As part of the Canadian Distributed Mentorship Program, here is where I will be posting my journal entries relating to the work done.

I've had random verses from The Gashlycrumb Tinies by Edward Gorey stuck in my head for most of the weekend.

My mother, who duly reads this journal, has asked me to write more serious entries, with less of a focus on llamas. I foresee problems, as there is so much one wants to read about the adventures of the undergrad in research-land, but we'll give it a try.

Nortel has finally sent us the simulator data, so we can begin the project. Doina, Pablo, Eric and I had a meeting early in the week, to discuss the project. We are expected to handle the "code blue" emergencies, based on the training data provided. We will probably use decision trees to generate the rules, as they are fairly simple, and there exists a large machine learning library, called Weka, that has all the tree-building and rule-learning algorithms needed.

The other approach would be a clustering algorithm. Rather than the top-down approach of trees (massive amount of data that gets structured), clustering starts from the bottom-up. Basically, the algorithm tries to bundle training instances up in a relevant fashion, and each of these "bundles" will be a rule. The great thing about the Weka library is that you needn't be concerned with finding the algorithms, as they are all hidden behind a very nice interface.

This week, Eric and I are mostly trying to figure out what attributes we can use in the trees. I can't really talk about the data that we have, but you can imagine that not all of it is useful in learning.
This weekend I think I might write some code and use the Weka library to solve a decision-tree-like problem, so that next week I can try to actually use our data with it.

As the weather has been atrocious the whole week, I have taken up math as a hobby. Prof. Panangaden, also part of the Reinforment Learning Lab, has been giving my office-mates and I interesting math problems every week. I think I might latex-ify the questions, and put them up on the webpage, to keep track of them.