Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Evaluation: how big the total of bonus and penalty scores?

Author: Amir Ban

Date: 03:02:40 11/15/00

Go up one level in this thread


On November 15, 2000 at 03:14:15, Pham Minh Tri wrote:

>Thank you all for advice. I totally agree that we each should have our own right
>weighting and therefore the evaluation functions could be very different from
>each other. Thanks for all recommendations on how to build it.
>
>However, I believe that there are many common threshold values for chess
>programs and they are always correct with the majority. For example, the
>threshold of 1.5 pawn is very popular – we could apply it to the iteration
>functions or somewhere in programs. That value is taken from an old research
>that statistically surveyed hundreds of grandmaster games (correct me if I am
>wrong). I also believe that chess program is a game of balance of many factors –
>one of the keys of success. You should set many values in the logical ranks if
>you do not want to risk your programs. For example, I do not think someone would
>penalize a couple of isolated pawns by value of queen.
>
>Back to my question, I hope that there is a threshold of _total_ of bonus /
>penalty scores (not detail scores). If you have any experimental/guess one, it
>could help the beginners like me very much with implementation.
>Pham

Chess praxis shows that you can be an exchange down with positional
compensation. There are positions where you can, with justification, be a piece
or more down, but they are rare. Positional compensation for a pawn is routine.

The problem here is mainly of accuracy and correctness of formulation. If your
program will routinely add up vaguely defined positional terms to a pawn or
more, it will also routinely make a fool of itself.

Most programmers would handle this by setting their terms to a size that is big
enough to help their program but not so big as to dominate its action and commit
the program to its correctness. This would mean that positional terms are
undervalued in most positions, so as not to overvalue them in some positions.

Amir




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.