Author: Dieter Buerssner
Date: 15:32:03 12/27/01
Go up one level in this thread
On December 27, 2001 at 09:26:46, Uri Blass wrote:
>I understand from other posts that it was fail high that
>was not confirmed but I still do not understand how fail high
>can be wrong.
GCP had given some points. For the null move: One could formulate this a bit
more general. Any pruning (that depends on the window) can yield in such
behaviour. Similar, extensions, that depend on the window. Because in the
research, with different window borders, different pruning/extension decisions
can be done. Pure alpha beta without pruning will not show this behaviour.
Gian Carlo also pointed out hash entries of deeper searches, that may be
available at one time and not at another time.
Even with hash entries of the same depth and path dependent extensions, such
things can happen. A path dependent extension would be a recapture extension.
For example
1. capture1 quiet_move1 2. quite_move2 capture2 {recaptures from the first move}
compared to
1. quiet_move2 quiet_move1 2. capture1 capture2
Many programs would extend the 2... move in the second line. Assume the first
line was searched first, and the info after 2. quiet_move2 is stored in the hash
table. Now, when we search the 2nd line with the same search depth, we might
cutoff the search before the second capture, because of the hash, while without
a hash hit, the 2nd capture would be searched with an extension.
One could argue, that one should not allow the cutoff with the same depth, when
the move is a recapture. But this will not help much. Perhaps the best move is
not the recapture and there is no way to detect it. Or more general: I can see
no way to detect, that in a transposition the same moves will be searched with
the same depth. Similar problems come from 50 moves rule and repetition rules.
Dennis Breuker calls this GHI (game history interference) in his thesis, IIRC.
Regards,
Dieter
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.