NicMcPhee
commented
9 years ago 
This plot is with N=10 (the hardest version) with the default contribution type. These were really fast, so we can certainly do the other three contribution types just for fun. It definitely looks like we're winning, with XO bias definitely improving things at every tournament size!
The ECJ implementation supports up to N=20, with the problems presumably getting harder and harder as N gets larger (and perhaps the runs taking longer and longer, but I'm less clear about that). We don't want to do all 20 versions of the problem. I suspect that the smaller values of N are too easy to be interesting, but I don't know where the cut off point is. We might start with something like N=5, see what that does, then try N=10, etc.
The ECJ implementation provides four different contribution types that determine how the values of leaves are incorporated. I assume these all come from the paper where this paper is introduced, and I have no idea which of these we want to use. I guess there's no particular reason not to just try all four versions since we have the cluster. It might make sense to try one of them and see how long that takes, and then do the others if they're not going to take a zillion years.