Re: modifications of runk selection method
- From: Lionel B <me@xxxxxxxxxxx>
- Date: Tue, 29 Aug 2006 10:17:33 GMT
Kent Paul Dolan wrote:
Natalya wrote:
But have any algorithm's modifications/enhancements of "rank-proportional" selection been worked out?
Yes.
As a population converges to the/an optimum, the fitness of the
population members becomes very nearly identical, and so "rank
proportional" selection becomes nearly useless, the roulette wheel
slots are nearly identical in width..
No, that's for *fitness*-proportional selection. In rank-based roulette
wheel selection - what I called "rank proportional" selection - the
roulette wheel slots are proportional to rank, which conventionally runs
from 1 to population size (so the population member with smallest
fitness gets a slot of size 1, the next-smallest in fitness a slot of
size 2 and so on). Thus some selection pressure is always maintained
even if the spread of fitness is small - although this does make less
sense when there are many identical copies of genotypes within a
population, in which case ranking becomes somewhat arbitrary.
If instead, the fitness range from the best to worst element, plus
some token amount to prevent the least fit individual having zero
selection probability, becomes the "proportional" part, with the rest
of the fitness subtracted away, then the differences among the
population are magnified and selection pressure to improve fitness
remains high.
/.../
I hope that's comprehensible, I'm a bit too tired to be writing technical prose..
Maybe yes ;-) What you've described is indeed a well-known scheme for
maintaining selection pressure within a population with a small spread
of fitness values, but rank doesn't come into it (at least not
explicitly; you would, however, expect there to be some correlation
between rank and actual fitness).
--
Lionel B
.
- References:
- modifications of runk selection method
- From: Natalya
- Re: modifications of runk selection method
- From: iAgent
- Re: modifications of runk selection method
- From: Natalya
- Re: modifications of runk selection method
- From: Thorsten Kiefer
- Re: modifications of runk selection method
- From: Lionel B
- Re: modifications of runk selection method
- From: Natalya
- Re: modifications of runk selection method
- From: Kent Paul Dolan
- modifications of runk selection method
- Prev by Date: Re: modifications of runk selection method
- Next by Date: Re: modifications of runk selection method
- Previous by thread: Re: modifications of runk selection method
- Next by thread: Re: modifications of runk selection method
- Index(es):
Relevant Pages
|
|
