Author: Dann Corbit
Date: 17:40:38 08/29/05
Go up one level in this thread
On August 29, 2005 at 03:46:33, Majd Al-Ansari wrote: >On August 28, 2005 at 09:27:21, Roy Brunjes wrote: > >>Almost all have seen how strong Fruit 2.1 is by now. It plays very well even in >>the endgame. And yet, it does not use tablebases. It seems about 2 years ago or >>so that there was beginning to develop a community consensus that tablebases >>were essential to strong endgame play by an engine. And yet, here is Fruit >>winning endgames vs strong engines such as Shredder (and others) with only the >>knowledge built into its relatively small executable. >> >>Does anyone else find this remarkable? I hope this has not been discussed >>already and my searches through the archives were the wrong ones. >> >>Roy > >While Fruit plays endings very well, it definetely can use EGTB. I have several >games which were completely won but Fruit was unable to get the point. In one >case I remember Fruit being unable to win a qr vs r+p endgames. When I put the >same game on shredder it immediately saw the mate while in Fruit's case it was >stuck with a draw by 50 move rule. So EGTB do matter. Unless the probing of EGTB tables causes the search to slow down so much that it plays less well in general. There has never been an experiment that shows any strength gained from EGTB tables for any program. There have been experiments that showed strength gained from bitbase files. There have been three EGTB experiments, with lots and lots of games, that showed EGTB files to neither add nor reduce strength. In other words, the value of the gained knowledge from the lookup is about the same as the value lost from the search strength due to the time wasted by EGTB lookup. I expect that very fast disk subsystems (or better yet ramdisk) will change that in the favor of tablebase files, but that is speculation on my part.
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.