[brettd]

I'm thinking of embarking on a data mined approach to generating angle plays on my chosen sport. Yes, I understand that angles can be data mined garbage, but I looking to implement a penalty function whereby the more complexity, the greater the penalty. This would act as an initial filter prior to evaluating the most promising angles for their inherent logic.

I played with Bayes' Information Criterion and Akaike's Information Criterion whilst evaluating forecasting models in my stats Master's, but have only made the conceptual leap to thinking about them in evaluating heuristics. Has anyone ever used these when evaluating sports models/angles? Or does anyone think they would be useful?

http://en.wikipedia.org/wiki/Bayesia...tion_criterion

http://en.wikipedia.org/wiki/Akaike_...tion_criterion

[uva3021]

if you are doing some sort of regression won't regularizing add a penalty, say

[tex]Function + \frac{\lambda**{n** \sum_i^n x_i[/tex]

sbr please add latex

F + c * sum(x_i)