I'm building a model ensemble with caretStack (package caretEnsemble). Here is a basic example :
library(caret);library(caretEnsemble) models <- caretList( x=iris[1:100,1:2], y=iris[1:100,3], trControl=trainControl(method="cv"), methodList=c("lm","rpart", "glm") ) meta_model <- caretStack(models, method="rf")
First, I would like a confirmation that I understand how this function works. Correct me if I am wrong :
- caretList uses lm,rpart and glm to fit x1 and x2 to y (i.e., y~x1+x2)
- This gives three predictions : plm, prpart, pglm
- Then caretStack creates the model ensemble and provides a global prediction with rf such as y~plm+prpart+pglm
Second, I think that, depending of the predictor values, one model or another will work better, thus I would like that caretStack takes into account the x1 and x2 predictors y~plm+prpart+pglm+x1+x2
Thus, is it possible to include predictors (eg x1, x2 in my example) in the caretStack ensemble modeling ? Thanks.