Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Non-deterministic behaviour of Deep Shredder - REALLY interesting

Author: Charles Roberson

Date: 17:00:57 01/01/06

Go up one level in this thread



 Ah, now I understand your statement. I have two more thoughts on it.

  The first is your probability statement doesn't sound strong to me. The
ordering of search trees make no guarantees on placement of the optimal
solution. Thus, the last 30,000 nodes of 10,000,000 nodes could hold the optimal
path. We use move ordering heuristics that we hope will put the optimal path
early in the tree, but there are no guarantees, hence the use of the word
heuristics.

   The second thought pertains to periodic updating vs real-time updates. Seems
to me the main reason for periodic updates is to speedup the search process.
This is a problem when the updates block each other, thus your solution is a
reasonable thought. However, I do recall that Dr. Robert Hyatt and Tim Mann
wrote a paper on a lockless method for updating. Such a method would reduce the
penalty for realtime updates.

   From Dr. Hyatt's home page:
Robert Hyatt and Timothy Mann (Compaq Computer Corp) "A lock-less transposition
table implementation for parallel search chess engines," Journal of the
International Computer Games Association, Vol. 25, No. 2, June 2002 (63-72).



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.