Third homework: Neural Networks/Evolutionary Computation

I give up. All homeworks are due by December 14th, before midnight.

For this homework you can choose to work with neural networks or with evolutionary computation. Below are options for each. As usual, you can choose to work on something else, as long as you clear it with me first.

1) NN option.Modify the code you used for induction of decision trees so you can use it to train a neural network. The file generatedby your code (Breve or otherwise) should now create an SNNS pattern file. Select part of this file to use for training, and leave part for validation. Create a NN inside SNNS. Train the network with the generated pattern file, either from within SNNS (using the control panel) or via batchman. You are free to choose the learning function (back propagation, quickprop, etc.), learning parameter, momentum, weight decay, etc. that will work. Report on the efficiency of the trained NN on your chosen task.

2) Evolutionary computation option. By now you have files that implement genetic programming within the scope of Quidditch in Breve. Use these files to evolve a program that controls when the simplepayer code chooses to shoot at goal.