Author: Robert Hyatt
Date: 19:58:38 09/13/01
Go up one level in this thread
On September 13, 2001 at 20:53:21, Vincent Diepeveen wrote:
>On September 13, 2001 at 17:59:44, Uri Blass wrote:
>
>>On September 13, 2001 at 17:26:21, K. Burcham wrote:
>>
>>>[Event "IBM Kasparov vs. Deep Blue Rematch"]
>>>[Site "New York, NY USA"]
>>>[Date "1997.05.04"]
>>>[Round "2"]
>>>[White "Deep Blue"]
>>>[Black "Kasparov, Garry"]
>>>[Opening "Ruy Lopez: closed, Smyslov defense"]
>>>[ECO "C93"]
>>>[Result "1-0"]
>>>
>>>1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.O-O Be7 6.Re1 b5 7.Bb3 d6
>>>8.c3 O-O 9.h3 h6 10.d4 Re8 11.Nbd2 Bf8 12.Nf1 Bd7 13.Ng3 Na5 14.Bc2 c5
>>>15.b3 Nc6 16.d5 Ne7 17.Be3 Ng6 18.Qd2 Nh7 19.a4 Nh4 20.Nxh4 Qxh4
>>>21.Qe2 Qd8 22.b4 Qc7 23.Rec1 c4 24.Ra3 Rec8 25.Rca1 Qd8 26.f4 Nf6
>>>27.fxe5 dxe5 28.Qf1 Ne8 29.Qf2 Nd6 30.Bb6 Qe8 31.R3a2 Be7 32.Bc5 Bf8
>>>33.Nf5 Bxf5 34.exf5 f6 35.Bxd6 Bxd6 36.axb5 axb5 37.Be4 Rxa2
>>>38.Qxa2 Qd7 39.Qa7 Rc7 40.Qb6 Rb7 41.Ra8+ Kf7 42.Qa6 Qc7 43.Qc6 Qb6+
>>>44.Kf1 Rb8 45.Ra6 1-0
>>>
>>>last book move for shredder 5 was 19. a4.
>>>so from move 20 to final white move of 45. Ra6, deep blue had 25 moves.
>>>of the 25 deep blue moves, shredder5 chose 20. 80% of deep blue moves chosen.
>>>GT.2 was almost identical to this.
>>>
>>>of the 24 GM moves , after book, shredder5 chose 14 moves. for 58% of GM
>>> moves.
>>>
>>>two significant moves 26.f4 and 37. Be4.
>>>todays programs can find both of these moves in tournament time.
>>>
>>>OF COURSE THE MOST SIGNIFICANT MOVE. 45...Qe3.
>>>with this move kasparov had several combinations to draw.
>>>but instead kasparov resigned. he did not see the draw
>>> that todays programs are very capable of playing.
>>>
>>>so the following is some of the commentary during the game.
>>>you can see these commentators did not anticipate the possible
>>>draw with 45...Qe3, or the resign from kasparov.
>>>also you will notice kasparov had 58 minutes remaining.
>>>
>>>Patrick Wolff wants to play 45...Ra8.
>>>kasparov wants to resign. which he did.
>>>here is the final position.
>>>
>>> [D] 1r6/5kp1/RqQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 b - -
>>>
>>>
>>>
>>>
>>>MAURICE ASHLEY: As we take a look at the clock it seems like Kasparov has more
>>>time than the computer does, than Deep Blue does because he has 58 minutes
>>>remaining and Deep Blue has 23 minutes remaining. That is in fact not true. The
>>>one at the bottom of the screen, the /TK-P side shows it's still in that first
>>>time control, while Garry is in the second time control, you see the two. So
>>>Garry has 58 moves remaining until move 60 while Deep Blue has in fact an hour
>>>and 23 minutes remaining on its clock. And that would be -- that will be
>>>revealed as soon as it passes the 00 mark. So plenty of time for both sides. And
>>>the game -- it's heating up. I mean a mate threat has occurred. And that
>>>suggests certain actions, definite things are beginning to occur. Garry has to
>>>be very precise in his decision-making or else he could easily mind himself just
>>>completely lost instantly
>>>
>>>
>>>
>>>PATRICK WOLFF: My point being, if we go all the way down here. I want to make
>>>two points. First of all, this is Kasparov's last line of defense. The last line
>>>of defense is what we call perpetual check. Continually checking the opponent's
>>>king. There's no danger of checkmate. That's why it's call perpetual check, but
>>>it goes on forever, it's perpetual, and because it's continually check, check,
>>>check, check, check, check, check, the game could be a draw. Whether or not this
>>>is perpetual check requires very, very precise analysis.
>>>
>>>YASSER SEIRAWAN: d5xc6, absolutely.
>>>
>>> YASSER SEIRAWAN: I think that the ending, when the pawn is on d5, the ending
>>>with the pawn on d5, black has excellent holding chances. With the pawn on c6, I
>>>think he's gone. This is a very bad pawn. Because with a pawn on c6, white's
>>>bishop will be able to go to the d5 square check. The e6 square, and then when
>>>he moves Ra7 and c7 he will just win a piece and the game. So after this what
>>>appears to be a forced exchange of queens, I don't see how Garry is going to
>>>defend.
>>>
>>>PATRICK WOLFF: Exactly. Let's take a look at Rc8. Now, the idea of Rc8 is white
>>>was threatening this check. Now, if that check should happen, black can block it
>>>with the rook. But I think we'll see that there's still a lot of trouble for
>>>black here. Ra7+ I think is still very strong. Now, Bc7 is very bad because
>>>white plays Rb7. Black cannot defend like that, because that guy goes. So after
>>>rook b7 this b5 pawn is in bad shape. Therefore, after Ra7+, Rc7 is pretty much
>>>forced. But now we play Ra8.
>>>
>>>YASSER SEIRAWAN: Ra5.
>>>
>>>PATRICK WOLFF: Also Ra5, good point. Itself going to -- I was going to suggest
>>>that black would be tied up with Ra8, but that would be bad enough, but
>>>
>>>YASSER is correct.
>>>
>>>MAURICE ASHLEY: And looking at Garry Kasparov now, he does not look like a happy
>>>man. He has all these variations --
>>>
>>>{{as you can see by the above conversation, Yasser did not anticipate the
>>> resign in this position. and also not only did kasporov miss the
>>> 45...Qe3 for the draw, Yasser missed the draw also.}}
>>>
>>>
>>>this is one of the drawn positions, that kasparov could have played into
>>> with, 45...Qe3. instead as you can read Mr. Ashleys words,
>>> kasparov resigned the position.
>>>
>>> [D] 4r3/5kp1/R2Q1p2/1p1Pp2p/1Pp4q/5B2/5KP1/8 w - -
>>>
>>>
>>>MAURICE ASHLEY: Garry Kasparov has resigned the position. (Audience applause.)
>>>(Long and sustained audience applause.)
>>>
>>>MAURICE ASHLEY: We were indeed impressed by the way the computer handled this
>>>position. We hope, we'll cross our fingers, but don't count on it that Garry
>>>will come on stage and explain why he lost. I don't think so.
>>>
>>> MIKE VALVO: Several Grandmasters were astounded by this game. It's clearly the
>>>best game ever played against Kasparov by a computer.
>>>
>>>YASSER SEIRAWAN: The question was, during the game, was there ever a negative
>>>valuation. In other words, did Deep Blue ever feel itself to have a worse
>>>position at any time, and what was the final position. JOEL BENJAMIN: Short
>>>answer, no, it always felt that it had a slight advantage, until near the end
>>>when it felt it had a larger advantage, and then the score in the last few moves
>>>went very high and it was something like two pawns up.
>>>
>>>
>>>Question for Dr Hyatt: in the final position, Joel Benjamin said deep blue
>>> eval was close to +2.00. shredder5 finds 45...Qe3 in 30 seconds with
>>> 1500 mhz. eval starts out at +1.88, with shredder5, and drops quickly.
>>>in 3 minutes shredder5 eval drops to +1.40. in 15 minutes shredder5 eval
>>>drops to +1.26. i do not understand how deep blue eval can be +2.00.
>>>deep fritz and GT2.0 are similiar in this 45...Qe3 eval. if deep blue
>>>was so far ahead of these rpograms in kns, it seems in this final position
>>>that deep blue eval should be somewhat lower than +2.00 for 45...Qe3,
>>>you would think in this position deep blue could see the drawn position
>>> of 45...Qe3. and if you say it could see the draw, then why did Joel
>>> say eval was close to +2.00.
>>
>>My opinion is that deeper blue simply had inferior evaluation in that position.
>>
>>If you try older programs like Genius3 and even Junior6 then they also cannot
>>see Qe3 in a reasonable time.
>>
>>Hyatt claimed that no program including deeper blue can see the draw.
>>I agree that the top programs do not see 0.00 score for Qe3 and the score is
>>dependent on evaluation.
>>
>>The question if to evaluate Qe3 as +2.xx or +1.xx is also dependent in the value
>>that chess programs give to pawns and if you reduce the value of pawns you are
>>going to find that finding Qe3 with better score than 2 pawns against black is
>>hard task for chess programs.
>
>Diep sees Qe3 as about draw. Not a 0.00 but very close to that and
>it sure isn't eval only, it's basically caused by seeing the critical
>lines very deeply. At 14/15 ply DIEP sees of course more than Deep blue
>when talking about tactics. That's it i guess. Diep versions that use it
>also have singular extensions just like DB. I'm not amazed that many programs
>who don't have SE do not find this as a draw. Any program with SE implemented
>more or less bugfree and which do checks in qsearch or extend them some
>near qsearch, those should find Qe3 to be near draw score.
>
>Of course this means that DIEP doesn't play Kf1?? a few moves before this.
>Diep plays Kh1! there of course which wins as Seirawan indicated in his
>analysis AFTER the game, as published in ICCA journal june 1997.
>
>I wouldn't be amazed if many of todays programs still do not find Kh1,
>and that these diep versions are an exception to it (when i throw out SE
>of course the near to draw score goes away for Qe3 but it still
>plays Qe3 of course, where perhaps Kh1 gets done because of
>positional reasons though).
>
>I'm not amazed that DB didn't find all this. In log files you can see
>that deep blue doesn't search much deeper in this position as what it does
>in complex middlegame positions, whereas because of transpositiontables all
>software programs get very deep here (and very shallow in those complex
>middlegame positions). The extra search depth here sees for DIEP the draw
>here then.
>
>That some lines are nearly 60 plies are not nonsense, but not really
>true also because the few search lines to see that Qe3 is a draw
>which are so deep that are search lines which are completely forced.
>
>In 1997 i figured out that a program without dangerous extensions would
>need at most 22 ply to see Qe3 to be a draw.
>
>>Uri
I would be willing to bet any amount of money you want to bet that you have
_not_ implemented singular extensions as they were implemented in Deep Blue.
Not even close. If you want, I will ask some key questions about specific
issues that the DB guys addressed and let you post code from Diep to show how
it handles 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.