Author: Uri Blass
Date: 02:34:14 12/06/02
Go up one level in this thread
On December 06, 2002 at 05:23:02, Richard Pijl wrote: >On December 05, 2002 at 11:40:34, Dieter Buerssner wrote: > >>On December 05, 2002 at 07:51:02, Gian-Carlo Pascutto wrote: >> >>>Some programs have a workaround that makes the ugliest >>>appearances of this problem go away. >> >>Can you give an example, preferably from a real game, where known workarounds >>would not make the problem go away? >> >>In the discussed case, probably even using the normal search time instead of >>exiting search early after a mate score woulh have helped. >> >>>But that doesn't >>>fix the problem - it still occurs everywhere else in the >>>search tree. >> >>I don't understand this. Why should there be any problem in the search tree? >>Until a "critical" situation is at the root, what should be the problem? >> > >Tablebases might force a program in a position it doesn't know how to win, for >example because the position is an exception to general rules. These positions >are possibly avoided by the chessknowledge of the program, but lured into it >because it sees a forced mate from the table bases. It might even sac material >because of it. > >Richard The main problem are with incomplete tablebases are not positions that the the program does not know to win without tablebases so I believe that in the general case using incomplete tablebases when you have good rules when to stop to use them is productive. I mentioned the problem of the 50 move rule as a reason not to score mate from tablebases as mates. Lack of chess knowledge is another reason and programmers may consider not to score mates in less than 51 moves from tablebases as mates in case that their program has incomplete tablebases for the relevant tablebase. They may translate mate in 50 from tablebases to a constant (+4) mate in 49 to (+4.2),mate in 48 to +4.4 so if they have a simple way to win they will prefer the simple way and not the tablebases. This should be done only if the program has not all the sons of the relevant tablebases but a program that is using tablebases correctly should know if it has the sons or does not have the sons. 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.