Author: Bruce Moreland
Date: 09:56:28 03/22/99
Go up one level in this thread
On March 22, 1999 at 12:41:44, James T. Walker wrote: >Last night I ran the 2nd 100 games at game/5 minutes between Crafty 16.5 with >and without tabelbases in an attempt to determine the value of the tabelbases. >It is almost like an instant replay. 2nd 100 games score: > >First 50 Crafty w/tb +16 =20 -14 = 26 to 24 >Second 50 "" " +13 =21 -16 = 23.5 to 26.5 >Second 100 games " Total = 49.5 to 50.5 >Total for 200 games Crafty with tb =100.5 to 99.5 > >I only have the 3/4 man tablebases and about 6 of the 5 man tabelbases. Most >notably absent is the KRPKR tablebases. I can't generate them on my computer >and AOL will not leave me alone long enough to download them. [:-) >So far it looks like the tablebases are useless !? I can't believe this though. > I really love to see a program announce "mate in 21" when playing 5 minute >blitz ! Even if that is the only value I get out of the tablebases I still >think they are worth while. I will play a couple hundred more but it looks like >the advantage is minimal without the KRPKR tablebase which I believe will yield >the most advantage. >Jim Walker No, what you are seeing here what life is like in statistical hell. The Elo number is going to be small, and in order to demonstrate it convincingly you would have to play a zillion games, it isn't worth it to even try. It is way easier to show that A is at least a little better than B, if in fact A is way better than B. But if A is really just a little better than B, it is very hard to demonstrate. You'd think that tables would make you better, because you evaluate better and sometimes you clearly shift the game a column or two in your favor. But they have drawbacks as well. You do disk accesses in the non-contentious parts of contentious 7-man endings, which slows you down so perhaps you lose before you make it to the 5-man case. Also, with some implementations you will avoid trading into difficult lost table endings, in favor of staying in easy lost non-table endings, for example, you would rather be in a dead-lost KRPP vs KR, rather than win the pawn and force your opponent to count down from mate in 50 in a hard KRP vs KR. So it's possible that they don't help. I think they do, but I think it would be hard to measure. bruce
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.