Mu plus lambda and environment selection

[baldini...@gmail.com]

Hi there,

I am using deap as base library for building my own evolutionary strategy for unsupervised learning application.

In my previous experience with genetic optimization, I have never seen mu+lambda. For example, MATLAB evolutionary scheme uses selection operator for picking individuals to be mated and/or mutated that are going to compose the next population for generation i+1 adding also some elite individual.

In DEAP, I found interesting  mu plus lambda since does not need to set in advance the number of individual to be mutated, recombined or reproduced.

Since this method produces the next population thanks to the selection pressure applied by the selection procedure I wonder if it is necessary that the selection strategy is totally fitness based, that is totally deterministic based only on the rank of individuals build on the individual fitness values.

Thank you so much.

Luca

[François-Michel De Rainville]

Hey Luca,

Randomized selection that is not fully elitist has been found to help to exit local optimas and converge more steadily to global optimas.

You could try it on your own by using selBest and compare your final result.

Cheers,

--
You received this message because you are subscribed to the Google Groups "deap-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to deap-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/deap-users/aeef1a2b-86e8-4c00-9a45-c14e32ccfe4bn%40googlegroups.com.

[baldini...@gmail.com]

Thank you!