Computer Chess Club Archives




Subject: Move Extensions

Author: Roberto Waldteufel

Date: 13:38:43 05/04/98

I have read about the singular move extension heuristic as used by
Deeper Blue, and indeed about extension heuristics in general. I have
also experimented with some variable depth schemes in my search trees.
One quite promising idea is to set the depth to some arbitrary value at
the start of each iteration (this value being increased a little prior
to each successive iteration), and then decrease the depth at each odd
level by an amount proportional to the logarithm of the number of legal
moves for the opponent at that level, and calling a quiescence search
when this depth drops below zero. The depth remains unchanged at even
levels. This has the effect of pusuing forcing lines for the machine
much deeper, even if some quiet moves are interspersed between the
checks. It works very well for some tactical positions where the machine
can mate or win material through a long sequence of moves, most, but not
all of which are checks.

I understand that in the singular move extension scheme, the node is
extended if its evaluation is significantly higher than that of its
next-best sibling, but I do not understand exactly how this condition
can be tested when a beta-cutoff occurs, since the evaluation scores for
some of the sibling nodes is then not known to the algorithm. Can
anybody enlighten me?

Robeto Waldteufel

This page took 0.03 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.