SelectionStrategies that can select the same element more than once

[Alfonso Garcia]

Again, question to see if someone can shed some light on this: I'm trying different selection strategies and have run into cases where the same element is selected more than once. Is this expected, or I'm doing something wrong, or have I just hit some corner case?

Thanks for the help.

[Klaas Hölscher]

Hey Alfonso,

This is expected behaviour. For the rouletteewheel each element is chosen with propability proportional to fitness, so good  members may be selected more than once - maybe even the same for every roll oft the wheel, but that would be very unlikely.
I cant remember how the other selectors work, but thats enough to understand the principle.

Regards,
Klaas

Am 20.10.2014 20:32 schrieb "Alfonso Garcia" <barb...@gmail.com>:

Again, question to see if someone can shed some light on this: I'm trying different selection strategies and have run into cases where the same element is selected more than once. Is this expected, or I'm doing something wrong, or have I just hit some corner case?

Thanks for the help.

--
You received this message because you are subscribed to the Google Groups "Watchmaker Framework for Evolutionary Computation" group.
To unsubscribe from this group and stop receiving emails from it, send an email to watchmaker+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Alfonso Garcia]

Thanks for your response, Klaas.

I think that I painted myself in such a corner case when I saw all the population being made up of the same element :-) From an evolutionary perspective (and I'm saying this from intuition, not because I have a formal proof) it does make sense to bias the selection preference towards the best rated members, but selecting the same element more than once decreases diversity in the population. Which could be bad because crossover with itself cannot yield anything interesting, and once you reach that state, only replacement and mutation can improve the score.

But I can see also that, as you say, there could be models where this is actually desired and expected, so I'll take a look at a custom selection strategy that avoids repeated selection of the same element for my problem domain.

Regards,

Alfonso