Author: Robert Hyatt
Date: 17:23:48 08/04/99
Go up one level in this thread
On August 04, 1999 at 17:00:03, Ricardo Gibert wrote: >On August 04, 1999 at 16:34:40, Robert Hyatt wrote: > >>On August 04, 1999 at 16:24:58, Ricardo Gibert wrote: >> >>>I think I figured out what the problem is. >>> >>>Consider W: Kf2, Re2, Pa2; B: Kh2. >>> >>>Distance to mate will return: >>>1 Re3 Kh1 2 Rh3# >>> >>>Distance to mate or winning pawn move or winning capture will return: >>>1 a4 Kh3 2 a5 Kg4, etc. >>> >>>The pawn move is a 1 mover that leads to a winning position, while the rook move >>>is a 2 mover to lead to mate. To avoid this idiotic behavior, distance to mate >>>must be used at the expense of extremely rare positions that will instead of win >>>due to the 50 move rule. The rest of the program must deal with 3-fold >>>repetition (easy) and the 50 move rule as best as it can. >> >> >>you say "deal repetition (easy)". I don't see how it is possible, much less >>easy. Unless the program plays the whole game from start to finish. Because >>tablebases are positions that say "mate in N moves from this position" and that >>is a reference to a path of moves that if followed, end up with mate in the >>number of moves given... But if someone has played some non-optimal moves, >>this won't work, because between "here" and "mate" lies a whole lot of other >>positions that have to be stepped over. And it is just possible that one of >>those positions has already been hit twice before the computer gets a chance to >>do the lookup. And I don't see _any_ way to repair this. > >Yes, I see what you mean now. You can deal with some cases, but not others. The >"winning" line can be quite long and a search can be prohibitive to avoid the >3-fold repetition. > right... this is a common problem... we have it in transposition/refutation (hash) tables for those of us that store mate scores there. Because a mate from the hash table might require you to search thru a 3-fold repetition to reach it, which you can't do but you can't see since the path from the root hash position to the final position is not stored in the entry (too large). >> >>of course, it isn't a problem if the computer plays from the beginning. But >>the lack of path information from the tablebase hit to the final tablebase >>mate position is a killer, otherwise...
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.