creating multi-panel ROC curve plots with lattice - r

Question

I would like to create a plot which contains two panels, and each panel contains two ROC curve. To introduce my (failing) approach, I generate data frame containing the true label, labels for four methods (each method corresponds to)
N <- 20
TF <- rep(c(0,1),each=N/2)
pred <- method <- true <- NULL
for (imethod in 1 : 4){
pred <- c(pred,seq(-1,1,length.out=N) + rnorm(N) )
method <- c(method,rep(imethod,N))
true <- c(true,TF)
}
dat.roc <-
data.frame(true=true,pred=pred,method=method,panel=rep(1:2,each=length(method)/2))
xyplot(true ~ pred|panel, data=dat.roc,groups=method,
xlim=c(0,1),xlab="1-specificity",
ylab="sensitivity",
panel=function(x,y,...){
DD <- table(-x,y)
sens <- cumsum(DD[,2])/sum(DD[,2])
mspec <- cumsum(DD[,1])/sum(DD[,1])
panel.xyplot(mspec,sens,type="l",...)
panel.abline(0,1)
})
The plot have two panels, each of which has only ONE ROC curve (with two colors)! How can I correctly specify lattice to return two ROC curve in each panel?

Answer 1

Since you're using cumsum here in your panel function, you want to make sure that you are creating different plots for each group, not just each panel. One way to do this is to use the panel.superpose panel function So you would change your code to
xyplot(true ~ pred|panel, data=dat.roc,groups=method,
xlim=c(0,1),xlab="1-specificity",
ylab="sensitivity",
panel=panel.superpose,
panel.groups=function(x,y,type, ...){
DD <- table(-x,y)
sens <- cumsum(DD[,2])/sum(DD[,2])
mspec <- cumsum(DD[,1])/sum(DD[,1])
panel.xyplot(mspec,sens,type="l",...)
panel.abline(0,1)
})
which produces the plot