Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: sliding attacks in three #define

Author: Tord Romstad

Date: 08:10:51 04/14/04

Go up one level in this thread


Hi Martin!

On April 14, 2004 at 10:35:26, martin fierz wrote:

>On April 14, 2004 at 10:07:35, Tord Romstad wrote:
>
>>On April 13, 2004 at 20:39:23, Vasik Rajlich wrote:
>>
>>>Well, the most obvious explanation is that there's simply a speed vs knowledge
>>>tradeoff.
>>>
>>>However, there are some interesting data points regarding evaluation functions.
>>>
>>>One is that Fruit is not so bad, considering that it apparently lacks the one
>>>single most important middle game evaluation factor. (King safety.)
>>
>>Or even Crafty.  Have you seen Crafty's king safety eval?  I almost couldn't
>>believe my eyes the first time I saw it.  Apparently king safety is no longer
>>very important if your engine is sufficiently fast.
>
>i use crafty 19.~6 as a sparring partner for my engine. it is far better than
>muse of course. but i have seen *many* *horrible* moves by crafty, regarding
>king safety. i never checked it's evaluation, and i don't intend to.

Basically, what it does (except for some pawn shelter analysis) is to
measure the geometrical distance of the enemy pieces to the king.  Pieces
are simply given a bonus for being located close to the enemy king.
Whether they attack any important squares is hardly considered at all.

>i just want to say that when crafty loses against muse, it is more often than
>not that crafty's king safety eval seems bad to me.
>
>on the other hand, the main reason i decided to use attack tables in muse is for
>king safety evaluation. they really pay off there, but the price i'm paying is
>huge - i search even less nps than gothmog ;-)
>i think writing a good king safety eval requires a lot of computation and then
>you're just back to complex+slow or cheap+fast.

Yes.  But although attack tables are the most natural and obvious tool to
use, and I use them myself at the moment, I think there are better and faster
ways to evaluate king safety.  I will experiment with other techniques in my
new engine, and of course let you all know if I discover something useful.

Tord



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.