CSCE 633
Machine Learning (Spring 2010)
due: Tues, Mar 23

Project 2
---------

Goal: Implement and test a Back-prop neural network.  

Implement it as feedfoward, fully-connected between layers.
Allow 0-2 layers, and allow number of nodes in hidden layer(s) as
an argument.

Remember to use unthresholded output units for regression (prediction
of continuous target values; remember to adapt your update function
accordingly for the output node), or thresholded for classification
problems (targets: 0, 1; or 0.1 and 0.9, which sometimes helps
convergence).

You will have to use a validation set to monitor mean-squared error
and determine stopping criterion.

Please evaluate the effect of number of hidden layers and/or hidden
nodes on the accuracy of your network.

You should compare your neural network's performace to your implementation
of a decision tree (with pruning) on several databases, and discuss
which is better and why (using proper statistical comparisons).