My experience with using a linear model for Tic Tac Toe learning

I have put up my results here: http://skothawala.com/2014/12/15/my-experience-with-using-a-linear-model-for-tic-tac-toe-learning/

The results, in a nutshell, were that given a set of features, the learned linear model failed to show any significant improvement over one using random weights. It did, however, show a significant improvement over a player making moves at random. The credit, I conclude, is due to the algorithm used by the learning player to make a move.

I don't have much formal training in ML and am trying to learn on my own. I would very much like to learn about any bugs or flaws in my approach, and any suggestions from more experienced practitioners will be much appreciated.

Thanks for checking it out.