Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: DIEP in WMCCC2000 London - long story

Author: Robert Hyatt

Date: 08:18:31 08/30/00

Go up one level in this thread


On August 30, 2000 at 05:24:22, Vincent Diepeveen wrote:

>On August 29, 2000 at 22:06:33, Robert Hyatt wrote:
>
>>On August 29, 2000 at 14:26:44, Vincent Diepeveen wrote:
>>
>>>On August 29, 2000 at 14:14:51, Robert Hyatt wrote:
>>>
>>>>On August 29, 2000 at 11:50:00, Vincent Diepeveen wrote:
>>>>
>>>>>On August 29, 2000 at 11:44:24, Uri Blass wrote:
>>>>>
>>>>>>I can say that Fritz did not get a winning position out of book against Crafty.
>>>>>>
>>>>>>I found that Crafty17.11 can find 21.Nxe6 against Fritz with more time.
>>>>>>[D]1r2kb1r/2qb1p2/p3pP2/1pp5/3NP2p/P1N2Q2/2P4P/1K1R1BR1 w k - 0 1
>>>>>>
>>>>>>
>>>>>>Unfortunately Crafty did not have good alpha that it probably needed in order to
>>>>>>find this move.
>>>>>>
>>>>>>Uri
>>>>>
>>>>>Oh well Fritz searching 13 ply extending lots of checks and doing
>>>>>checks in qsearch against crafty with crafty a king nearly mated and
>>>>>searching 12 to 13 ply without doing checks in qsearch. Note that
>>>>>fritz has also a lot of threat extensions.
>>>>>
>>>>>Crafty repeated exactly the same line as against Nimzo. Very dumb.
>>>>>
>>>>>So it was dead lost from all sides.
>>>>>
>>>>>Play 100 games crafty here against DIEP, Fritz, Nimzo, Zchess, SOS,
>>>>>Shredder or anything that is doing either a lot of checks or doing
>>>>>checks in qsearch. Bye Bye crafty. Crafty in Najdorf is a zero in advance
>>>>>at icc it basically plays e6 d6 systems. NOT aggressive najdorf systems.
>>>>>
>>>>
>>>>
>>>>I think checks in the q-search is an exaggerated advantage.  I play all of
>>>>the above programs on ICC, all the time, and I am not getting rolled into a
>>>>ball by any of them.n  Didn't you play a bunch of games vs crafty using a
>>>>quad 550 xeon for Diep?  Did those q-search checks make a big difference?
>>>>
>>>>
>>>>
>>>>
>>>>>So selecting this opening is stupid in advance anyway, like i discovered
>>>>>with DIEP in Caro-Kann.
>>>>
>>>>
>>>>I don't think it is a bad opening.  I generally don't play the Sicilian as
>>>>black against computers.  As white, I alternate between d4 and e4 openings
>>>>so black will get a chance to play Sicilians.  At a reasonable time control,
>>>>my quad will find nxe6.  My PII/400 notebook fails low at 5 minutes.  The
>>>>xeon would probably fail low at the 1-1.5 min mark.  It would then use as
>>>>much time as needed and would certainly find Nxe6.  I am surprised that the
>>>>alpha didn't have time to find this, or at least have time to fail low at
>>>>depth-14, which is strange.
>>>
>>>OH dear. How do i start explaining here. I'm talking to a mountain here.
>>>
>>>Najdorf is a very specialistic part of the sicilian where the black king
>>>is facing bunches of checks in the center. Wanting to castle there,
>>>OR being tactical weak there OR falling into a book line means you lose.
>>>
>>>The rest of the sicilians i don't talk about here. We talk here about
>>>crafty single cpu at 300k nps against an opponent at a million NPS.
>>
>>Crafty should have been able to find Nxe6.  Why it didn't, I don't yet know.
>>It seems that it was moving far faster than it should have.  I can't reproduce
>>the fast moves here, so it was something that happened in London.  I may have
>>given Graham a bogus command that screwed up the time control.  He may have
>>entered a time control wrong.  I might have a program bug that asks for seconds
>>but means minutes, somewhere.  Until I know more, I don't have any idea why it
>>would be moving in just over a minute.
>>
>>However, I would take the white position as a human, any day.  Until someone
>>convinces me that black wins no matter what.
>>
>>
>>
>>>
>>>We don't talk about a few blitz games B01 or something at icc.
>>>
>>>We talk about WMCC and about Crafty versus Fritz where crafty
>>>repeats a line which and loses chanceless, even though everyone
>>>in this room felt it wasn't smart to repeat the same B99 mainline
>>>as against Nimzo.
>>>
>>>Still finding Nxe6 doesn't solve the problem for you.
>>>
>>>You lose a few moves further then.
>>
>>I am not convinced of this, yet.  I have seen Fritz make tactical errors.
>>I have seen everyone make them.  I don't think this is as clear as you
>>imply it is...
>
>The only reason you are arguing here is because fritz is not perfect,
>where crafty has the habit to lose somewhere in najdorfs.

It doesn't have a "habit" of losing at all.  I don't fix these kinds of
problems by saying "OK, I just won't play that opening".  I did this for
years with CB, but decided in Crafty to teach it how to play with a bishop
on g7 or g2 in front of a castled king position.  Now rather than worrying
about transpositions into such openings, I ignore it because Crafty can play
those positions just fine.



>
>Note that fritz also made severe error in blitz against diep even though
>i was outsearched by 2 or 3 ply or so. This was not in the mainline
>najdorf but the modern mainline.


What is the point?  I simply don't think q-search checks are a requirement for
a strong program.  And I don't think this particular opening has to be avoided
by programs that don't use them.

Why don't we play a long match playing only that opening from both sides?  If
you win them all I will re-think.  Crafty plays it all the time on ICC, and
isn't getting rolled into a ball all the time.  Otherwise book learning would
have disabled the opening completely.



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.