Author: Álvaro Begué
Date: 13:45:00 12/12/05
Go up one level in this thread
On December 12, 2005 at 14:52:05, Dann Corbit wrote: >On December 12, 2005 at 14:39:20, Álvaro Begué wrote: > >>On December 12, 2005 at 14:07:17, Dann Corbit wrote: >> >>>On December 12, 2005 at 14:01:45, James Constance wrote: >>> >>>>When a tablebase says that a position is a win in more than 50 moves does it >>>>take into a account the 50 move rule i.e. that a piece has been taken or a pawn >>>>moved? >>> >>>No. But it's trivial to count it yourself. >> >>It's not that trivial. If your database gives distance-to-mate values, it is in >>possible in principle that it will spend over 50 moves without making a pawn >>move or a capture, which results in a draw. A good database should know about >>distance to an irreversible winning move, and consider anything with a distance >>over 50 to be a draw. Then you can try to win faster if you want. I don't know >>if any tablebases currently available do this correctly, but in practice it will >>probably not matter. > >It will multiply the size of the database by a very large number. >You follow your projected pv to the end result. If you do not win, but only >draw, you were not going to win anyway. So there is not value at all to adding >the 50 move rule and a huge cost associated with it. First of all, the size of the database wouldn't increase at all. You still store one value per position, with distances to mate if you want. The difference is that positions that would have gotten a distance to irreverible move larger than 50 get a score of "draw". You may actually draw a won game by using the wrong database. Imagine a position with 7 pieces where by exchanging a rook for a pawn you get to an endgame that can be won in 126 moves, but which requires more than 50 moves without an irreversible move. Your engine would exchange the pawn thinking that it's winning, and then the game would be a draw. > >e.p. is another matter and one that could have more importance, I think. >It's not inconceivable to let a pawn sit for almost the full game (especially a >king shelter pawn). e.p. is already considered in Nalimov's databases, and I guess in others as well.
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.