Author: Thorsten Czub
Date: 10:15:46 08/27/02
Go up one level in this thread
On August 27, 2002 at 12:21:36, Uri Blass wrote: >Why not develop wipe out everything style. >This can save a lot of developing work;-( > >Uri :-))) the first thing i do begin is to play against the strongest program. when i have a styling that gets more than 50% against the leader of the ssdf list, i do test "weaker" programs. this method seems to be very logical IMO. why should i test against xyz first when fritz leads the ssdf-list ? i do understand that chessfun has fun ... but i have fun too. its not very easy in our days to make a program beat fritz :-)) it was much easier when fritz was more stupid. today it needs some fine tuning :-)) and a program that has the potential to do so. The new rebel XP HAS this potential since ed has programmed some nice new algorithms that evaluate different than century4 does. see http://members.aol.com/mclanecxantia/myhomepage/rebelxp.html and watch out for games by center style outpost style and ks-edx2 style. its all within the games. you only have to replay them. these new knowledge stuff needs of course fine tuning, this is almost always the case when new ideas get into a program, that you have to tune on the right weigthing. this is a process than can take much time. the new algorithms disturbs the evaluation function first. you have to find the right balance that other values are not overwritten. this can be done by styling. i think any chess programmer has to do so. and knows what i mean. the right weigthings of evaluation function terms/parts. when you think you have the "right" ones, you can prove them with playing bean-counting autoplayer games to get enough statistical data that proves your tuning.
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.