Re: Lineage selection

> When keeping track of n parents then you have a maximum guaranteed depth
> of log(n)(base 2) , so with 100 parents the maximum number of generations
> you can keep track of is 6.
Jou. So tree structure is needed then, then it don't grow but is 100
generations * pop_size. Big array.

> There are space efficent ways to keep track of parents but then you've got
> hassles when searching with them that will lead to long search times.
And yes it needs to be searched then I see it ok anyway.

> It all depends upon the length of your genome , the number of parents per
> generation and how far back you wanna keep.
Yes, so it is not _so_ good. Anyway wondering why it is not researched more.
But I don't like
individuals and compex comparisons with structures, as Java searching is
O(log) quite easily. And this works for any genome structures same way.

> Eventually you will reach a point where the further you go back the more
> likely you are to encounter every parent in that generation being an
> ancestor , so you are going to have find a balance between depth of
> ancestry and the number of children.
Yes, now I have tool to really see that. That is good point also, because
that I try to avoid. I am gonna use/test it together with fittness based
selection function.

> Maybe shallow depths are suitable but i have my doubts , as your solution
> doesn't handle two individuals that have evolved the same genome over
> (depth +1) time.
That is not so. I use fitness based tournament selection typically, lineage
selection is for supporting different species. There typically exists
tournament selection also/or fitness is used in selection function
offcourse, if I am not investigating just simulation behaviour.

> A similarity based method will handle this.
It is just not enought good my opinion. And needs some coding depending
genome type.
I have used that (was it DifferenceSelection method in
sourceforge/projetcs/narugo), and compared it success with pure tournament
selection. Found no big difference, propably because target problem was TSP
and difference was badly calculated.

t. Harri



.

Relevant Pages

  • Re: Lineage selection
    ... >> individuals and compex comparisons with structures, as Java searching is ... And this works for any genome structures same way. ... > every element in one array finding it another so you end up ... I use fitness based tournament selection typically, ...
    (comp.ai.genetic)
  • Re: Evolution increases the computational ability of organisms.
    ... >> Fluctuating selection would satisfy Gould equally well. ... >> Probably the major theme is that complexity follows a random walk ... You were trying to justify a claim that genome length increases through ...
    (talk.origins)
  • Re: DNA and Information
    ... That natural selection can increase the information ... content of a population genome (which is a slightly different thing ... the entropy of that message source both before and after selection. ... sequences of random letters into lines from Shakespeare doesn't strike ...
    (talk.origins)
  • Re: Non-random mutation of the genome from apes to humans
    ... If portions of the genome can become fixed ... Of course I understand how natural selection works. ... onto but the aids virus with already existing mutations can survive to ... genetic differences between humans and chimps. ...
    (talk.origins)
  • Re: DNA and Information
    ... That natural selection can increase the information ... content of a population genome (which is a slightly different thing ... I would say rather that mutation creates ... the entropy of that message source both before and after selection. ...
    (talk.origins)