> library(caret)
>
> #Sample data
> x1<-rnorm(1000,0,1)
> x2<-rnorm(1000,10,10)
> x3<-rnorm(1000,5,2)
> x4<-rnorm(1000,2,1)
> x5<-rnorm(1000,-2,3)
> yraw<-x1*0.4+x2*2+x3*1.5+rnorm(1000,0,7)
> yfin<-ifelse(yraw>(mean(yraw)+.5*sd(yraw)),1,0)
>
> df<-data.frame(y=yfin,x1=x1,x2=x2,x3=x3,x4=x4,x5=x5)
> df$y<-factor(df$y,levels=c(0,1),labels=c("peace","war"))
>
> #Split the data
> set.seed(123)
> z<-createDataPartition(df$y,p=0.75)
> TrainSet<-df[z$Resample1,]
> TestSet<-df[-z$Resample1,]
>
> #5-fold cross validation
> tc2<-createFolds(TrainSet$y,k=5)
> tc<-trainControl(index=tc2,
+                method="cv", number=5,
+                  summaryFunction=twoClassSummary, # provides ROC summary stats
+                  savePredictions="final",
+                  classProbs=T)
>
> model<-train(as.factor(y)~x1+x2+x3+x4+x5,
+              metric="ROC", method="glm", family="binomial",
+              trControl=tc, data=TrainSet)
>
> #Get held-out predictions
> z<-model$pred
> print(nrow(z)/nrow(TrainSet))
[1] 4
> z.ordered<-z[order(z$rowIndex),]
> head(z.ordered)
     parameter  pred   obs      peace          war rowIndex Resample
1         none peace peace 0.99999902 9.843203e-07        1    Fold1
1202      none peace peace 0.99999905 9.519363e-07        1    Fold3
1804      none peace peace 0.99999992 8.283829e-08        1    Fold4
2404      none peace peace 0.99999844 1.563241e-06        1    Fold5
2         none   war   war 0.01125721 9.887428e-01        2    Fold1
1203      none   war   war 0.01052763 9.894724e-01        2    Fold3
> print(head(tc2$Fold2))
[1]  1  2  6 11 13 15