On Feb 21, 2014, at 9:08 PM, Victor Jatobá <[log in to unmask]> wrote:
> So, I stil have a question. Two individual are selected for crossover and mutation process, but is generate two copies and not the real individuals,
> after the processes spoke before, these two clones are added in the new Subpopulation and the two parents are mantained in the Subpopulation too (right?).
The parents won't go in the subpopulation unless you have some kind of elitism or ES. That's the standard approach for most EAs.
> If I have many Subpopulations (> 1000), in the beginner generations, the algorithm is very fast but after to more generations the algorithm is not too fast as before. Why does it occur? It seems that the subpopulations is accumulating the number of individuals. How the algorithm define the fitness value to kill the weakest individuals?
If your number of subpopulations is staying constant, and the number of individuals in a subpopulation is constant, and the size of an individual is constant, and your evaluation process is constant regardless of the individual, then ECJ should never slow down. This suggests to me that one of these is not true about your system. Are your individuals growing in size?