Re: A new algorithm: Explore


I don't know if those randomers are the best idea? Seems like alot of
computational time effort to evalute individuals that arn't likely to
get anywhere near a good solution, also wouldn't this just be
n-dimentional brownian motion? Purhaps a better solution would be to
have an anti clustering coefficent. Something like if it gets to
croweded around a given solution some of the population shoots of in
various directions away from the central cluster.

""
Swapping: It is an operation which will interchange the positions of
two individuals without changing their group.
"""

This seems like an odd funtion, it seems like you will have some folks
walking in brownian circles no where near the optimal, then you will
swith them out with people who will charger towards the optima, and
the other folk will just walk around in very near (sub)optima and
completly ignore gradient. I don't see how this will help?



Provide a graph of the maxfitness as a function of fitness evaluation,
for various ratios of randomers to target. Similarly for the
introduction and removal of the "swaping" algorithm, but provided the
meanfitnees, as well as the max.


Cheers
- Haz


--
I run faster with a knife, everyone runs faster with a knife.

.

Relevant Pages

  • Re: A new algorithm: Explore
    ... I don't know that the randomers would really do that. ... > mean when you say "maxfitness as a function evaluation". ... The x-axis should be the number of times you have called the "fitness ... I run faster with a knife, everyone runs faster with a knife. ...
    (comp.ai.genetic)
  • Re: A new algorithm: Explore
    ... parmeters of the randomers. ... As i don't think they are helping, ... I run faster with a knife, everyone runs faster with a knife. ...
    (comp.ai.genetic)