Boosting: Summary

• Select a number of iterations T
• Initialize the weights for the training set
• For each iteration
  • Train a new hypothesis H, using the weighted training set
  • Compute the error for H
    • Sum of weights of misclassified training examples
  • Recalculate the weights using the newly calculated error
  • Calculate the weight for H
    • log((1-error) / error)
  • Append H to the list of hypotheses
• Return the list of hypotheses