Author: Bruce Moreland
Date: 14:43:58 12/15/99
Go up one level in this thread
On December 15, 1999 at 12:04:05, Dan Homan wrote: > >In a game recently sent to me by a tester, EXchess reached the following >position against another program > >8/8/8/7r/2K5/8/6kp/4R3 b - - > > +---+---+---+---+---+---+---+---+ >8 | |:::| |:::| |:::| |:::| Black to move > +---+---+---+---+---+---+---+---+ >7 |:::| |:::| |:::| |:::| | castle: - > +---+---+---+---+---+---+---+---+ >6 | |:::| |:::| |:::| |:::| > +---+---+---+---+---+---+---+---+ >5 |:::| |:::| |:::| |:::|<R>| > +---+---+---+---+---+---+---+---+ >4 | |:::| K |:::| |:::| |:::| fifty: 0 > +---+---+---+---+---+---+---+---+ >3 |:::| |:::| |:::| |:::| | > +---+---+---+---+---+---+---+---+ >2 | |:::| |:::| |:::|<K>|<P>| > +---+---+---+---+---+---+---+---+ >1 |:::| |:::| | R | |:::| | > +---+---+---+---+---+---+---+---+ > a b c d e f g h > >In this position, EXchess probed tablebases, found a mate and moved >Rh4+? > >It turns out that the tester has the KRKRP tablebase, but not the >KRKRQ tablebase. So as long as the pawn was on the board, EXchess >saw a mate, but with the pawn promoted to a queen - EXchess would >have to search several plies to see the mate (after an exchange into >one of the 3 or 4 man tablebases). As a result, EXchess will >prefer Rh4+ in this position for several iterations - unfortunately, >I have the search set to terminate if a mate is found and held after >the 3rd iteration. > >Even if I didn't have the search set to terminate under these conditions, >selecting a move like this would still be a problem if EXchess was under >any kind of time pressure. > >Has anyone else dealt with problems like this in their programs? I >assume this kind of thing has come up before... One quick fix would >be to turn off all tablebases if a promotion is possible at the root >position and the corresponding tablebase doesn't exist. There might be >other variations on this problem that this fix doesn't cover, however. > > - Dan You have to have all common sub-databases. You might be able to get away without having extremely uncommon ones, for instance you might not need KBN vs KN in a KNP vs KN ending, even though there are some cases that can only be solved by promotion to a bishop. But refusal to promote is an obvious problem when you don't have the databases that you need, and you need to make sure that your testers don't blow it. bruce
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.