Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Artificial Intelligence in Computer Chess - *DETAILS* as promised

Author: J. Wesley Cleveland

Date: 09:43:46 03/30/04

Go up one level in this thread


On March 29, 2004 at 11:14:17, Anthony Cozzie wrote:

[snip]
>
>What I am proposing is this:
>
>Create some code that stores for each transposition table key the best move at
>each ply.  Requires maybe 1/2 GB of memory for a reasonable depth search (my
>back-of-the-envelope unoptimized computation).
>
>E.G.:
>
>Key 0x8432762183
>Q-Nxd5
>1-Nf3
>2-Nf3
>3-Qc3
>...
>
>Run search, record node counts.  Keep track of every time the move ordering was
>wrong (both at PV and CUT nodes).
>
>Flush transposition table. Run search iteratively until move ordering is
>perfect. You can use the new table to simply always choose the best move first.
>This might require a few runs since the tree might search different nodes as the
>move ordering gets better.
>
>At that point you will have a search of muse with perfect move ordering.  Record
>the number of nodes visited versus a regular search. Of course, this is not
>practical for a tounament game (for obvious reasons).
>
>Some interesting sub-experiments: move ordering differences by ply - 3% at 5 ply
>vs 100 % at 8 ply (?)  Also, move ordering at PV nodes vs fail-high nodes.
>
>anthony

There is a simpler way. Just keep track of the minimum necessary nodes as you
search. At any ply with n possible moves, if it is fail low, the mininum number
is n + the sum of mininum nodes from the ply below for all moves. If it is fail
high, the mininum number is 1 + the mininum nodes from the ply below for the
fail high move. You should probably turn off transposition tables if you want
more exact numbers.



This page took 0.01 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.