Ensemble Learning

Boosting

Reference Foundations of Machine Learning Chapter 6
Idea: Build a strong learner from a set of weak learners.

Adaboost

Learn a linear combination of our weak learners.

Given a sample of size m
for i=1:m
  d_i=1/m
for t=1:T
  h_t <- classifier
  alpha_t <- (1/2)log((1-eps_t)/eps_t)
  z_t <- e[eps_t(1-eps_t)]^(1/2)
  for i=1:m
    D_{t+1} <- (D_t(i)exp(-alpha_t*y_i*h_t(x_i))/z_t
g <- sum alpha_t h_t

Bagging

Bagging Predictors
Bootstrap aggregation
Idea: Given a sample S, bootstrap from the sample to get m samples S_1,...,S_m.
Then build m classifers from those samples
Your new classifier is a linear combination of those classifiers

References

• An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees: Bagging, Boosting, and Randomization