usage

# S3 method for predictCI
plot(x,
ncomp=1, level=1, las=2, col, title, name.var = TRUE, abline=TRUE,...)

Arguments

x

a `predict.CI' object, as one resulting from CI.prediction, or from prediction$out.CI.

ncomp

Component to be plotted

level

One of the levels of the outcome to plot

las

las argument for the x-axis labels

col

Color for the confidence Intervals. Can be a single value or a different value for each sample.

title

title of the plot

name.var

If TRUE, add the name of the variables on the x-axis. A vector of length the number of variables can be supplied for customized names.

abline

logical. Add two horizontal lines: 0.5 and 1/the number of levels of the outcome. Note that if the outcome only has two levels, only one line is plotted.

…

not used

Details

Plot the Confidence Intervals for a specific component and a specific level. In the special case of the "max.dist" being used (in the bootsPLS function), a line at 0.5 and 1/number of levels is added to the plot with abline=TRUE. These lines highlight the predicted class of the samples. For example if the level 1 out of 2 levels is displayed: any sample above 0.5 is predicted as level 1, any samples below 0.5 is predicted as level 2, and samples with a confidence interval overlaying 0.5 are not a clear cut. Another example if the level 1 out of 3 levels is displayed: any sample above 0.5 is predicted as level 1, any samples below 1/3 is predicted as not level 1, and samples in between 1/3 and 0.5 can't be decided from this graph only (you can display other levels, or find the predicted class in your object). See example below.

In the case of another distance being used, the lines should not be added to the graph abline=FALSE.

See Also

CI.prediction, prediction

Examples

# NOT RUN {
data(MSC)
X=MSC$X
Y=MSC$Y


# with a bootsPLS object
boot=bootsPLS(X=X,Y=Y,ncomp=3,many=5,kCV=5)
fit=fit.model(boot,ncomp=3)

CI=CI.prediction(fit,X.test=X)
plot(CI)

pred=prediction(fit,X.test=X, CI=TRUE)
plot(pred$out.CI, ncomp=2)
# we color each confidence interval by the predicted class
plot(pred$out.CI, ncomp=2,
    col = color.mixo(factor(pred$predicted.test$"max.dist"[,2]))) #second component


# because bootsPLS was used with dist="max.dist" (by default)
# and because there are only two levels in the outcome,
# everything above the 0.5 line is predicted as MSC

par(mfrow=c(3,1))
plot(pred$out.CI, ncomp=1, level="MSC")
plot(pred$out.CI, ncomp=2, level="MSC")
plot(pred$out.CI, ncomp=3, level="MSC")


#we can do the same things for the second level (Non-MSC)
par(mfrow=c(3,1))
plot(pred$out.CI, ncomp=1, level="Non-MSC")
plot(pred$out.CI, ncomp=2, level="Non-MSC")
plot(pred$out.CI, ncomp=3, level="Non-MSC")

# }