Author: Barry Culp
Date: 08:26:49 02/12/06
After running several engine matches using the 20 position endgame suite suggested by Sune Larsson (http://www.talkchess.com/forums/1/message.html?485148) I decided to concentrate on one position and run a double round robin tourney with 3 different EGTB setups : One tourney with no EGTB's, one with all the common 3-4-5 man EGTB's, and one with some 6 man EGTB's added to the 3-4-5. [D] 8/4k3/p1B4p/2K5/1P4bP/8/8/8 b - - 0 1 The comments that came with this position are : "Eliskases-Capablanca, Semmering 1937. Another classic ending. White has little trouble winning the a-pawn, but will this be enough to win the game? White is handicapped by the fact that his bishop is the wrong colour for the h-pawn, so even winning Black's bishop may not be sufficient. In the game White won, but various analysts who have examined this ending in detail have come to the conclusion that Black should draw with accurate play. But if even the great Capablanca could not find the draw, can a machine? " The 3 double round robin tourneys consisted of 20 games each, 5 engines: Rybka beta 13b-x64, Fritz 9, Fruit 2.2.1, Hiarcs 10, and Deep Shredder 9.12-x64, all engines that support EGTB's. Time control of 15 min + 0 was used. In each tourney, every engine played the above position 8 times, 4 times as black and 4 as white. Results comparing white v black winning percentage in the 3 different tourneys : Tourney #1 : no EGTB's White - +8/-0/=12 70.00% 14/20 Black - +0/-8/=12 30.00% 6/20 Tourney #2 : 3-4-5 EGTB's White - +10/-0/=10 75.00% 15/20 Black - +0/-10/=10 25.00% 5/20 Tourney #3 : 3-4-5-6 EGTB's White - +15/-0/=5 87.50% 17.5/20 Black - +0/-15/=5 12.50% 2.5/20 White's winning percentage increased as EGTB's were added. Specific EGTB's that probably played a role are 4 man KBp.K, 5 man KBp.KB, and 6 man KBp.KBp + KBpp.KB . Can one conclude from the small number of games (60) that EGTB's made a difference in engine play for this particular position ? Barry ***************************************************************************** Engines - Rybka 1.01 Beta 13b-x64 - defaults Fritz 9 - defaults Hiarcs 10 - position learning turned off Fruit 2.2.1 - defaults Deep Shredder 9.12-x64 - position learning off, 1 thread Time control - 15 min + 0 Book/DB - Sune Larsson 20 position endgame suite (position #8) System - Windows x64, 3gb RAM, dual AMD Opteron @2.4ghz EGTB- 3 different setups GUI - Fritz 9 (date Dec 20) Hash - 512mb each engine Draw - Late, Resign - Late Permanent brain - On - (1 cpu per engine)
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.