I am using a cooperative coevolutionary genetic algorithm, the algorithm is given in A cooperative coevolutionary approach to function optimization by DeJong et. al. (1994), for optimizing a 62 dimensional function. One difference being that I am minimizing instead of maximizing. I am also using roulette selection, however since I need to minimize, the fitness is defined as the (maximum_fitness - actual_function_value).

I tried running the algorithm with 10, 20, 50 and 100 individuals, each about 5 runs. The one with 20 individuals had the lowest value of convergence while the others converged at what seems to be a higher local minima.

As far as I know, the more the individuals, the higher is the chance of escaping local minima. Is there any reason why that is not true for my case?

  • 2
    $\begingroup$ Maybe this question can be answered from general principles but it looks an awful lot like you're asking us to debug code or an algorithm that we haven't even seen, which seems pretty impossible, to me. $\endgroup$ – David Richerby Jul 21 '16 at 16:56
  • $\begingroup$ Thank you for your comment David Richerby. Sorry, I see that providing a solution for this case specifically is quite difficult. In general (for a simple GA) is there any concept which might explain why increasing the number of individuals is disadvantageous? $\endgroup$ – Leela Prabhu Jul 21 '16 at 17:44
  • 1
    $\begingroup$ Welcome to CS.SE! I doubt that this question is answerable from the information you've provided us; I suspect you'd need to provide a lot more information. Have you tried re-running the experiment and seeing if the effect how consistent the effect is? Did you run them all for the same number of iterations? Also, we expect references to fulfill the minimal scholarly requirements and be as robust over time as possible. Please take some time to improve your post in this regard. We have collected some advice here. Thank you! $\endgroup$ – D.W. Jul 21 '16 at 17:49
  • $\begingroup$ D.W.♦ yes I ran each experiment at least 5 times. I ran them till they met the stopping criteria (not more than 1% change for 100 iterations). So no fixed number of iterations. Ok, I'll improve the post. $\endgroup$ – Leela Prabhu Jul 21 '16 at 18:11
  • $\begingroup$ Thanks for your comment. I don't feel like this gives me any sense of how consistent or reliable the effect is. It might help to show actual data in the post: e.g., for each number of individuals, show the results of the 5 runs (number of iterations until convergence and value of the objective function after convergence, for each run). At least then we could test whether the difference you are measuring is statistically significant. (Have you tried to do that already?) $\endgroup$ – D.W. Jul 22 '16 at 1:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.