Some things

What I've been learning about lately is genetic programming. You can read about the description of it at Wikipedia if you're unfamiliar with it.

I found a couple of interesting papers, one of which is available online in PDF form: Integrating Genetic Algorithms and Text Learning for Financial Prediction by Thomas and Sycara of CMU. They give very interesting (profitable) results from creating a simple committee (weighted average) system from both evolved trading rules and the output of text classifiers, both working on downloaded stock bulletin board data. Unfortunately, I think manually training a text classifier just doesn't scale up to large data sets, mainly because I'm lazy. Also, they don't take into account transaction costs.

That's why I like clustering, self-organizing maps, and other unsupervised methods so much: I hate tagging and labelling stuff by hand.

last updated 2 years ago # history