Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: High NPS vs. Low NPS

Author: Michael Borgstaedt (GOLIATH CHESS)

Date: 06:22:15 11/26/01

Go up one level in this thread


On November 26, 2001 at 07:56:42, Steve Maughan wrote:

>Uri,
>
>Please note I'm not critisising Goliath just using it as an example of a fast
>nodes / second engine.
>
>>What do you mean by piece-table evaluation?
>
>The normal definition e.g. a knight on e5 always gets +350 from a pre-computed
>table
>
>>Goliath does not use piece-table evaluation
>
>I'd be amazed if it didn't.  The nodes / second are too high for it to be
>spending any significant time in the evaluation function.  I guess it's possible
>that Goliath doesn't have the definition of nodes / second.
>
>>you cannot evaluate passed pawn only by piece square table and it seems that
>>goliath knows to evaluate passed pawns and here is the proof:
>>
>>goliath can see positive score for white inspite of the fact that it does not
>>see the promotion
>>
>>Goliath Light 1.5 - Blass,U
>>7k/7r/2pp4/2p5/2p5/8/PK6/7R w - - 0 1
>
>The example shows that Goliath does know about passed pawns and the squared rule
>for them being unstoppable.  In this sense you're right it isn't a *pure* piece
>square table engine.  However from the n/s I believe that it must be mainly
>piece square table based.  The example you give could be solved using a simple
>pawn hash - so I guess that Goliath does a leaf evaluation of pawns coupled with
>piece square tables for the other pieces.
>
>>How do you know this information about goliath and chessmaster?
>>The source code of Goliath or Chessmaster is not free so the only way that you
>>can know is if you talk with the programmers.
>
>Frans Morsch in London 2000 said that Chessmaster evaluated the control of each
>square - I don't know from where he found this.  As regards Goliath, I'm
>assumming from the nodes / second.
>
>Regards,
>
>Steve

Less than 5% of the evaluation in Goliath (Light 1.5) is done by "piece-table
evaluation". Besides, nearly no prescanning is done. The latest version (which
won the blitz tournament during the WM/Maastricht) comes without any type of
prescanning or piece-table evaluation. Little/Light Goliath uses much (!) more
chess knowledge than many users would expect. But Goliath is still extremely
fast and the only "secret" is the way, how I realized the cooperation between
search and knowledge. This has nothing to do with "lazy evaluation". Goliath
uses a special "oracle" which decides, how much and what type of knowledge will
be used for the reached leaf position. And, of course, the essential parts of
the program (search, make/unmake move and so on) are highly optimized for speed.

Michael



This page took 0.03 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.