Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A question for programmers of top chess programs

Author: Vincent Diepeveen

Date: 11:35:53 06/17/01

Go up one level in this thread


On June 17, 2001 at 12:56:20, Uri Blass wrote:

>On June 17, 2001 at 09:09:27, Vincent Diepeveen wrote:
>
>>On June 17, 2001 at 04:11:26, Uri Blass wrote:
>>
>>>Are there top programs that are designed to see always simple tactics after
>>>enough time?
>>>
>>>Is it possible to change them to see always simple tactics after enough time
>>>when the price is only being 0.1% slower by using 2 engines when one engine is
>>>usually used only for only 0.1% of the time or maybe my math is wrong and only
>>>using another engine makes the program already more than 0.1% slower because of
>>>the fact that the program is bigger(even if the other engine is simply the
>>>original engine without null move pruning)?
>>
>>Here i can jump in with double nullmove. A few years ago some people
>>doubted nullmove to be a correct way of searching in *any* respect.
>>
>>Nothing would beat fullwidth. Hence Deep Blue being fullwidth.
>>Nowadays thinking has changed after everyone has been beaten by
>>nullmove.
>
>Not exactly truth.
>
>Junior is still one of the top programs and it is not a null mover.

Junior sees because of this positional uncommon lines to very small
depths like 5 ply at most in total when being at a 17 ply search
depth, as each move eats up 3 ply possibly.

>It has other problems of not seeing simple tactics(for example uncorrect
>stalemate detection that happens in some rare cases so the engine can say mate
>against itself when there is a stalemate combination).

Junior is only searching the tactical search space as it positionally
will of course never be in the same league as todays programs.

I'm amazed it's still scoring so well. Of course junior is tactical
one of the greatest as it is focussing on seeing tactical lines very
deeply.

This whole junior system will fall apart as soon as you try to improve
its knowledge. The reason for that is that one needs to do more near
the leafs then. So not using a qsearch is out of the question then.
So are very dubious ways of pruning.

But in chess there is only one truth when talking about score, and
that's the lemma that the weakest chain of a program decides whether
you win or lose. So far junior team did a very good job in not having
weak chains, but IMHO the positional chains are already its weakest
chain quite some time.

The bitter conclusion for those who want to focus on search only
nowadays must be that junior is one of the last programs that is
like this. I can't imagine any new program getting in the top which
is not focussing on evaluation rather than tactical search.

This at the same time is also the reason junior is still alive.
Nowadays it is bigtime fahsion to overevaluate attacks to the king,
and overevaluate rooks on 7th rank and being on half open files to
the opponentking.

Being one of the few programs with a very careful scoring here
brings junior in a position that the other programs are basically
tuned to beat an entire different type of program as junior is.

Best regards,
Vincent

>Uri



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.