The results from Fowler and Laver's Tournament of Party Strategies (previously blogged) are up here. I was chuffed to see that my strategy (Raptor) came in 8th - just beating Hunter, the pre-submitted strategy on which it was based. They plan to rerun the tournament using only those strategies which beat Hunter. I'm optimistic that with the smaller set of strategies used, Raptor (which uses a predefined step size to recognize and cooperate with conspecifics) will continue to do well.
Interestingly, Fowler and Laver would also like to develop evolutionary rules that allow successful party strategies to evolve and mutate. They worry about how to parametrize party strategies. One option would be to use genetic programming (not to be confused with genetic algorithms): each party strategy is a parse tree of R statements, and each round, the parse tree is set to work on the inputs, and outputs a new location. Mutations could consist of prunings and duplications of the tree, as well as adding new branches.
No comments:
Post a Comment