Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Deep Blue's 8.Nxe6 in Game 6 a forced win?

Author: Robert Hyatt

Date: 09:22:08 09/22/03

Go up one level in this thread


On September 22, 2003 at 10:31:53, Uri Blass wrote:

>On September 22, 2003 at 08:46:18, Vincent Diepeveen wrote:
>
>>On September 21, 2003 at 22:10:52, Dave Gomboc wrote:
>>
>>>On September 21, 2003 at 20:04:21, Vincent Diepeveen wrote:
>>>
>>>>On September 20, 2003 at 16:12:18, Dave Gomboc wrote:
>>>>
>>>>no way they could predict he would go caro-kann, so the statistical chance in a
>>>>4000 move book this line was there till nxe6 is like zero.
>>>
>>>The GMs doing DB's opening tuning had that position up and running on DB --
>>>briefly -- as they were booking it up.  It could be that they saw it would play
>>>Nxe6 out of the extended book, saw that it liked its position as White there,
>>>and left it alone.  Whether it was in the small book or the extended book, the
>>>important thing is that DB was comfortable playing Nxe6 on its own, and they let
>>>it do so.
>>>
>>>Dave
>>
>>That's just another BS marketing story which is not true.
>>
>>Just like the marketing story that in a 12 stone double rook endgame deep blue
>>managed to keep a draw because every move was perfectly played thanks to an
>>incredible big endgamedatabase.
>>
>>In fact marketing department said something even more overreacted, 30 seconds
>>after the game had ended in a draw.
>>
>>It's here just like that too.
>>
>>Those GMs didn't even have the time to prepare caro-kann further than move 5,
>>let me assure you that.
>>
>>4000 bookmoves that's something very very tiny!
>>
>>Also you enter 4000 bookmoves eyes closed in 8 days.
>>
>>Kure told he usually enters around 2000 bookmoves an hour.
>
>If your claims is correct then it means that he cannot have a good book.
>You cannot be sure of no errors with 2000 moves per hour.
>
>There are a lot of lines that you do not know how to evaluate
>and you need to analyze them before deciding if to put them or not to put them
>in a manually generated book.

Uri...

You and I have had _many_ discussions.  From them I have concluded that you
are a very bright person.  But when you try to talk reality with Vincent, I
get the _opposite_ opinion.  :)

Of course you can't do 2000 moves an hour.  Unless you have an already-prepared
set of moves and are just typing them in.

Vincent just rambles, and is really an _excellent_ random number generator
since most of his numbers are random, irrelevant, and always inaccurate.

>
>Analyzing them is not only deciding if they are good or bad but also analyzing
>if the specific program knows to play well after them.

Correct.  And this takes time.  I did this for 15 years with Cray Blitz.  I
did good to enter a few _lines_ in an hour, where the lines might have 50-100
moves total, if I was lucky.

>
>
>>
>>So for the experienced openingsbook creators here, 4000 bookmoves is just 2
>>hours.
>>
>>That typically describes the contribution of the GM's to the openingsbook.
>>
>>Hsu always believed in autogenerated books.
>>
>>Only in october 1997 that believe was interrupted rudely by Kure who with a very
>>poor program, managed to score many points, just based uponb ook.
>
>Nonsense.
>
>Kurt played a tournament when both sides use the same book and Fritz scored
>better than programs like ruffian.
>Fritz is not the best program today(shredder is better) but it is not better
>than Ruffian because of opening book.

Vincent is convinced that the opening book is the end-all for computer chess.
It gives him a convenient excuse every year "I got into a killer-book line
prepared against me..."  It is important.  It is not _that_ important.


>
>I also remember that
>Kurt also found that Ruffian does slightly better against the commercial
>programs when both sides use their own books and not when both sides use an
>external book and play both sides of every opening.
>
>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.