Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SEE & accuracy

Author: Stuart Cracraft

Date: 14:25:56 08/21/04

Go up one level in this thread


On August 21, 2004 at 15:18:45, David B Weller wrote:

>Hi Stuart,
>
>I follow your threads with great interest, for it seems we are at a similar
>level with our engines [I doubt as a programmer though]
>
>You may have addressed this once already:
>
>GES [my engine] is relatively slow, and thus prone to losing tactics. So I too
>use WACNEW.epd extensively to judge the merit of certain changes.
>
>But I have seen clearly, that what might increase correct answers, average
>depth, fail-high-1, branching factor, and any other metric I follow, will often
>make GES play worse.
>
>What I do, is just make sure the results from wacnew stay within a certain range
>[eg., >= 240 correct at 1sec/pos on AMD 1Ghz] and then test in some games with
>competitive opponents.
>
>Even if you only play a couple games, I think its safer, than making conclusions
>just after running wacnew.
>
>Ive seen this happen very clearly with GES.
>
>that'll be 0.02$,  :)
>
>-David

Hi David -- I think your program is better. Believe me. I've seen the code
(but haven't borrowed anything yet.)

I see what you and others are talking about in this regard and have considered
the same idea. I may just end up throwing larger problem sets at it. ECM
seems to be rather hard.

My program is mostly just for private research and I don't have anyone
other than me who wants to play it but I should renew my account as a
computer account on FICS/ICS, etc. Problem is, I don't know that I'd
be playing against a human. Everyone there seems to use computers to
get the edge.

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.