Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Questions to be answered in the WMCCC.

Author: Uri Blass

Date: 09:38:29 08/28/00

Go up one level in this thread


On August 28, 2000 at 10:11:27, Vincent Diepeveen wrote:

>On August 27, 2000 at 17:58:14, Uri Blass wrote:
>
>>On August 27, 2000 at 17:25:55, Marcus Kaestner wrote:
>>
>>>>>Which program had the best opening book? Shredder 5.0
>>>
>>>you must be kidding!
>>>shredder´s book was so very bad that shredder was named there "houdini-chess"
>>>because first he has had to untie himself.
>>>
>>>sorry, but it was one of the worst one´s.
>>
>>A good book is not a book that leads you to better positions but a book that
>>lead you to better result.
>>
>>I read that the 1.e4 c6 2.d4 d5 3.f3 line shredder has more than 90% result and
>>if it is the case  this line was a very good line.
>>
>>I read that the line against Sos was also good.
>>
>>I do not say that book of shredder was the best but I disagree that it was one
>>of the worst
>>
>>
>>>alex´book was BY FAR the best and there is NO doubt about that.
>>>having killer line´s but also with hardly a whole.
>>>
>>>>>Which program had the worst opening book?
>>>
>>>here you can say shredder.
>>>
>>>>>Which program had the best middlegame? Junior
>>>
>>>i prefer fritz
>>>
>>I was impressed by the way that Junior played the middle game against
>>Shredder,Rebel and tiger.
>
>I was not very impressed by Junior in the middlegame, it had everywhere
>a 0.00 score as if it's completely ignoring the world. It looks to me
>that juniors search is getting outdated, but that the bad way in which
>most programs attack and bad finishing dominated, like DIEP-JUNIOR could not
>get decided by DIEP in the middlegame.
>
>For sure Junior defended very well, but usual after doing some real stupid
>moves as it hardly got a worse middlegame position.
>
>For example the DIEP-Junior game, here DIEP predicted dxe5 instead of
>fxe5?, after fxe5 junior was in big trouble. Junior obviously didn't
>see this because it doesn't evaluate pins like DIEP does.

It probably could find dxe5 with faster hardware.

Junior understood after fxe5 fxe5 that it cannot play Bg7.
Seeing more 2 captures is only one ply for Junior so I guess that it could see
it with faster hardware and avoid the problem.

>
>>Fritz could get nothing in the middle game against Junior so I prefer
>>Junior(london) in the middle game.
>
>I thought fritz improved a lot, but it seems basically improved
>for blitz to me. It sure isn't a preprocessor anymore, but the
>knowledge inside its evaluation is stil limiting its great search.
>
>>>>>Which program had the worst middlegame?
>>>>>Which program was the best in the endgame? Shredder 5.0
>>>
>>>yes, but even close rebel and tiger.
>>
>>>>>Which program was the worst in the endgame?
>>>>>Which program is best in closed positions? Rebel
>>>
>>>ok
>>>
>>>>>Which program is best in noticing attacks and attacking? Shredder 5.0
>>>
>>>i would prefer junior.
>>>
>>>>>Which program is best in defense? Shredder 5.0
>>>
>>>no doubt.
>>>
>>>>>Which program is best in playing human players? Shredder 5.0
>>>
>>>no, no, no. much better is rebel.
>>>
>>>>>Finally, which program is best overall? Shedder 5.0
>>>
>>>not so very clear. rebel has the best evaluation, tiger the best search and
>>>fritz the best combination of these two points. bt for me it is clear, that
>>>these four programs are dominating at the moment.
>>
>>It is not clear.
>
>>Junior was the most interesting program in the event.
>>I think that this program did the biggest change relative to previous versions.
>
>I don't know what Amir changed to junior, but it sure looked drastically
>to me.
>
>When talking about number of lines of source code changes, i'm sure that
>i changed more at DIEP as i changed really a lot in DIEP. This resulted
>however in bad performance, so probably no one cares (as people kick
>on results), but it's looking real good for DIEP's future.
>
>>Sometimes the change was good(see the games against Rebel and tiger)
>>Sometimes the change caused trouble when Junior played h5 against shredder or
>>let nimzo to win it by sacrificing the exchange.
>
>Nimzo is always overestimating passed pawns, so it was no big deal that
>nimzo found this sacrafice. In Nimzo-The King the same thing happened
>but with opposite results. In Nimzo-Shredder , there nimzo had a won
>position but couldn't profit for it as it feared the passed pawns
>too much. Against junior nimzo was just lucky in my eyes, even though
>i predicted during the game that nimzo would perhaps win it with an
>exchange sacrafice. I just said this because i find juniors search for
>the future really outdated. pruning interesting moves with 3 plies or
>something similar is really not gonna improve it for the future. It
>will always miss logical human continuations in the future. It has
>a certain level and without changing its search its gonna get real hard.
>
>>I know that Rebel is a preprocessor so I doubt if it has the best evaluation.
>
>I agree here completely. Rebel sure isn't bad compared to most, but
>it's really getting a bit old now.
>
>>I do not like preprocessors for analysis.
>
>Idem
>
>>I may analyze a position when the computer say that white have an advantage when
>>after castling the computer suddenly say that black has an advantage when it
>>cannot see deeper.
>>
>>I know that not all the programs behave like this.
>
>Oh well another zillion things go wrong with preprocessors, the
>big ADVANTAGE is however that computerchess is hard, so making a good
>non-preprocessor is real hard, where making a preprocessing kind of
>thing is very easy to debug and it finds shots fast.
>
>>Amir admitted that there are cases when Junior is a preprocessor but Amir hope
>>to get rid of it.
>
>Oh well, Amir will face a real big problem then. He searched
>this championship 17 ply. He searched also 17 ply at a quad xeon in
>paderborn. So obviously he has not slowed down Junior.

Depth 17 of Junior does not mean 17 plies.
It means only 9 plies if you talk about brute force search and it means more
than 9 plies

Junior does not do a lot of preprocessing.
It may be a processor of something but usually not of the root.

I usually do not get situations when Junior shows draw evaluation and after
castling or trading queen it shows a clear advantage(I did not test Junior6 but
only previous versions but I do not think that Junior6 is worse).

Junior can learn from previous search and usually have the same evaluation if
the opponent does the expected move.

I know that part of the other programs do not show this behaviour.

Uri



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.