Re: A new algorithm: Explore



"Einstein" <paras1987@xxxxxxxxx> writes:

> Check it out at:
> http://www.paraschopra.com/articles/explore/data/ratio.jpg
>
> I don't know why, changing the ratio does not change up the time
> taken to find a solution significantly. Is it because my choice of
> fitness function wrong??
>
> Atleast there must be some variance if a change no. of randomers/no
> of targeters..
>
> What do you suggest??
>

Very clean, nice and telling diagram, (not a big fan of excel graphics
but thats beside the point). I suspected that this was the case and
that the randomers were not helping the convergence in any signifigent
way. I'm not sure really how to explain it at the moment, but if you
think of how the randomers are simply walking in circles untill they
are called upon to be reactivated. What in effect they are doing as
far as i can see is acting similar to what happens when you lower the
probability of crossover.

At any rate i think you should explore other options, such as sub
populations to prevent premature convergence.

Cheers
- Haz

--
There are three types of posters to alt.binaries.*
- Binary Content Posters
- Those who thank Binary Content Posters
- Spamers
.