Author: Vasik Rajlich
Date: 08:01:36 02/06/06
Go up one level in this thread
On February 05, 2006 at 14:03:49, Ray Banks wrote: >CCRL (Computer Chess Rating Lists) > >We are a group of testers formed with the goal of producing further rating lists >that will go hand in hand with other established rating lists in showing the >relativity in engine strength. > >At present CCRL has the following testers: >Graham Banks, Ray Banks, Sarah Bird, Kirill Kryukov, Tom Logan, Andreas >Schwartmann, Charles Smith and Chris Taylor. > >We have concentrated at this stage on a 40 moves in 40 minutes repeating rating >list, calibrated to Athlon 64 3800+ hardware, but also intend to produce a blitz >list (40 moves in 8 minutes repeating) shortly. Future plans include the >possibilty of "endgame rating lists" at the same time controls based on play >proceding from a series of endgame positions. > >Each engine uses a generic opening book with an upper limit of 12 moves per >side. Ponder off, no learning, EGTBs, hash size 128mb or 256mb each (deep >engines get twice this size). For blitz it will be 64mb hash > >For our first 40/40 rating list, we took a list of 18 engines with the intention >of running 30 game matches (all play all) so that we could establish a base from >which to develop. > >We now present to you the first CCRL 40/40 rating lists at the following link: > >http://kd.lab.nig.ac.jp/chess/CCRL-4040/ > >There is a best versions list, an all versions list and a pure list that removes >rating distortions. > >There are also many stats tables available for your interest at the above link >including: >- pairwise results >- game length statistics >- engine correlation statistics (expected move table, evaluation difference >table, engine distance table). > >In addition we have opened a CCRL discussion board. You are all welcome to join >so that you can discuss our testing and make comments or suggestions on an >ongoing basis. > >http://kd.lab.nig.ac.jp/chess/discussion-board/viewforum.php?f=7 > >Please note though that general questions about engine testing, not directly >related to CCRL, are better posted in other discussion boards, for example WB >Forum or here. You will certainly reach more people and get better answers >there. > >Comments and questions are welcomed. There are some interesting statistics on this web site. For example, the eval correlations: http://kd.lab.nig.ac.jp/chess/CCRL-4040/eval-difference-table-best-limited.shtml Honestly, I always liked Junior's evaluation the best :) One odd thing is that the engine whose eval is furthest from Rybka is Fritz 9, and yet this is the engine against which Rybka ponder-hits the most. Vas
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.