Week 6

Today we discussed the work I had done on implementing the M-estimate in Weka. Dr. Dutta suggested going in a different direction, working with the J48 tree on its own instead of modifying the Weka GUI.

We reviewed the Multi-class ROC paper and went more deeply into its formal definition of the AUC measure. Dr. Dutta described the problem she would like to work on, connecting the Laplace correction to the AUC, based on this definition.

We reviewed the paper on shrinking trees and discussed its relevance.

On Dr. Dutta's suggestion, I abandoned my earlier efforts in working with the Weka GUI and instead ran the J48 tree on its own from the console, a solution that was less exciting but much more practical.

I spent quite a lot of time making sure I understood the formal definition of the AUC measure and thinking about how the Laplace correction might relate to it.


[Previous] [Next]