Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Evaluation testposition

Author: Gian-Carlo Pascutto

Date: 10:38:59 03/20/02

Go up one level in this thread


On March 20, 2002 at 13:17:36, Scott Gasch wrote:

>Wow.  My static eval for white here is -0.5 pawns.  With a search the engine
>thinks it can force a draw and is happy to do so.  As for _how_ I evaluate it,
>the major component of the eval is the white passer.  However it's blockaded by
>an enemy piece sitting on a square controlled by the enemy so it has no >prospect of advancing.  That chops down the bonus to that passer severly.

Ok, I do something similar here.

>I do give some credit to white for aiming its rook/queen at the enemy king but
>monsoon does not understand the mate threat it poses.

Hmm. A large component in my eval here is that white's king is unreachable
for black, while black's has lost all it's protection. Add the threat on f7
and the mate threat on g7, and my thing thinks this is about as bad as it
gets.

I have a system that reduces the score as pieces go off, but it considers
a rook and a queen still plenty dangerous. I fiddled a bit with so the
reduction is a bit bigger, and it also enforces a maximum kingsafety penalty
that may be applied, and now it's down to more reasonable levels (+1.5-+2.0)

It seems to work, at least here, but it still feels somewhat like a hack,
especially because most are getting such very _low_ scores here.
Is there something intrisic in this position that makes white's threat just
worthless that they understand?

>I think my eval here is
>just as skewed as sjeng's, just in the opposite direction.  Do you actually
>evaluate the mate threat?

It understands that this configuration of pieces leads to a mate threat
for white. (It'll also understand it has to be more carefull with pruning)

--
GCP



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.