Re: Genetic Algorithm

On 13 Oct 2003 02:18:37 -0700, [EMAIL PROTECTED] (May) wrote:

>I want to know how rank selection work in detail.
>
>And I have also read this linear function for rank selection.
>   
>     prob(rank) = q - (rank -1)^r
>
>where q = a user defined parameter ( q varies between 1/pop_size and
>2/pop_size)
>
>This function returns the probability of an individual ranked in
>position rank
>(rank = 1 means the best individual, rank = pop_size the worst one) to
>be selected.
>
>e.g; pop_size=100 and q=0.015 then r=q/(pop_size-1) =0.00015151515
>
>and prob(1)=0.015
>    prob(2)=0.0148484848
>    :
>    prob(100)=0.00000000000000000051
>
>    I have got these probabilities and how can I use these
>probabilities to select chromosomes.
>    I also want to use this rank selection method in Evolutionary
>Planner/Navigator. Please suggest me.
>
>Thanks in advance.
>
>With Regards,
>May

The GA package, Generator,
(http://www.nli-ltd.com/products/genetic_algorithms/demos.htm, uses a
simple linear rank-based mate selection scheme, P= rank/(n^2-n), where
rank is the rank of a potential mate in the population from which the
mate-seeker has been removed.  

Each individual in the population gets to find a mate, but most likely
will mate with a high-fitness individual.  The higher an individual's
fitness, the higher the number of matings it gets to participate in.

It generally works very well.  In a large population it would make
sense to de-linearize it so that the high-ranking individuals get a
greater share of mates.  

Steve McGrew

(out of N-1; best has rank of N; worst has rank of 1)