Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty - Forward vs. Backward Analysis

Author: Robert Hyatt

Date: 07:58:41 06/10/02

Go up one level in this thread


On June 10, 2002 at 08:30:01, Steve Coladonato wrote:

>Is there any significance in the quality of the analysis that is produced when a
>program (Crafty) performs its analysis forward, i.e. starting at move 1, through
>the game vs a program(s) (Chess Tiger, Shredder) that perform analysis backward,
>i.e. starting at the last move.  In the annotate command, I set the "margin" to
>be .10 pawns and the "time" to be 90 secs.
>
>Also, can anyone tell me what the recommended environment/settings should be for
>Crafty on Linux?  I'm running on a Gateway 733MHz Pentium III Laptop.
>
>Steve


Here is the "thinking" about going backward:

as you back up thru the game, scores from later searches are stored in the
hash table and they propagate back up the game as a result, letting the program
often see some tactical flaw earlier in the game (because it knows about the
tactical problems that occur later in the game first).

I don't like it.  And here is the "why"...

When you go thru a game in the forward direction, the program will spot any
tactical oversight its search is capable of finding.  When it complains that
you made a worse move than necessary, this will be based on its search, and it
will be repeatable across games.

When you go backward, you hope that the important tree search results stick
around long enough to be used a few moves back in the game, giving you better
scores.  But this is based on a lot of serendipity (luck).   If key table
entries survive, you get better analysis.  If they don't, you don't.  This
means that the "analysis" by the engine has a strong luck component in how
accurate it is.

I personally prefer "consistency" to "spotty genius-like" analysis because
I want the same quality of analysis for each of my games...  It makes it easier
to understand what the machine sees.

Remember that the computer will produce some analysis at the point in the game
where it sees the score drop.  Going forward, it will happen at about the
same "depth" each time.  Going backward, it will vary significantly...



This page took 0.01 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.