Re: Crossover operator for a superposed individual set

"Steve McGrew" <[EMAIL PROTECTED]> escribió en el mensaje
news:[EMAIL PROTECTED]
> On Wed, 19 Nov 2003 20:56:52 +0100, "José Gomila Fàbregues"
> <[EMAIL PROTECTED]> wrote:
>
> >Hello,
> >
> >I'm a telecommunication science student, and I am working with GA. We
have
> >to optimitze a very hard problem. Our goal is to find a solution that is
> >formed by a superposed set of individuals. That is, individuals that are
> >forming our population are a set of parametres that are repeated because
are
> >superposed. An example:
> >an simple individual can be formed by 4 parameters: X,Y,Z, (for spatial
> >position) and V (a value in that position), or a superposed set of those
> >parametres. We want to find an hipotetical composed individual formed by
a 4
> >simple individuals that are representated by
> >        X1,Y1,Z1,V1
> >        X2,Y2,Z2,V2
> >        X3,Y3,Z3,V3
> >        X4,Y4,Z4,V4
> >The individual are formed by 4x4 parameters now.
> >How can i write a good code to implement the crossover operator? Some
body
> >have an idea? May be a correlation between paramters can be a good
option,
> >but in the first generations there is not correlations.
> >We work now with one point classical crossover operator and we have som
> >goods solutions, but can it be optimitzed?
> >Thank  you.
> >
>        Will your solution be a weighted sum of several individuals?
> E.g., A{x1,y1,z1,v1} + B{x2,y2,z2,v2} + ...?
>
>       It is important to think about what aspects of a trial solution
> should survive crossover in order to ensure that the result a) is a
> valid trial solution and b) has a reasonable likelihood of retaining
> the aspects of the trial solution that give it a high enough fitness
> to be participating in creating the next generation of trial
> solutions.  For example,  should crossover just involve exchange of a
> subset of the individuals in each of two sets?  Or can it involve
> exchange of, say the z coordinates from one subset with the z
> coordinates of a corresponding subset of another set?
>       One of the sample problems given with the Generator download
> http://www.nli-ltd.com/products/genetic_algorithms/demos.htm, "Wasps",
> looks for the optimum way to distribute several "bug bombs" in order
> to kill the largest number of wasps whose nests are distributed over a
> region.  A trial solution is the set of coordinates of all the "bug
> bombs".  As it turns out, in that particular problem a very simple
> two-point crossover operator works fine.
>       However, the best crossover and mutation operators for any
> particular problem depend strongly on the internal structure of the
> problem.
>
> SPM

Thank you very much Steve,

    I will try 2-point crossover operator.
    The problem is that 1-point crossover operator cant guarantee that the
good part of the individual will be mixed with the good part of the others
individuals.
        individual A:      xa1 ya1 za1 va1
                                xa2 ya2 za2 va2
                                xa3 ya3 za3 va3
        individual B:      xb1 yb1 zb1 vb1
                                xb2 yb2 zb2 vb2
                                xb3 yb3 zb3 vb3
        children:           xa1  ya1 za1 va1    (by 1-point xover)
                                xa2 yb2 zb2 vb2
                                xb3 yb3 zb3 vb3
Because the subsets of the indivual are not puted in order, is too difficult
belive that the children has the best parts of each parent. In the subset
formet by both parent individuals (xa2 yb2 zb2 vb2), is xa2 with the others
parameters yb2 zb2 vb2 a good combination?
Well, I have to study with 2-points crossover. I will explain my experience
with it.
Thanks.
Pepe.