ICS 273A
Winter 2011
Homework 7
Read section 4.1 and chapter 14 from Bishop Book.
1) Implement Adaboost on the Iris data. Use your logistic regression classifier (LRC) for
your weak learner. Try making the LRC weaker by applying it to a random subset of
features and/or training it only for a few iterations before adding it to the pool. Plot both
training error and testing error as a function of boosting round.
2) Derive the expression for SB and SW in kernel space.
3) Implement Kernel Fisher Discriminant analysis in the Iris data using an RBF kernel.
(use all 3 class labels). Make sure you add a regularization term to make the whole
procedure numerically stable. Plot the Iris data onto a 2-D plane obtained from KFDA.
Play with the amount of regularization and the width of RBF kernel and observe how the
projections change.