From news-rocq.inria.fr!jussieu.fr!oleane!news.maxwell.syr.edu!cs.utexas.edu!news.unt.edu!news Tue May 11 15:47:53 1999 Article: 10258 of rec.games.corewar Path: news-rocq.inria.fr!jussieu.fr!oleane!news.maxwell.syr.edu!cs.utexas.edu!news.unt.edu!news From: "Deathlok the Demolisher" Newsgroups: rec.games.corewar Subject: Re: Genetic Algorithms Date: Mon, 10 May 1999 05:13:53 -0500 Organization: University of North Texas Lines: 85 Message-ID: <7h6bd1$aao@hermes.acs.unt.edu> References: <7h29ra$cck@hermes.acs.unt.edu> <37359564.5308@dial.pipex.com> NNTP-Posting-Host: pras85.local.premium.dialup.unt.edu X-Priority: 3 X-MSMail-Priority: Normal X-Newsreader: Microsoft Outlook Express 5.00.2014.211 X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2014.211 Robert Macrae wrote in message news:37359564.5308@dial.pipex.com... > Deathlok the Demolisher wrote: > > > I know this isn't particularly useful, since it will > > evolve a warrior designed _only_ to fight a single > > enemy, but when I speed up the emulation a bit, > > Good approach. It is always best to tackle some easy > problems first because the nature of the solution helps > explain the things you hit on more complex cases. > What kind of strategies are teh dwarves using? Why are > the 2:1 winners better than your starters? Just as an example, here is the initial dwarf program. I matched this against the Cannonade program by Paul Kline, for about 3 hours, now (the program is still running) the current "best" program the genetic manipulations have been able to come up with is seen below. While _far_ from perfect, it has been able to increase its record from an almost "perfect" streak of losses to a few wins and more draws than losses... While not good enough to be considered a success, the longer the program is allowed to run, the better (in theory) redcode warriors should be generated. However, at the moment I'm being a bit limited by the slowness of my emulator...so I'm unable to run the large populations needed in order to create genetic diversity...currently I'm running with only 10 genetic warriors in the population...and even that few is quite slow.... :\ pmars output from 1000 rounds: Dwarf by Anonymous scores 1 Cannonade by P.Kline, Paul Kline, pk6811s@acad.drake.edu scores 2998 Results: 0 999 1 ;redcode ;name Dwarf JMP START bomb DAT -1 START MOV -1,-1 SUB #1,-1 JMP START END pmars output from 1000 rounds: Genetic by Anonymous scores 596 Cannonade by P.Kline, Paul Kline, pk6811s@acad.drake.edu scores 1835 Results: 9 422 569 ;redcode ;name Genetic SLT < -2632, < -66 MUL @ -1405, > 746 CMP @ -855, < -3284 DAT < 410, > -3684 SPL @ -2853, $ -3128 SLT $ -1843, $ -1341 ADD # -3251, < 722 MOV $ 3276, $ 141 ADD @ -1753, < 3971 SPL @ -2853, $ -3128 SLT $ -1843, $ -1341 ADD # -3251, < 726 MOV $ 3276, $ 141 SPL # -3307, < 1493 ADD < -1134, < -1330 ADD < -766, < -1330 DIV # 1386, < 3042 SUB # 37, $ -1 JMP $ -2, $ -1746 (before anyone says anything, I know that the program is basically generating random garbage....but it'll be interesting to see if it can come up with a version of the dwarf program that can at least match cannonade...well...maybe I'm the only one who is interested... :P ) (And just as a note....the random garbage _has_ been getting progressively better scores against cannonade as time goes on... so it must be doing something right) Deathy