Author: Robert Hyatt
Date: 07:17:25 01/10/02
Go up one level in this thread
On January 10, 2002 at 05:45:18, Sune Fischer wrote: >On January 09, 2002 at 15:20:03, Robert Hyatt wrote: > >>>>I don't remember the positions, but Crafty once announced a mate in 103 on >>>>ICC in a KNN vs KP ending, and Ferret beat that with a mate in 104 announcement >>>>in the same kind of ending... >>> >>>I doubt if they can win against correct defence. >>>The position may be drawn by the 50 move rule and a draw like this already >>>happened in comp-comp games. >>> >>>Crafty should never announce mate in 103 based on existing tablebases and a >>>score of mate in 103 should be translated to some material advantage based on >>>the estimated probability to win. >>> >>>Uri >> >>Why? Just because some arbitrary rule says the game is a draw? Even if that >>rule has been eliminated and then reinstated a few years later? mate in 104 >>might be right again one day as fickle as FIDE is. :) > >The point is, that the programs shouldn't trust the tablebase score if the >distance to mate is above 50, then the score should be draw and not a mate >score. >We always assume that the opponent will make the best possible defense, don't >see why this should be any different. > >Exceptions: the distance to mate is 53 moves and the engine is playing a human, >then it is probably okay to assume the human will "blunder" by playing slightly >inaccurate, enough to mate within 50 moves. > >On the otherhand, if 99% of the mates in the tablebases are in less than 50 >moves then it might not be worht checking? > >-S. There are not many of those > 50 mates. I only recall having seen that one mate in 103 from Crafty in all the years it has been playing on a server, although I wouldn't begin to claim that there were no other such mates found as I don't look at that very closely... Another issue is that _many_ of the mates in > 50 are perfectly valid, particularly if there are 5 pieces on the board, as there is a trade or the win of a piece at some point which resets the counter. In the case of Crafty it may well escape some of those 50 move draws because I don't probe the tables at the first couple of plies, which will let it see the 50 move draw and (if possible) do something to prevent it from happening (pushing a pawn, trading a piece or whatever). Of course, there _will_ be those positions that it thinks it is winning, but it will end up drawing due to the 50 move rule. But so far they have been _very_ rare. Against humans, and that is the primary kind of chess I am interested in, this is not nearly such an issue. Against another computer with tables, both will be wrong as one will think it is winning, the other will think it is losing, when in fact it is a draw...
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.