Author: blass uri
Date: 21:56:05 10/09/98
Go up one level in this thread
On October 09, 1998 at 22:54:49, Roberto Waldteufel wrote: > >On October 09, 1998 at 18:46:36, David Eppstein wrote: > >>On October 09, 1998 at 15:24:26, Robert Hyatt wrote: >>>This doesn't matter however, because *every* possible position must be accounted >>>for, with an exact distance to mate for the side on move with that specific >>>piece configuration. So they *all* have to be computed to build the next one >>>after them... >> >>You don't need distance to mate if you are searching from a non-tablebase >>position trying to reach a tablebase position (and aren't worried about the 50 >>move rule, but you must not be since you're using distance to mate). >> >>If the actual game reaches a won tablebase position, you need some way to force >>progress, but it doesn't have to be distance to mate. If you can always search >>deeply enough to find a conversion (capture or pawn move), you can use distance >>to conversion, and only store win/loss/draw in the tablebase. In any KXP-KP or >>KX-KPP endgame, searching deeply enough to find a conversion should be easy >>(there are fewer than a million distinct positions in which at most one pawn has >>moved, so you can load just that part of the tablebase into memory and use the >>hashtable to do the search quickly no matter how deep it is). >> >>But I agree with your main point, that the heuristics suggested by the poster >>you were responding to aren't good enough -- the information needs to be exact, >>and you need to compute lots of other tablebases before you can think about >>KPP-KP. > >Even if the search were too deep to be feasible, eg an ending like KBBKN, it is >still possible to reduce memory access requirements during the search by storing >only win/loss/draw information, if we maintain a separate tablebase (not used in >the search) which simply contains the best move You need memory to store the best move for example in KBBKN for the stronger side the maximal number of legal moves is 8+13+13=34 legal moves and you need 6 bits for a move so I think you do not save memory by this. If you have a good order of moves and always 1 of the first 32 moves is best you can need only 5 bits If you store distance to conversion with my idea you need only 5 bits for 1-2,3-4,...49-50,draw,loss You need to do a search but I do not think there is a problem with small search. Uri
This page took 0.01 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.