She couldn't send this to the list because she's not a member.
Sean
> From: Sara Silva <[log in to unmask]>
> Date: September 6, 2009 11:36:09 AM GMT+02:00
> To: Sean Luke <[log in to unmask]>
> Cc: ECJ Evolutionary Computation Toolkit <[log in to unmask]
> >
> Subject: Re: test best individual off line
>
>
> Hi Sean!
>
> I don't know if this will go through the mailing list, probably not.
>
> GPLAB implements two different protected divisions. The one from
> Koza is called 'kozadivide' and returns 1 when the denominator is 0.
> The other one is called 'mydivide' and returns the numerator when
> the denominator is 0. I don't recall why I implemented the second
> one... and I have no idea which one is best. Anyway, no bugs this
> time!! :)
>
> Cheers,
> Sara
>
>
> Sean Luke wrote:
>> On Sep 5, 2009, at 1:44 AM, Zheng Yin wrote:
>>> Found my problem. It turned out the answer is: the protected
>>> division in
>>> ECJ (the one in ecj\ec\app\regression\func)and the one in Matlab,
>>> which is
>>> from GPLAB are different. The answer from ECJ protected division
>>> equals
>>> one if denominator is zero. The answer for GPLAB's protected
>>> division is
>>> numerator if denominator is zero.
>>> :) which one is better?
>>>
>>> Zheng
>> ECJ follows the Koza tradition. GPLAB may not. Koza's approach
>> may not be the best, and there's no reason for GPLAB not to do
>> something else. However it's possible this wasn't intentional.
>> I'm cc:ing Sara Silva, who wrote GPLAB, in case it might be an
>> undocumented bug. [Hi Sara!]
>> Sean
>
>
|