Author: James T. Walker
Date: 10:16:44 05/01/99
Go up one level in this thread
On April 30, 1999 at 19:02:14, José de Jesús García Ruvalcaba wrote: >On April 30, 1999 at 18:54:24, Chuck wrote: > >>On April 30, 1999 at 18:37:49, José de Jesús García Ruvalcaba wrote: >> >>>On April 30, 1999 at 18:31:16, James T. Walker wrote: >>> >>>>On April 30, 1999 at 12:03:21, blass uri wrote: >>>> >>>>> >>>>>On April 30, 1999 at 11:37:23, James T. Walker wrote: >>>>> >>>>>>Hello, >>>>>>Food for thought. Which programs are considered by most people as simply fast >>>>>>searchers and which are Knowledge based? >>>>>>Examples? >>>>>>"Fast searchers" >>>>>>Fritz >>>>>>Junior >>>>>>Nimzo >>>>>> >>>>>>"Knowledge based" >>>>>>? >>>>>> >>>>>>I think many claim to be in between. CM6K,Hiarcs,MchessPro? >>>>>> >>>>>>I would like some opinions and why. >>>>>>Jim Walker >>>>> >>>>>I think that no program of today should be called knowledge based. >>>>> >>>>>I looked at the evaluation of hiarcs7 and chessmaster6000 in one game that they >>>>>drew. >>>>>Chessmaster had KBPP against KR but could not make a progress. >>>>>Both programs evaluated the position as a clear advantage for the KBPP. >>>>> >>>>>It is clear to humans that the position is a draw not because of some static >>>>>evaluation function but because the pawns can do no progress. >>>>> >>>>>A program that cannot understand thess simple things that humans understand is >>>>>not deserved to be called a knowledge based program. >>>>> >>>>>Uri >>>> >>>>Hello Uri, >>>>I'm not sure I follow your logic. You seem to be saying based on one position >>>>again that a program can not be knowledge based if it does not understand this >>>>one position. What about all the other positions that it does understand? >>>>Also, you seem to be putting Hiarcs in the "Knowledge based" program category >>>>and then saying that it can't be knowledge based because of the one position. >>>>Why did you pick Hiarcs to begin with. Some people may say that it is only one >>>>of the "Middle of the road" programs that is a mixture of knowledge and search >>>>speed. But what about a program that searches only 200 nps but plays 2100 level >>>>chess? Is this probably a knowledge based engine? What if it still can't >>>>understand the one position you mention? The speed of 200 nps would not qualify >>>>it as a "Fast searcher". What if you take this same program and increase it's >>>>speed to get 6000nps and now it's playing 2600 chess? What will make it a >>>>"Knowledge based" program? Suppose it still cannot solve your specific position >>>>then what? >>>>Jim Walker >>> >>> I do not know exactly what is "knowledge based". It seems to me that you >>>consider "fast searchers" not knowledge based (please correct me if I am >>>misunderstanding your statements, I do not want to put words in your mouth). I >>>think a fast searcher can have lots of knowledge (not in the evaluation >>>function, of course). Still I do not know if they qualify as "knowledge based". >>>José. >> >>I think Jim is trying to make a distinction - fast searching programs just look >>at tactics and thus reach high numbers of nodes/second. For "knowledge-based", I >>don't think Jim is using the term in the purist AI sense but to mean that the >>program derives its strength as much or more from chess knowledge like attacking >>weak squares than from merely searching deep lines of moves and picking the one >>that has the best score. I think it is easy to see by watching the programs >>play, tactical programs play "weird-looking" sequences of moves, the >>knowledge-based ones play more logical sequences. What I mean is that you should >>be able to take a group of several moves from a game and see an overall purpose >>to those moves. Some won't go with the group, guarding against a tactic or >>responding to a capture, but overall, a theme should reveal itself, attack >>against the King, a buildup against an isolated pawn, etc. Don't look at one >>move and make a judgement, look at sequences of moves. >> >>Chuck > > Still, I can easily imagine a fast searcher doing good sequences of moves that >pursue a goal (as long as the rewards of this goal are within the computer's >horizon). I think it is not easy to tell if the good sequence of moves is due to >the deep search or due to some other factor. > I think the fast searchers can have knowledge in their search policy, that may >allow them to look deeper at the more relevant variations, and reject poor moves >after a shallow search. > I do not think that a fast searcher that looks at all the variations at the >same depth (i.e. does not have knowledge in the search policy) can compete >succesfully against strong engines on similar hardware. >José. It seems that there is more than one kind of "Knowledge" to consider. In one you talk about knowledge in the search policy. I guess meaning when to extend and how far to extend, things like that. I was considering more the type of knowledge that is basic to chess. The more basic knowledge of "The opposition" as an example. It seems to me that a program with the knowledge of the opposition can in 1 ply decide to take the opposition and "Know" it can save the 1/2 point. A program without this knowledge may have to search 15 plies to see that making this same move saves the 1/2 point. Then with proper extensions it could do this very quickly. I'm wondering how can you classify a program as being one or the other. Seems like a very tedious balancing act by programers to decide what "Knowledge" to install in the programs and what knowledge the program can just find out through search. Jim
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.