Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Knowledge V Search Depth Illustrated With Diagrams

Author: Graham Laight

Date: 02:00:34 05/12/00

Go up one level in this thread


On May 11, 2000 at 20:52:54, Bruce Moreland wrote:

>Chess programs search to variable depth, and their evaluation function is
>accurate with variable degree of confidence.  Strength is search plus positional
>understanding of key groups of positions.
>
>Sometimes a human will win a game on brute tactics, by seeing more in a tactical
>position.  The odds of this are reduced with better hardware, since when you add
>hardware you improve the search aspect of strength most.
>
>Other times a human will win because the game is driven into a situation where
>the computer's evaluation is consistently useless or even worse than useless. I
>call this a "dead space" in the tree.  Sometimes the whole tree is dead space.
>In these cases increased search can't defend a program adequately, so improved
>hardware doesn't improve these cases significantly.  Sometimes it is possible to
>improve these cases by improving eval, but other times it is very difficult to
>do anything about them without affecting general-case positions adversely.
>
>Effective strength can change dramatically based upon the root position you
>search from, and not in every case do you get significantly better if you add
>horsepower.
>
>bruce

Hi Bruce,

A picture paints a thousand words, so let me try to paraphrase you
diagramatically:

What you seem to be saying is that a chess programs' knowledge (or positional
understanding) + search extensions may give it a knowledge profile as
illustrated below:


 ply |-------------------------------------------------------------|
     |  $$$$$                                      #               |
25   |                           #                 #               |
     |                   #      ##                 #               |
20   |                  ##      ##                 #       #    #  |
     |                  ##     ###                 #      ##    #  |
15   |        ##   #   ###    ####               ###      ###  ##  |
     |   ########  # ##############         #########  ###### ###  |
10   |#############################################################|
     |#############################################################|
5    |#############################################################|
     |#############################################################|
     |-------------------------------------------------------------|

                         Breadth of knowledge

However, if the significant knowledge required is in this instance is where the
"$" symbols on the diagram are, then all the hard work the computer is doing is
effectively going to waste.

-g



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.