Author: Uri Blass
Date: 14:04:34 10/11/03
How do you solve instability problem in null move after fail low? The problem is the following: The program fail low and suppose that it finds mate against itself after move A. Move B is also a bad move that lead to mate against the program for the same reason but the problem is that the reply of the opponent does not threat mate when it is searched to reduced depth so the program fail high. After the window is opened the program see that the reply of the opponent threats to win a pawn so it searches it and find that the reply leads to mate. The result is that the program often fail high on bad moves only to fail low later and needs more time to find the right move to prevent mate. I think that it may be possible to avoid the problem by changing the definition of a threat after a big fail low when threat is not relative to the score but relative to the score of previous iteration but before I do something about it I am interested to know what other do about that problem because I guess that it is a common problem. Maybe storing in the hash tables not to do null move because in previous search null move was bad is a better solution but I think that this solution does not always work because it is possible that the relevant positions were not searched thanks to alpha beta. I am also not sure if it is a good idea to store not to use null move regardless of beta because if the score is changed I want also the concept of threat to be changed(otherwise I may search too many lines that really have no threat). 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.