We spoke about the classification problem in terms of a linear separator, presenting the ideal probability estimate as a density of the distance from this margin. This can provide a benchmark with which to compare the estimates we obtain from our experiments. Dr. Dutta asked me to read a textbook to introduce me to linear algebra.
We discussed how the way you partition data affects whether your results will have ties, and how shrinking trees deals with this. Dr. Dutta asked me to re-read the shrinkage paper I had read two weeks ago, focusing on the sections on recursive shrinking and relating pruning to recursive shrinking.
On the programming side, Dr. Dutta asked me to look for java code implementing shrinkage, and to implement the AUC score in Eclipse.