Author: Bruce Moreland
Date: 14:44:00 11/16/00
Go up one level in this thread
On November 16, 2000 at 15:36:46, Bob Durrett wrote:
>On November 16, 2000 at 13:35:17, Bruce Moreland wrote:
><snip>
>
>>Why is evaluation more intelligent than search? If you have a program with a
>>very simple search, that spends a large percentage of its time in evaluation,
>>and you make the search more clever, so that a higher percentage of the time is
>>spent in search, did the program become dumber?
>
>"make the search more clever" sounds like a great idea! In future programs [new
>paradigm? (sp=?), one might consider making the search considerably more
>"clever."
>
>It would be interesting to see ideas on how that might be done.
>
>My first thought would be to find a way to incorporate "chess planning"
>knowledge into the search strategy so as to favor evaluation of lines which are
>consistent with appropriate chess plans [which, in turn, are based on earlier
>position evaluations made to determine what the appropriate (or feasible) plans
>are for each side.]
An obvious idea is to make a program understand that it is being pressured. You
can do this by counting legal replies. If you have very few, you may be being
pressured.
Another way to make a program understand that it is being pressured is to cause
it to realize that it has only one move that keeps the score above alpha.
Another thing to do is to make the search realize that the opponent has no
threats in a position.
All of this has been done.
>>And what about programs that do incremental evaluation? A good incremental
>>evaluation should produce the same results as the same evaluation carried out at
>>the tips, and it should do it more quickly. Is this program dumber?
>
>To elaborate on this very interesting idea: Why should a position be evaluated
>"from scratch" each time a new position is reached? That's not the way I do it
>when I play chess. In my own serious chess games, information I found out in
>earlier positions typically do influence my evaluation of the current position.
>Perhaps this is true also for all chessplayers. Why not also for the computer
>program?
>
>>
>>I think that one of the greatest wastes of time is trying to assign categories
>>to these programs.
>
>But . . . people keep asking "is this a knowledge based or a bean counter
>program {also called "brute force")? What shall we do with those people?
>
>Inherent in their questions about terminology seems to be the more important
>question: "Which approach should I take in the design of my new chess-playing
>program?" "Which approach offers more promise?"
Another aspect of the question is that it gives you an indication of what
program you should buy.
People don't want to simply get whacked by their programs, they want to get
whacked well. They want to feel like they've been whacked by a human, so they
can practice against the computer and feel like they are practicing against a
strong human. The idea being, of course, that eventually they can beat strong
humans, which will make them strong humans, too. And that is their overall
goal.
The concern is to avoid moves that look computery, and it's been considered by
many that a brainless eval results in computery looking moves.
Some people have also made claims that a smarter program will perform better
given a hardware increase.
And so on, it goes on forever, although people rarely talk about subjective
stuff like contrasting styles or the "educational" strength of chess programs.
All you hear about are match results and computer vs computer ratings and other
silliness. There are lots of specific topics that can be discussed for instance
how easy it is to king-attack various programs, or which programs will trade
into a lost K+P ending rather than keeping the last pair of rooks on, or the
benefits of various pawn structures, but none of this gets discussed.
I think silliness is fine, but we've had years of it here, there are a lot of
people spending a lot of time spinning their wheels.
bruce
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.