Unfortunately I'm having problems getting results with BPGEP. It seems to find functions alright, but its terribly slow. I had to scale the parameters down enormously to get it to a reasonable time (It would have taken days if not weeks to do a single run with the parameters in the paper "Investigation of Constant Creation Techniques in the Context of Gene Expression Programming"). I'm okay with using different parameters because the point of the paper is to investigate BPGEP, not solve problems.
I would like to use the more complex function in the paper, the "V" shaped one, because it may be solved more easily when I add lots of constants to the terminal set, which I plan on doing.
Incidentally- why are all the fitness functions in symbolic regression papers so strange? The one in the original GEP research doesn't seem to force the fitness to be positive (unless I'm missing something) and the one in the paper mentioned above doesn't seem to make sense when the individual fitness is equal to the best found so far. The highest possible fitness would be .5, and if the residual is 0, the fitness would be 0. Its like I'm not interpreting it correctly.
No comments:
Post a Comment