You would like to build a classifier that is robust against errors such as the ones described in the previous example.

You would like to build a classifier that is robust against errors such as the ones described in the previous example. Namely, your classifier should be robust against errors that "erase" up to half the hyperplanes you create. In particular, you need to create 2L hyperplanes, knowing that, your classifier will use only L of them, but you do not know in advance which L. You should use the implementation template given in part I to build your classifier, but you can design how you select the hyperplanes in a different way. Your implementation of the classifier should have the following additional method:

• TestCorrupted: this method should be used to test input data on the classifier with corrupted hyperplanes. The method takes as input 1. a data point x to be classified, 2. a vector I which contains the indices of the corrupted hyperplane. The function would then attempt to classify the input x using only the hyperplanes in t1, 2, ¨ ¨ ¨ , 2LuzI, that is, the hyperplanes other than the ones indicated in I. Your classification proceduce should be the same, regardless of which hyperplanes are corrupted. In other words, you should not have a classification procedure for one set I1 and a different procedure for another set I2 and so on.