Author: James Swafford
Date: 12:26:23 07/14/05
Go up one level in this thread
On July 14, 2005 at 14:55:00, Mathieu Pagé wrote: >Hi, > >I was wondering do someone know of an engine that do some opponent modeling? Has >it ever been tried? Is there paper/documentation about such an experimentation ? > >I don't know to wich level it could be used, but it seem to me that a learning >strategie like the one used by crafty for it's opening book could be used on a >per oppenent basis. I think this could be proved really efficient against other >engines. > >In particular, such an engine could learn not to be trapped in closed position >(so called anti-chess) against some player, but not change is opening theory >against other opponents. It could even learn to close the position against >stronger computer opponents (Am I dreaming?). > >Opponent-modeling could also be applied to evaluation function weight tunning. >This, however, would require lots of games to be played. One could make it's >engine play against another engine X for some weeks and get an anti-engine_X >weight set. > >What are your comments? Is there some documentation avaible about >opponent-modeling in computer chess? > >Mathieu Pagé >mathieu.page@gmail.com I had thought to do something along those lines. About a year or year and a half ago I wrote a perl script that would log the ICC games for any account I told it to. I scheduled the script to run every half hour or so. When it logged in, it would pull a 'history' of every account. If there was a game in the history it hadn't logged yet, it would get the moves for that game and log them into a Postgres database. This worked pretty well, but I haven't run it in a long time, and I haven't done anthing with the data yet. -- James
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.