Author: Christophe Theron
Date: 22:18:21 08/31/01
Go up one level in this thread
On August 31, 2001 at 16:05:48, Peter Berger wrote:
>On August 31, 2001 at 13:05:35, Christophe Theron wrote:
>
>>It does not make any sense to talk about "optimizing for 386 or 486 tournament
>>time controls".
>
>How come ? It makes perfect sense IMHO : the conclusion ( and a very possible
>one) is that it never existed - if this conclusion ( or any other one btw) is
>reached it was useful to talk about it as a problem that obviously was of
>interest to some ( they took the effort to post ) was there and was resolved .
>People who think it is futile can keep away from the thread or ask for
>moderation if they think it hurts their general reading experience. People who
>have valuable information and feel like joining can provide it and help the less
>knowledgeable. If someone has no new information, opinions or questions he can
>still read and learn as long as interested.
>
>I sometimes think this policy could be useful in some of the Deep Blue threads
>also.
>
>And I don't see I suggested anything that contradicts your statement anywhere
>anyway - as I agree to your opinion.
>
>>
>>It would take years to achieve a task like this, and while this optimization job
>>would take place the author would not be able to make any serious change in his
>>program.
>
>Maybe your opinion is too extreme here ( or better your idea how such an
>optimization might happen) . It might be more about ways of testing . An extreme
>example : an author tests every major change he makes in 1000 1/0 bullet games
>against GNU on his dedicated test computer . The engine might end up being
>overtuned for being successful against GNU in the end - and it is conceivable it
>will be stronger in Bullet games than at slower time controls.
>
>I have read a few posts from chess programmers and beta-testers that explained
>how they do their tests and I think some of them seemed to show something that
>points into a similar direction but I am not the right person to discuss this.
>
>>
>>An author simply tries to make his program stronger, and that's already a task
>>difficult enough, from the human point of view.
>>
>>I do not know of any improvement that would be a blitz improvement only (I mean
>>an improvement that would only help in blitz and not at longer time controls).
>>Likewise, I do not know any improvement that would only help at long time
>>controls.
>
>I think some of the things Genius _seems_ to do might be better in blitz than in
>longer games- for example the way it seems to evaluate some pawn structures ,
>but I won't fall in the trap to talk about things I don't really understand and
>won't go on .
I have a simpler explanation: Genius is handicapped at longer time controls
because of its higher branching factor.
This theory:
1) can be proved (measure the branching factor of Genius and Tiger on some large
sample of positions).
2) does not call for any other explanation like an hypothetic optimization for
some sort of time controls.
>>"Optimizing for blitz or tournament time controls" is a fantasy in the mind of
>>many readers of this discussion group, but as many other things I see discussed
>>here, it does not exist.
>>
>> Christophe
>
>OK - a sentence like this will somehow always be right - or will never be proven
>wrong.
It's always the same principle I am using: instead of trying to add new ideas to
a theory, start by checking if facts cannot be explained by the existing theory,
or by an even simpler theory.
It's the principle of economy in sciences. It's pretty successful.
I'm not smashing the idea of "optimized for a given time control" because I
don't like it, I smash it because I don't need it in order to explain what I
see.
Why should I make the world around me more complex than it is?
I can explain the mechanics of the whole universe by stating that there is a
little god governing any object, but does it help me to solve any serious
problem?
>A question about testing the Palm Tiger added : I assume computer opponents on
>68000 processors would be a fair match as this is a similar architecture ?
Yes, approximately.
Christophe
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.