rasenqt
commented
11 months ago Thank you for the review :) . About the results, using hamming distance or entropy is very computationally expensive (for example hamming distance O(population^2), because of that i didn't try to find best results due to my pc limitation and i tried only to figure out if others metric could works on.
Hi Michelangelo I am writing you this review hoping you will enjoy it.
First of all I thank you for the readme you wrote which allowed me to understand your idea. Next time I suggest you to put markdown comments also before the various code sections so that it will be more understandable.
Your idea of using metrics other than fitness seems interesting and certainly from the results you have shown it allows you to reduce the number of fitness calls while still getting good results. I am not entirely sure that using other metrics besides fitness is "standard" procedure for an EA but after all this seems to work. The results obtained for the vanilla verison seem a bit low to me, I think the reason is that you imposed a fixed number of generations rather than letting the algorithm go. You could set an infinite loop that interrupts if the fitness value reaches 1 or if you don't notice any substantial improvement in the last x generations, this certainly increases the number of fitness calls but may also increase the result. Finally, I suggest you introduce some graphs to show the learning of the algorithm.