January 2008


Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Bradford Barr <[log in to unmask]>
Reply To:
ECJ Evolutionary Computation Toolkit <[log in to unmask]>
Sat, 5 Jan 2008 18:42:33 -0500
text/plain (45 lines)
I realized that after double checking on all of my params. Sorry  
about the bad subject for the email. I'm going to see if it is a  
rounding error. Thanks for all your help again.


On Jan 5, 2008, at 4:34 PM, Sean Luke wrote:

> Brad, the ERC classes and KozaFitness have literally nothing to do  
> with one another.    So I'm not sure that's the problem, unless  
> your ERC class does some bizarre like reach into the individual's  
> fitness and fools with it (not likely!).
> Perhaps there's a rounding error or a simplification in what's  
> getting printed?  KozaFitness doesn't define an ideal individual as  
> having an adjusted fitness of 1.0: it defines an ideal individual  
> as having a *raw* fitness (and thus a standardized fitness) of  
> *0.0*.  You might check to see if the individuals you have with 1.0  
> adjusted fitnesses indeed have a raw fitness of precisely 0.   
> Perhaps some raw fitnesses very close to 0 might produce 1.0 in a  
> Java float.
> Sean
> On Jan 5, 2008, at 10:19 AM, Bradford Barr wrote:
>> Hey list,
>> I have a question.
>> I created an integer ERC class using RegERC as a template (the one  
>> that comes in the regression package), but when I use it my  
>> finesses, and some of my params start to work funny.
>> I have quit-on-run-complete = true which means it should quit when  
>> it finds a solution, but it doesn't. It reports in the stat file  
>> (I'm using KozaStatistics and KozaFitness) that the best  
>> individual has an adjusted fitness of 1.0 but it doesn't quit.
>> I was wondering if this has happened to anyone else and how I can  
>> fix this problem.
>> Thanks in advance,
>> Brad