Evolutionary algorithms like genetic algorithms (GAs) are typically run multiple times and the outputted results are averaged across successive runs.

However, in the case of long-runtime algorithms (e.g., due to large population sizes or algorithm complexity), is there any justification for running GAs (and the like) only once? Clearly, statistical evaluation of variability is not possible in such a case.

I've not been able to find anything in the literature on the subject, so am asking here to gain some insight from the CS community.

  • $\begingroup$ For the 2nd paragraph: isn't what you say a valid justification? That is, if I can only afford to run my algorithm once, I just run it once? Of course, you might be unable to argue reliably or convincingly something from a single run. $\endgroup$ – Juho Aug 9 '19 at 18:51
  • $\begingroup$ Yes, that's true I suppose. From the literature, I think too, a way around the issue is to simply lower the population size. Runtime will be faster of course, but solutions may be suboptimal (trapped in local optima). Lowering the population size will enable statistical testing however since the algorithm can then be run multiple times. $\endgroup$ – compbiostats Aug 12 '19 at 14:06

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.