Author: scott farrell
Date: 18:13:48 08/19/03
Go up one level in this thread
On August 17, 2003 at 13:37:20, Peter Skinner wrote: I guess it depends on how much your engine relies on the tb. My engine has trouble mating at blitz with KRK, and needs the tb (I have no specific knowledge for this, and the mate can be fairly deep as you know) - or even KQK sometimes. I use TBs accross the web to lokusoft, and now and again the connection doesnt work for several hours (damn dlink ADSL modem thinks loku is doing a DoS attack). I lose about 100-200 points on ICC at blitz, it starts to not only turn draws into losses, and wins into draws, even wins into losses. If the program is old enough (like crafty) - they had to mate with KRK long before TBs were around, so they have the knowledged coded - so the tb has less effect. I think many engines have lots of code to tell them when its a likely won/lost/drawn end game, my engine does not, so the egtb helps if it gets that far in the search. I guess I should really code more of this logic, but its not easy. Scott >Since my last inquiry of the importance of tablebases I have run various engines >on FICS over the last month, playing over 1100 games. > >In very few instances using tablebases did not change the outcome of a game ie. >a loss into a draw, a loss into a won game. In almost 96% of the games, the >outcome was pretty much decided by the time the 5 man tablebases were used. > >On FICS of course the time controls were of a fast nature, usually 5 3 blitz, or >30 10 "standard" time controls. > >The engines that I used in the test were Ruffian 1.0.5, Crafty 19.03, Chess >Tiger 15.0, and Junior 8. > >Out of all of these, Crafty seemed to do the best whether it had the tablebases >or not. Ruffian scored slightly under Crafty. Chess Tiger and Junior had very >mixed results. > >Now while the tablebases due in fact let the chess program gain the instant >moves needed in a fast time, I have found that without the access, the engine >usually comes up with the same move in 96% of the time. It also allows the >program to move instantly when it is searching the tablebases, thus in a short >time control this can be of service. > >I have read many tests where others are doing the same as I am, and I have come >to the conclusion that while tablebases certainly _can_ be benefitial, they for >the most part are not 100% required. > >If anyone would like the games, I would be happy to email them. > >Peter.
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.