How to use trained caret object to predict on new data (not used while training)?

Question

I am using caret package to train random forest model on training dataset. I have used 10-fold cross validation to get an object say randomForestFit. Now I would like to use this object to predict on new data set say test_data. I also want to get the respective class probabilities. How would I do that?

I have been using extractProb function as follows :

extractProb(randomForestFit, textX = test_data_predictors, testY = test_data_labels)

But it's giving me unexpected results.

Answer 1

From the extractProb help page example, you need to wrap the model in a list:

knnFit <- train(Species ~ ., data = iris, method = "knn", 
                trControl = trainControl(method = "cv"))

rdaFit <- train(Species ~ ., data = iris, method = "rda", 
                trControl = trainControl(method = "cv"))

predict(knnFit)
predict(knnFit, type = "prob")

bothModels <- list(knn = knnFit,
                   tree = rdaFit)

predict(bothModels)

extractPrediction(bothModels, testX = iris[1:10, -5])
extractProb(bothModels, testX = iris[1:10, -5])

So the following should work:

extractProb(list(randomForestFit), textX = test_data_predictors, testY = test_data_labels)

edit:

And yes, the preprocessing will be used. From the documentation:

These processing steps would be applied during any predictions generated using predict.train, extractPrediction or extractProbs (see details later in this document). The pre-processing would not be applied to predictions that directly use the object$finalModel object.