Computer Chess Club Archives


Search

Terms

Messages

Subject: HYBRIDing Fritz and HiArcs

Author: Steve Schooler

Date: 01:41:28 09/02/99


The Chessbase support notes indicate that the Fritz 5.32 engine is 10
times faster than the HiArcs 7.32 engine, because of HiArcs extensive
"positional library".  Based on recent postings, HiArcs 7.32 seems
(generally) more powerful, despite it's "slowness".  I believe further
progress may be made by "merging" the two engines (algorithms).

As the opening book is exited, use current HiArcs algorithm.  As
the game progresses and portions of the "positional library" become
redundant, eliminate those portions from algorithm.  During the
game, as more portions of HiArcs positional library are eliminated
from the algorithm its speed (nodes per second) should approach
Fritz'.

Perhaps I'm misunderstanding HiArcs' positional considerations and
the hybrid strategy is not feasible.  Alternatively, perhaps the
hybrid simply requires reorganizing HiArcs' positional library.
Pawn promotion considerations would be tricky.

If the hybrid is feasible even greater improvement may be possible by
adding a 3rd category within HiArcs' positional library; in addition
to Relevant and Permanently Irrelevant (unless a pawn is promoted),
also allow the category of Temporarily Irrelevant.  At the start of
the procedure used to evaluate a position (node), ignore the
Permanently Irrelevant portion of the positional library, and
(quickly) scan the remainder of the library to determine which portion
may be classified as Temporarily Irrelevant.  This would (hopefully)
further limit the portion of the library that needs to be accessed
when evaluating this node.

Again, I may be badly misunderstanding HiArcs' positional
considerations.  I request general programmer feedback; I'm especially
interested in Chessbase' reaction.



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.