Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: What's the Secret to Shredder 7.04 Success?

Author: Tom Kerrigan

Date: 20:14:26 05/03/03

Go up one level in this thread


On May 03, 2003 at 21:58:19, Jim Bond wrote:

>On May 03, 2003 at 20:58:31, Tom Kerrigan wrote:
>
>>On May 03, 2003 at 16:50:31, Jim Bond wrote:
>>
>>>On May 03, 2003 at 15:59:17, Uri Blass wrote:
>>>
>>>>On May 03, 2003 at 15:45:14, Jim Bond wrote:
>>>>
>>>>>On May 03, 2003 at 15:10:50, Uri Blass wrote:
>>>>>
>>>>>>On May 03, 2003 at 14:37:48, Jim Bond wrote:
>>>>>>
>>>>>>>On May 03, 2003 at 14:07:59, Uri Blass wrote:
>>>>>>>
>>>>>>>>On May 03, 2003 at 13:44:27, Jim Bond wrote:
>>>>>>>>
>>>>>>>>>On May 03, 2003 at 10:32:08, George Wilson wrote:
>>>>>>>>>
>>>>>>>>>>Hi
>>>>>>>>>>
>>>>>>>>>>  After playing Shredder 7.04 against Several Strong programs on two separate
>>>>>>>>>>computers I am very "Impressed" . I wonder though what distinguishes this
>>>>>>>>>>program from the other top programs, what gives it the edge in playing strength?
>>>>>>>>>> I think it is shredders fantastic endgame prowness. All six of the games it won
>>>>>>>>>>against century 4 was in the endgame, however rebel seemed to play even with it
>>>>>>>>>>tactically in all the middlegames
>>>>>>>>>
>>>>>>>>>This could be partly due Shredder's ability to probe ending game table base.  I
>>>>>>>>>tend to find that, in its analysis window, the tb numbers are much bigger than
>>>>>>>>>other engines.
>>>>>>>>
>>>>>>>>I disagree
>>>>>>>>Bigger is not better.
>>>>>>>>
>>>>>>>>I did not check your theory but if shredder probes ending tablebases more than
>>>>>>>>other programs then it suggests that shredder has not knowledge that it can
>>>>>>>>trust without tablebases.
>>>>>>>>
>>>>>>>>A program with knowledge is not going to probe tablebases in a lot of tablebases
>>>>>>>>positions because calculating the winner by knowledge is faster.
>>>>>>>>
>>>>>>>>Uri
>>>>>>>
>>>>>>>Calculating the winner by knowledge may be faster but is it more accurate than
>>>>>>>tablebase?  I am afraid not.  The table base is a superset of conventional
>>>>>>>theory or knowledge.  It is an oracle.  Shredder might be going for accuracy as
>>>>>>>oppose to speed.
>>>>>>>
>>>>>>>Jim
>>>>>>
>>>>>>It is not less accurate if you do it only in the right part of the cases.
>>>>>>The right part can be bigger when the program has more knolwedge.
>>>>>>
>>>>>>Uri
>>>>>
>>>>>Not less accurate?  So are you claiming you or someone can program chess
>>>>>knowledge equally or more accurate than the complete 5-men tablebase?  I believe
>>>>>the fact is no body can otherwise there wouldn't be tablebase at all.  People
>>>>>wouldn't have developed it in the first place.  Can anyone cover all the "right
>>>>>parts" as quoted from you?  Have anyone been able to?  I am afraid not.
>>>>>
>>>>>Jim
>>>>
>>>>I do not use tablebases in movei but there are cases that can be evaluated
>>>>correctly without tablebases
>>>>
>>>>I can give you examples:
>>>>
>>>>KQ vs K is always a win if the position is not stalemate and findin if the
>>>>position is stalemate is faster than calling tablebases.
>>>>
>>>>KBP vs K is a draw when the bishop is blind and the king is close enough to the
>>>>corner.
>>>>
>>>>KPP vs KP is a win for the side with the pp if this side is to move and has
>>>>unstoppable pawn  and the distance of the opponent's pawn to be a queen is
>>>>bigger by at least 2.
>>>>
>>>>It is possible to define a lot of rules that will be 100% correct and will cover
>>>>part of the cases.
>>>>
>>>>The main part when they can help is in positions when one side has 3 pieces
>>>>because these cases can be detected as wins in most of the cases with no errors
>>>>in the cases that they are detected as wins.
>>>>
>>>>Uri
>>>
>>>Thanks for sharing the rules.  However I wonder if anyone can program a rule
>>>that can accurately solve this Queens-pawn mate in 63 ending position (126 half
>>>moves).  The program would have to tell me 1, -1 or 1/2 accurately without TB.
>>>
>>>8/4q2K/1k5p/7Q/6P1/8/8/8 w - - 0 81
>>>
>>>My point is that in simpler cases, 100% accuracy it is possible, but in more
>>>complicate cases, without TB, it is not humanly programmable or computer
>>>searchable.  I am only guessing Shredder might be taking advantage of TB to a
>>>greater degree than other programs given TB is the "truth".
>>>
>>>Jim
>>
>>What's your problem?
>>
>>Uri never said that rules are accurate in all cases or that tablebases are
>>unnecessary and should never be probed.
>>
>>Of COURSE you can find cases where rules will not apply. That's when you do a
>>lookup. The fact that you can find these cases completely misses the point.
>>
>>-Tom
>
>You are welcome to adding your opinions.  I guess you want to clarified what Uri
>has been trying to say - more TB probing is bad.  Do you agree?
>
>Jim

Uri has been saying nothing of the sort--he's saying that a probe is not
necessarily good, i.e., when you can get a perfectly accurate rule-based answer
much faster.

-Tom



This page took 0.01 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.