Predict different models from K Best models in Caret I was wondering if there was a way to run predictions on the K best Models and not just the best Model in Caret. 
### Repeated CV ### 
tunedpar_RepeatedCV <- caret::train(x =XY_train[,-target], y =XY_train[,target], method= 
catboost.caret, metric= "ROC", maximize = TRUE, trControl=repeatedCV(YID=y,classProbsX=TRUE,methodX= 
"gls",CV_folds=5, CV_repeats=5),tuneLength=20)

predict(tunedpar_RepeatedCV,XY_test[,-target]

KBestModels=na.omit(tunedpar_RepeatedCV$results)
KBestModels=KBestModels[order(KBestModels$ROC,decreasing=TRUE),]        ### Order by ROC
KBestModels=KBestModels[(KBestModels$ROC-max(KBestModels$ROC))<CRIT,]   ### Subset to only Good Models

I just need to find out how to make predictions from KBestModels
 A: You can set savePredictions=TRUE under trainControl, and then go back after training to retrieve the predictions.
library(caret)
library(mlbench)
data(Sonar)

trControl=trainControl(method='cv', number=3,savePredictions=TRUE)

modl <- train(x = Sonar[,-ncol(Sonar)], 
y =Sonar[,ncol(Sonar)],method="gbm",trControl=trControl,tuneLength=10)

So the predictions are stored under:
head(modl$pred)
  pred obs rowIndex shrinkage interaction.depth n.minobsinnode n.trees Resample
1    R   R        2       0.1                 1             10     500    Fold1
2    R   R        6       0.1                 1             10     500    Fold1
3    R   R       13       0.1                 1             10     500    Fold1
4    M   R       16       0.1                 1             10     500    Fold1
5    R   R       26       0.1                 1             10     500    Fold1
6    R   R       27       0.1                 1             10     500    Fold1

Then using your code:
CRIT=0.01
KBestModels=modl$results
KBestModels=KBestModels[order(KBestModels$Accuracy,decreasing=TRUE),]
KBestModels=KBestModels[KBestModels$Accuracy[1]-KBestModels$Accuracy<CRIT,]

To get predictions do a merge:
KBest_pred = merge(KBestModels,modl$pred)
dim(KBest_pred)
[1] 624  12
> dim(Sonar)
[1] 208  61

