Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Disequilibrium schemes

Author: Vincent Diepeveen

Date: 16:23:22 10/23/03

Go up one level in this thread


On October 23, 2003 at 04:43:31, Sergei S. Markoff wrote:

>Hello!
>
>>In case you forgot, the evaluation can just return 1 score
>
>Oh when I was such simply guy
>Than I was clean and brave :)
>
>>score it can't return 2 scores for either positional or tactical matters.
>
>The matter is not about the returning more than one score in search. The matter
>is to use eval disequilibrium in current node to make some descisions. It can
>affect search or evaluation itself.

>But, anyway, it's not impossible to use two scores. One example Centaur program
>which uses two evals - pessimistic and optimistic. It's allows to produce good
>cut-offs where optimistic eval < alpha or pessimistic > beta. I don't know
>detatil but it seems to be interesting concept.

The 1997 centaur i exchanged for a copy of diep at the time.

When compared to what diep searched at the time versus centaur, centaur searched
everything but the girl.

>In my case I'm talking not about rebuilding whole search scheme but about more
>clever using of eval. to guide the search (basic SmarThink concept is to make

Do you want to store that in hashtable or something?

If diep must evaluate all its innernodes instead of get 'em out of hashtable
then my thing will slow down exactly 2 times.

So basically that's why i see search independant from evaluation.

Let's not slow down things too much will we!

>search knowledge-based). But it's also about evaluation - what score must return
>eval in a case of losing one or two pawns vs lot positional advantage (king
>attack, passed score)?

I get impression we both have a different viewpoint on search versus evaluation.
I see them as 2 different objects where search borrows at most a few functions
from evaluation. You seem to see evaluation as something inline in the search
that's getting performed at every node?

You search something based upon a search, and leaf evaluations get evaluated.

Using knowledge for extensions is cool, but those are at most a few functions
borrowed from evaluation, not entire evaluation.

That's just too expensive.

The argument that innernodes are in minority when depthleft increases also is
not counting, because it has been proven in history time and again that majority
of tactics you pick up near the leafs. So that's where extensions are most
interesting to do.

>Best wishes,
>Sergei




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.