Author: Robert Hyatt
Date: 14:46:57 06/30/04
Go up one level in this thread
On June 30, 2004 at 17:01:16, Uri Blass wrote: >On June 30, 2004 at 15:10:51, Robert Hyatt wrote: > >>On June 30, 2004 at 02:58:11, Mig Greengard wrote: >> >>>For storage space considerations, some impractical and unnecessary bases have >>>been left out of the DVDs. >> >> >>Again, kpppk is not that impractical. Particularly when all 3 pawns end up on >>the same rook file... >> >>But it really doesn't matter whether the ending is easy to win or not. What >>matters is to stop searching as soon as possible along a path where the result >>is known. having _all_ 5 piece tables makes that happen. Missing files are >>less efficient, period... > >Based on my knowledge yace does not use 4 vs 1 tablebases. Maybe not. But I'll bet it calls EGTBProbe() whenever a capture takes it down to 5 pieces, whether it is 4 vs 1 or 3 vs 2. And that means a hit stops the search where a miss will not.. > >I suspect that it may be cheaper to use static evaluation in >most 4 vs 1 positions to end the search and not to call tablebases. > >I suspect that a simple rule that there is a mate in at most 20 moves >if the following conditions even in 8 vs 1 and not only in tablebases position >may be correct(I did not prove it but I cannot think now of a negative example): kppp vs k is an example with problems... > >1)the side to move has at least one rook or one queen >2)the opponent has only king >3)the side to move does not control all the squares near the opponent king(to >prevent forced stalemate in the next move). >4)in case of a rook the rook controls at least one square that the king does not >control(to prevent a possible problem of white rook at h8 white knights at h7 g8 >and black king at g7). > >You can decide to be careful not to use this evaluation when the root position >is position when the opponent has only king and you have at least a rook so you >will not have problem of not getting progress in a winning endgame because you >only have mate in 20 evaluations. > >The program without tablebases may show score of mate in 24 some plies before >trading to the endgame when the opponent has only king and smaller score later >but it is not a problem. > >Uri
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.