Re: Selection Functions

[Heinzelmann Matthias]

Michael, why use individual modelling when what you really do is use an
algorythm that approaches the problem from the macro-level. Why not let
individuals meet and then let them decide to breed according to some
'attractiveness'. Seems to be more bottom-up.



On Fri, 5 May 2000, Michael Stillwell wrote:

> I'm working on a GA-like system in which the probability that a bug is
> chosen to breed is related to its fitness.  It appears that the
> traditional way to do this is to use "roulette wheel selection"--each
> bug gets a slice of the wheel in proportion to its fitness, the wheel
> is spun, and the bug associated with the point where the wheel ended
> up is gets to breed.
> 
> Do other "selection functions" exist?  My problem with roulette wheel
> selection is that it offers no real control over the distribution of
> the expected number of offspring.  For example, if all the bugs (or
> "solutions", etc.) are *equally* *fit*, then there is an e^-1 chance
> (approx 36%) that a bug will not be selected at all.  This doesn't
> seem to mirror what happens in nature, where one might expect each bug
> to have one child (or close to it).
> 
> (I am aware of one other "selection function"--Goldberg's "Genetic
> Algorithms in Search, Optimisation and Machine Learning" describes an
> "expected value model" that effectively reduces the size of the slice
> granted to a bug every time it is selected for reproduction, but I
> would be interested to hear of others.)
> 
> 
> 
> 
> 
> 
> Michael
> 
> 
>                   ==================================
>    Swarm-Modelling is for discussion of Simulation and Modelling techniques
>    esp. using Swarm.  For list administration needs (esp. [un]subscribing),
>    please send a message to <address@hidden> with "help" in the
>    body of the message.
>                   ==================================
> 



                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================