[galib] Fine tuning of Galib

Chris Chapman cnchapman at msn.com
Sat Mar 17 17:07:08 EDT 2012

Hi Paul --
You'd need to tell us a little more about the problem: e.g., a general sense of what you're optimizing, the range of solutions (is 4 particularly close to 0 or not?), and any insight you have about the granularity of the fitness landscape (e.g., should one realistically ever expect to reach 0)
In general, having partial credit in a fitness function (which it sounds like have) is a good thing; you might reconsider however that works and tinker with it (e.g., transform the distance; combine a second measure; etc). Trying a larger population and different crossover + mutation methods are the other obvious levers.  Beyond that and similar generic advice, there's not enough to go on.
> Date: Mon, 12 Mar 2012 22:12:25 +0000
> From: phhs80 at gmail.com
> To: galib at mit.edu
> Subject: [galib] Fine tuning of Galib
> Dear All,
> I am solving a problem with Galib, whose objective function is 0 at
> the (theoretical) solution. With Galib, I am getting solutions close
> to 0 but never 0 (sometimes I get 4 or even 1). I have played with
> several parameters of the GA, but with no remarkable success. Have
> more experienced users some suggestions to offer me?
> Thanks in advance,
> Paul
> _______________________________________________
> galib mailing list
> galib at mit.edu
> http://mailman.mit.edu/mailman/listinfo/galib
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/pipermail/galib/attachments/20120317/1533deb5/attachment.htm

More information about the galib mailing list