Any computer scientists in here?

[Kylearan]

Hi,

sorry for answering so late; I was busy during the last few days.

Reworking it in my head, O(nlogn) is right. On very large sets, it would matter, but the overhead of the modivied trunc would probably be like half of log n on average anyway.

Indeed. Normally, population size is between 100 and 500, less than 1000 most of the time - so the improvement from O(nlogn) to O(n) doesn't really matter (apart from it being theoretical aesthetic...) . And in many cases, selection isn't the most time-consuming process anyway - but it's quite hard to discuss this if we don't know exactly what you are trying to solve with the GA. ;) In the few cases I programmed a GA, calculating the fitness function dominated selection time-wise anyway, so I found it better to optimize that instead. And if you expect your GA to be time-consuming, flash might not be the best choice of language here anyway... :P

I'll post a link if it ever turns out.

Please do so! I'm curious about anything GA related. :)


By the way, do you know the works of Karl Sims? He has done amazing things with GAs. He simulated a physical world and used a GA to evolve "creatures" that learned to move. His representation of genes is nothing short of amazing, and best of all, he animated the results in a video than can be downloaded from his website, along with the scientific papers he published about the subject. Well worth to check out for anyone interested in GAs!

If you're interested, look for the paper "Evolving Virtual Creatures" and the corresponding videos on his website

-Kylearan