Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Adding knowledge to the evaluation, what am I doing wrong?

Author: Joachim Rang

Date: 11:35:15 03/13/03

Go up one level in this thread


On March 13, 2003 at 14:23:08, Albert Bertilsson wrote:

>Hi!
>
>I've just taken a major step in my engine, adding more knowledge than material
>and pst. As I know little of chess I have very small knowledge of the points to
>give to each knowledge but the major problem is that things seem to get worse!
>
>When I run a 10 position benchmark I've noticed that the knowledge the more
>nodes have to be searched, and branching factor gets higher and higher, this was
>not what I expected, my guess was that with better knowledge the program would
>pick better lines earlier and get a lower branching factor.
>
>The basic implementation know have pst values ranging from 0 to 0.2 pawns, and
>the following defined bonuses:
>
>Pawn = 100
>
>#define DOUBLED_PAWNS -30
>#define PAWN_NOT_ISOLATED 10
>#define BOTH_BISHOPS 25 //Added if both bishops are left
>#define ROOK_HALF_OPEN_RANK 20
>#define PAWN_CLOSE_TO_KING 10 //Added for each pawn withing one square from king
>#define QUEEN_CLOSE_TO_OKING 20
>#define KNIGHT_CLOSE_TO_OKING 10
>
>I know this is just the beginning but I still expected to get better results, or
>is there a flaw in my thinking.
>
>/Regards Albert

I know little about adding knowledge, but I think your values are generally too
high. In the following post on the message board of chessbrain you find all
default values for Beowulf described by Colin:

http://www.chessbrain.net/bbs/viewtopic.php?t=145

For example, colind add only 4 centipawns for a rook on a half-open file. Your
last two values I don't understand. do you add a bonus for the knight or the
queen to be close to its own king? This seems strange for me.



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.