Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: A Camparison of Fritz6 and Deep Blue1996 Deeper Blue 1997

Author: Vincent Diepeveen

Date: 14:00:07 09/15/01

Go up one level in this thread


On September 15, 2001 at 16:18:42, Uri Blass wrote:

>On September 15, 2001 at 15:56:06, ALI MIRAFZALI wrote:
>
>>Although much has been said about moves like Be4 in game 2 of the second match
>>with Kasparov and h5 in game 5 of the second match because of the great
>>controversy that these moves have generated;the positions that happen
>>after these moves are not a good comparison test for seeing if current
>>commercial programs running on ordinary machines play at or beyond the level
>>of deep Blue.Let me first turn my attention to the first match game 5.At move
>>29 Deep blue plays29.g3? Kasparov says that after 29.Ne2 Rxe2 30.Qxe2 Qa1+
>>31.Nc1 White may be able to hold but the chances are still in Black's favor.
>>Interestingly Fritz6 chooses 29.Ne2 in this position(under 3 minutes 450MHZ).
>>In the same game at move 12 Deep blue plays 12.Rae1 .This is the case of moving
>>the Wrong rook.In the same position Fritz6 plays Rfe1 instantly.There is no
>>doubt that these positions prove that chess Knowelgde is very important for
>>chess programs.Since the Hardware that the deep Blue program was running on
>>was more powerful than anything commercial programs are running on even today
>
>The position is easy for programs with the relevant chess knowledge
>but can be also solved by search for programs without chess knowedge.
>
>I remember from a post of the programmer of yace
>that yace could find 29.Ne2 after some hours by search when
>g3 failed low for kasparov's move.
>
>I believe that the search algorithm of Deep blue were simply
>inferior than the search algorithm of the chess programs of today
>(otherwise some hours of yace could be translated to some minutes
> of the chess programs of today)
>>;the question of the Camparison of a top commercial program like Fritz6 and
>>Deep blue comes down to this question:In a series of games will positions
>>that require chess knowledge for the selection of the best move occur more
>>often then positions requireing computational power? From the 2 matches with
>>Kasparov we can safely deduce that Deep Blue did not have much chess knowledge
>>in Camparison with Todays programs.Now let us see an example for Computational
>>Power.In the same game(game5 first match) at move 32 Deep Blue plays 32.f3
>>Fritz6 (450Mhz) even after ruuning for 20 minuetes plays 32.gxf4 which is worse
>>than f3 because it leads to a faster loss.
>
>I remmeber that Genius3 on p100 could see f3 in a few hours.
>I believe that finding f3 may be dpendent in the
>evaluation function of the program and
>I expect part of the top programs on good hardware to find
>it in few minutes.
>
>Uri

Please explain to me why f3 is better here than gxf4. I get direclty -2.xx
for gxf4 but after f3 you get an endgame which is so simplistically lost
that there is very little chance of ever even drawing it against a
strong player as you have a pawn less!

I don't see why gxf4 would be worse as this involves heavy tactical dogfights
which all are real good for black, but the alternative is a technical
simplistically won position!

Probably Deep Blue evaluated the endgame wrong after f3. Most likely
the border to play f3 nowadays is much tougher than back then. A lost
endgame shredder, diep and many programs evaluate as being -5.xx soon,
whereas old programs didn't come further than -1.x probably.

In that case i bet nowadays commercials would directly play f3.

Only objective search reveals of course at a certain time that gxf4
is that bad, that the alternative, a dead lost endgame with a piece less,
where one can nullmove, is what you do.

At 12 ply the score has dropped to 2.922 up for black for gxf4 still
f2-f3 not found here. But i bet that the same line which made DB decide
to play f3, that DIEP already saw that within a second here.

We do not have the logfiles from deep blue here, so it is kind of useless
to say DB is tactical strong here. I bet it sucked so much everywhere
(compared to todays standards) that it took f3 within a second as
it probably saw
  K+R+5 pawns versus K+R+B+3 pawns. So that's 3 for bishop - 2 for extra
  pawns is 1 pawn down. Add a very small positional score in endgame
  and it's 1.xx down for black there.

Where diep says directly that f3 is over 3 pawns up for black.

At 13 ply -2.857 for gxf4.

Deep Blue simply sucked in endgame that it played f3, that's the
logical conclusion. Nothing else than that.







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.