Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Thinking out eval improvements and a test pos.

Author: Robert Hyatt

Date: 19:16:16 08/06/04

Go up one level in this thread


On August 06, 2004 at 19:07:24, GeoffW wrote:

>On August 06, 2004 at 18:33:26, Robert Hyatt wrote:
>
>>On August 06, 2004 at 18:21:51, GeoffW wrote:
>>
>>>Hi
>>>
>>>My program has been playing some test games against Dorky and getting soundly
>>>beaten, it got me thinking about my eval function which is pretty dumb and badly
>>>in need of some chess knowledge !
>>>
>>>What I noticed in some of the games was that Dorky was seeing threats sometimes
>>>10 or 15 plies earlier than my program, as we were searching to similar depths
>>>it could only be that Dorky has much better eval function.
>>>
>>>Here is one of the positions that showed this effect, material is even, but
>>>black looks to me to be in a horrible position
>>>
>>>[D]1b3rk1/4qb1p/2r1pRpP/2ppP1P1/P3pQB1/1PB1P3/8/5RK1 b - - 9 44
>>>
>>>My program (Black)reckoned that it was only 0.25 of a pawn down, where as Dorky
>>>reckoned white was winnning by 2.2.
>>>
>>>I tried this position with a few programs, only Dorky and Crafty scored white
>>>ahead by > 1.5 pawns.
>>>
>>>Any suggestions as to what the important factors are to evaluate this position
>>>as strongly good for white ?
>>>My program hasnt really got any mobility scoring, no king tropism and rather
>>>basic pawn structure eval. Wondering which area I ought to try first ?
>>>
>>>   Regards Geoff
>>
>>
>>Crafty's static eval here is +1.5 as you mentioned...
>>
>>The primary points are (a) the f-file is open and occupied by white's three
>>major pieces.  The pawn at h6 makes g7 another weakness, not to mention being a
>>potential passer if the h7 pawn goes away.
>>
>>Main problem here is king safety, with the white majors pointed right at the
>>black king.
>
>Hello Bob
>
>Thanks for the quick reply, I was checking the score with Crafty as you posted
>
>note: scores are for the white side
>material evaluation.................   0.00
>development.........................   0.00
>pawn evaluation.....................  -0.13
>passed pawn evaluation..............   0.00
>passed pawn race evaluation.........   0.00
>king safety evaluation..............  -0.46
>interactive piece evaluation........   1.88
>total evaluation....................   1.29
>Black(1):
>
>king safety as listed here is actually good for black, the important factor
>though is the large interactive piece evaluation of 1.88

That is "first order king safety".  Second-order king safety includes piece
locations and coordination.  It is more important.  The first-order term is
purely pawn structure around the king...


>
>Is this term simply adding up the combined attacks against the enemy king , or
>does it look for more subtle factors as well?
>I assume Crafty must look at the squares adjacent to the enemy king and also
>ignore the defending rook and bishop ?

It is a combination of things.  But the main problem is the f-file.  White owns
it and has multiple threats because of this..  Black should never have allowed
the white heavies to triple-up uncontested...





>
>I guess the position above would have scored highly for this term as I think
>Crafty may have a non linear scaling for this type of multi piece attack ?

Correct.  It is an exponential-type function where each additional piece ramps
up the score exponentially.  IE 4 pieces are not twice as good as 2.  They may
be 8x better...


>
>Incidently, even a strong engine like Gothmog after analysing this position for
>30 seconds, still only reckons white is ahead 0.4.


Some types of attacks are very deep.  But with black paralyzed because of the
f-file, it appears to be only a matter of time...




>
>     Geoff



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.