Saturday, March 28, 2009

BinInt, finally!

After much delay, I finally got my BinInt GA problem running successfully in ECJ. Part of the issue was setting everything up again on a new laptop, part also was moving to ECJ 18, and part was my own stupidity. But it looks like it is all working now.

BinInt takes the same bit vector as MaxOnes and divides it into several subproblems. Each subproblem is treated as a binary integer, hence the name. The key new feature of this problem is the introduction of exponential scaling. Some bits are worth a lot more than others.

I wrote the fitness function so that it scales the fitness according to the size and number of subproblems. Best fitness is always 1. this way, we can vary the size and number of the subproblems and directly compare the trajectory of different parameter choices in solving the problem.

No comments: