Computer Chess Club Archives


Search

Terms

Messages

Subject: A problem with tablebases

Author: Jeremiah Penery

Date: 03:10:21 04/03/00


There is a problem with tablebases.

No, they don't produce wrong answers.  And no, they don't make engines play
weaker.  The problem is that engines would rather be -9 and not in TBs than see
mate against itself and be in TBs.  I have one good example, from a game my
modified Crafty played on ICC, and I may try to find it, if possible.  But I do
remember what happened:

I had something like a knight and a pawn, and my opponent had 2 knights and a
pawn, about to promote.  I had the choice to let him promote, and have a score
of about -9.  I would easily lose.  My other choice was to sacrifice my knight
for his pawn, and enter a lost KNNKP endgame, somewhere around Mate in 75.  I
would not easily lose this, unless my opponent had the TB, in which case it
could be a draw by 50-moves.

I forced Crafty to enter the KNNKP endgame, and got a draw out of it.

Had I not done so, Crafty would have preferred the -9 score, which would
eventually lose very easily due to the opponent's new Queen.

I'm wondering if anyone has experimented with making the engine prefer difficult
lost TB endings over easily lost non-TB endings, especially in cases like KNNKP.
 I think this could at least turn a few lost games into draws (as it did for
me), and possibly something better could come of it if some interesting new
ideas were formulated based on this.

Jeremiah



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.