We follow their approach because all our base
classifiers are equipped to output class probability distributions. In the second
part of chapter 3 concerned with meta classifier choice, we will show that this
extension is indeed competitive to Stacking with predictions