Computer Chess Club Archives


Search

Terms

Messages

Subject: The Centrality Perturbation to Move Selection Order

Author: Stuart Cracraft

Date: 21:33:32 01/29/98


Sorry about the absurd subject, but I had to get your
attention somehow. :-)

Here it is...

When I sort moves, most of the moves don't fall into
a standard category that gets assigned a non-zero score
(killers, captures, hash moves, history heuristic valued,
etc.) Instead, most moves end up in the "other" category.

This category ends up giving the move a zero score. Most
scores therefore were in the bottom of the heap and chosen
randomly, actually in the order that the move generator
generates them (knights first, then kings, then bishop,
rook, queen, pawns last), since my sort doesn't change
locations of moves or pointers to them and leaves the
zero-scored ones in the same order as the move generator
generated them.

So recently, I added a centrality perturbation that
added 15 minus the taxicab distance from the move's
destination square to the center of the board (chosen
as E4 since I wanted to avoid real arithmetic to 4.5.)

Anyway, the result was a major drop in search speed on
certain searches and a drop in results in Win-at-Chess
from 204 to 199 positions solved out of 300, just for
the lousy centrality perturbation of the move sort order.

Anyone else seen this kind of thing? I didn't even change
the static evaluation. Nothing else was changed.

--Stuart



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.