Assignment 5

A program to write out a solution is available at
http://www.cs.ubc.ca/~poole/cs422/2012/as5/code/dt_learner_sol.py
This program adds some extra arguments to the leaner and to the determining the error.
To use different datasets, change what is imported at the top to change the dataset.

When we are evaluating according to the absolute value, it predicts the mode at the leaves.
When evaluating for the other methods, it predicts a probability (with a pseudo count) 
at the leaves.

Note that for SPECT2 and SPECT3, none of the features myopically improved the prediction 
when splitting using the sum of absolute values. Thus the tree consists of a single node. 
Often the error are the same because the trees are the same.

Sam was suspicious of SPECT because the training and test sets contained very different 
proportions of the target feature.