Code created for PSU's machine learning class. Based around the UCI Letter Recognition dataset, it is not general purpose.
Letter recognition using simple binary perceptrons. Uses all-pairs method to classify instances and produces a confusion matrix of the results. Achieves aroung 70% accuracy on test set.
A multilayer neural net with 1 hidden layer. Performs best with eta=.3, momentum=.3, and at least 64 hidden units, achieving over 80% accuracy.