begin quoting Tracy R Reed as of Mon, Jun 06, 2005 at 03:37:13PM +0700: > Stewart Stremler wrote: > > I've been messing about with GAs too much lately. Lots of local > > minima blocking the preferred solution because to get there from > > here a little 'slop' is needed now an again. > > Now THIS is something worthy of kplug-list and I would like to hear more > about it. I have been intrigued by GA's ever since I read Steven Levy's > book "Artificial Intelligence" back in 93 or so.
"Artificial Intelligence" or "Artificial Life"? (No, I haven't checked...) > At one point I was > really wanted to use GA to write an optimal Core Wars program and win > the King Of The Hill competition. But since the Core Wars machine did > not actually have a real bytecode I could never really figure out a good > way to implement a GA using the text based assembler language of Core > Wars with my minimal knowledge. GAs are generally dial-twiddlers; GPs (as I understand it) are program-builders. Which one were you thinking of? > Also any calculation I did showed that I > did not have nearly enough computing power at the time to do anything > useful. You'd've wanted to build your beasty outside of the environment, not train' 'em up inside the environment. Core Wars had the constraint that cloned processes share their time-slice, so a program that tried various versions _in_ the game would run really slow. > Can you tell us what you are using GA's for and how it works? Any good > references on the subject? Non-linear scheduling with arbitrary constraints. And I'm mostly a sounding board, although I rewrote the GA into a framework because the previous codebase was a mass of spaghetti code, and changing something took a week, so there was no way to quickly see "if I tweak X, would that help?". (It's not suprising that just about everyone builds a new one or copies-and-modifies another GA, because the pressures on making "just a little change" are high. Plus, it's mostly Ph.Ds who are working on such things, and as a general rule, the amazing thing about Ph.D code is that it works at all.) A co-worker has a pretty good book on the subject, when he gets back into town I'll see if I can get a title/author for you. And isn't this a subject for LPSG? :) -Stewart "I should take an older version of the code and relese it." Stremler
pgpc1nw41OVSg.pgp
Description: PGP signature
-- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
