From news-rocq.inria.fr!jussieu.fr!oleane!plug.news.pipex.net!pipex!tube.news.pipex.net!pipex!lade.news.pipex.net!pipex!news.be.innet.net!INbe.net!news.nl.innet.net!INnl.net!hunter.premier.net!newsfeed.internetmci.com!solaris.cc.vt.edu!news.new-york.net Wed Jun 19 11:09:12 1996 Article: 5430 of rec.games.corewar Path: news-rocq.inria.fr!jussieu.fr!oleane!plug.news.pipex.net!pipex!tube.news.pipex.net!pipex!lade.news.pipex.net!pipex!news.be.innet.net!INbe.net!news.nl.innet.net!INnl.net!hunter.premier.net!newsfeed.internetmci.com!solaris.cc.vt.edu!news.new-york.net!news.stormking.com!iltzu@sci.fi From: iltzu@sci.fi (Ilmari Karonen) Newsgroups: rec.games.corewar Subject: 3-clear 0.5 Date: 18 Jun 1996 23:53:26 -0400 Organization: Storm King Ind. Inc. Lines: 74 Sender: server@news.stormking.com Distribution: world Message-ID: <19199643.Amiga@sci.fi> Reply-To: iltzu@sci.fi NNTP-Posting-Host: valhalla.stormking.com Originator: corewar-l@stormking.com X-Mailer: Air Mail V2.0 (Unregistered) -- Amiga Mailer by Danny Y. Wong Well, this (v0.3 actually) is the first of my warriors to establish a position on the beginner hill. Any ideas for improvement are welcome. Actually, I don't think there's much room for improvement without turning it into a completely different warrior. The original clear was a simplified version of the clear in Symmetry 0.4, my first warrior ever. (I think it reached the mighty age of 3 on the b hill.. =) I first tried making 3-clear use a multipass version of d-clear's clear, ie. SPL mixed with djn stream. This is what Symmetry 5.0 still uses. Unfortunately I found that no matter how many DAT passes I made, after 80000 cycles there were still SPL's present because of the djn, giving me too many draws. Changing the spl to go backwards seemed to improve the score greatly against everything else except it's own previous versions. =) I probably sent 3-clear to the hill at a very bad time; All the papers had just fallen off the hill, everybody was doing coreclears inspired by d-clear, and the top of the b hill was (still is) filled by big mean scanners/p-spacers. :( If the papers ever come back, though, we might even see 3-clear on the big hill someday. (wishful thinking) The other problem is that clears are strongly affected by starting positions and boot distances. (v0.4 actually got 4 more points on the hill than v0.5 just by luck) The current version still seems to bounce around the 10th position on the b hill, killing papers and all those fancy warriors (like Impfinity!) that just can't match a simple brute-force attack. =) OK, thanks for reading all that crap. Here's the program. Any newbies (and others too) are free to use it as a p-component etc.. ;redcode-b verbose ;name 3-clear 0.6 ;author Ilmari Karonen ;date June 13, 1996 ;strategy Cclr + djn (reverse djn version) ;strategy Simple and easy to adapt to different situations. ;comments 'd-clear' converted into a multipass cclr! ;history 0.1 - Works.. (SPL/SPL/DAT[/DAT]...) ;history 0.2 - Starts spl again after SPL/SPL/DAT/DAT. ;history 0.3 - Fixed a VERY stupid bug! ;history 0.4 - Test: splits less, clears faster? ;history 0.5 - Back to SPL/SPL/SPL/DAT[/DAT]... ;history 0.6 - Switched to backward djn.. hmm? Added more space. ;planar clear, stun ;size 3 ;speed ~1c ;assert 1 == 1 org split tail equ (kill+1) toe equ (head+MAXLENGTH) space equ (MINDISTANCE) head bomb dat #stun, #toe+space dat 0, 0 stun spl #start-bomb, #tail-bomb start spl #split-bomb, #tail-bomb split spl #kill-bomb, #tail-bomb loop mov.i *bomb, >bomb mov.i *bomb, >bomb djn.f loop,