Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Incremental Evaluations

Author: Peter McKenzie

Date: 12:54:25 10/27/98

Go up one level in this thread


On October 27, 1998 at 09:16:59, Mike Stoker wrote:

>Does anyone know if any of the current chess programs perform incremental
>evaluations of positions.  I.e. To evaluate a position after a move is played,
>just consider the changed features of the position.  I think a possible way to

Most programs maintain at least part of their evaluation incrementally, for
example material, and the piece square table based evaluation.  For the pawn
structure evaluation, using a separate hash table is even faster than
maintaining the score incrementally.  For some other things, incremenatal
evaluation might get pretty complex.

>do this is to remove all evaluation code from a program and to hand code it into
>a neural network type of structure.  At the lowest level, there are the
>positions of individual pieces.  At the highest level, there is the positions
>evaluation.  At level inbetween, there are scores for certain features and
>partial evaluations.  In order to find the change in evaluation after a move is
>made, simply trace upwards from the bottom nodes which specify the changed piece
>position, and recalculate each nodes that the changed nodes form an input into.
>Similarly, if the output from a re-calculated node changes, re-calculate the
>nodes that IT feeds into.  In this way, only the relevant features of a position
>are re-calculated.
>
>Any thoughts?

Interesting idea, my advice would be to start simple and build on that.

>
>Regards,
>Mike.

cheers,
Peter



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.