Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The "Correct Assessment" of a Chess Position

Author: Mike S.

Date: 10:02:57 01/11/04

Go up one level in this thread


On January 11, 2004 at 12:01:45, Bob Durrett wrote:

>(...)

>These two authors [primarily Gufeld] conveniently split position assessment into
>what they call "intuitive" assessment and assessment via "concrete calculation."

It's typical for (some) russian chess writers, that a lot of "blah" is written
using complicated wording, around things which are actually simpler. I know that
i.e. from Suetin's text paragraphes in many books. His examples and in-game
comments are usually very good and instructive though, and a pleasure to replay
in addition to the intended learning effect. All the text is just decoration
around the real chess contained.

>(...) This allows them to see
>positional sacrifices in their proper context, i.e. pursuit of strategic goals.

This idea is on a strength level way beyond 2000 (or even 2200) though.

>If the position cannot be properly evaluated by "intuitive assessment," then
>some specialized "searching" would have to be accomplished before the position
>assessment could be completed properly.

I'm no programmer, but I think the easiest example of this is, when an exchange
variant isn't finished at the current depth level. Then, the so called
*quiescence search* must be performed. This is even known since Shannon, ~1950.
A chess program without that would be chaotic.

But that mainly relates to the material balance, and maybe to checks, not to
positional things. A "silent" position can always be evaluated by a chess
program sufficiently so to speak, I assume.

>If I understand the bulletins posted here at CCC, chess programmers typically
>use "position evaluation" primarily or solely for the immediate determination of
>which move continuations to search.

I don't think so. The moves are not evaluated (themselves), but by the positions
*at the end of the variants* which are started by these moves, each. This is
repeated on each ply level, when the program increases the search depth by one
ply (basically, +extensions). What to search deeper and what not, is decided by
the mechanics of the (alpha-beta etc.) seacht algos, which I admit that I don't
understand much of their details.

This is certainly simplified; maybe programmers can add if it's basically
correct.

(So far it was sufficient for me to understand the minimax tree concept :-))

>Also, it seems to me that some of the "searching" should be done for the purpose
>of completing the evaluation of a specific "internal node" position [other than
>the position on the board in the game].  This implies that the general nature,
>or character, of such searching might be different from the "ordinary"
>searching.

As far as I've understood of all this so far, the normal evaluation actually is
always the one for a "future" position **at the end of the primary (main)
variant,** maybe 20 or more plies ahead. This *is* the ordinary searching.

(Maybe I don't understand what you mean. :-)

A "correct" assessment of a position is most often impossible, except in forced
mates or forced repetitions etc. In all other normal positions there must remain
some uncertainty, because the calculation depth has a limit.

IOW, only a proven evaluation of 1-0, 0-1 or 0,5-0,5 can by entirely correct.
The rest are estimations based on the continuation which the engine expects as
the one with the highest probability, based on search and evaluation. Blah yadda
yadda. :-))

Regards,
M.Scheidl



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.