15.1 Process Description

  1. Set up the ensemble by
  • Defining a list \(L\) of tuned based learners
  • Defining a meta learner algorithm (usually some form of regularized regression)
  1. Train the ensemble by
  • Training each base learner
  • Performing k-fold CV on each of the base learners and collect the cross-validated predictions from each, to avoid overfitting as they would be predicting new data.
  • Creating the \(Z\) feature matrix of \(N \times L\) known as “level-one”
  • Training the meta learning algorithm on the level-one data \(y = f(Z)\)
  1. To make predicts
  • Generate predictions from each base learner
  • Feed those predictions into the meta learner to generate the ensemble prediction