Computer Chess Club Archives




Subject: Re: evaluation tuning tricks

Author: Fabien Letouzey

Date: 07:33:33 03/17/04

Go up one level in this thread

On March 17, 2004 at 10:25:24, Tord Romstad wrote:

>On March 17, 2004 at 09:55:59, Fabien Letouzey wrote:
>>No offense Tord, but I don't understand  why programmers think the move ordering
>>is "perfect" if a fail-high move is found first 100% of the time.
>Because the percentage of first-move fail-highs is easily measured?  Of course
>I agree that the size of subtrees is important, but I don't see how you can
>determine how often a fail-high move has a smaller sub-tree than all other
>fail-high moves at the same node, except by searching the whole minimax tree.

You can do a "minimax" search for the first few plies and then use alpha-beta to
approximate the size of the minimal tree.

Of course it's a lot of work.


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.