next up previous
Next: Bagging and Boosting [6+2* Up: MLA_Exercises_160106 Previous: AdaBoost [3 P]

AdaBoost Applet [3 P]

Use Yoav Freund's applet at to analyze what happens in single iterations of the AdaBoost algorithm. Try out the algorithm on two different datasets as shown in Figure 2. Make screenshots to show the decision boundaries and the sum of the weak learners' hypotheses at interesting iterations. Describe the behavior of the algorithm on the different datasets and document training and test error. Design a dataset where the algorithm fails. (Hint: Use a large number of data points. The weak learners here are axis-parallel splits.)
Figure 2: Target distributions for AdaBoost Applet
\includegraphics[scale = 0.8]{boost.eps}

Pfeiffer Michael 2006-01-18