Extra Information for the Project

Here are some datasets that your algorithm should achieve near 0% error on:
equal4.arff, equal6.arff, equal8.arff, equal10.arff.
For each dataset, the classes are whether there are more zero bits than one bits, an equal number of zero bits and one bits, or more one bits than zero bits.

Using the perceptron algorithm, 100 epochs, and a learning rate of 0.1/(number of attributes), I obtained the following results (10-fold cross-validation).

Dataset     Error
glass       37%
hypothyroid  6%
iris         4%
letter      28%
segment      8%
vowel       33%

This is using the updating procedure in the project description.