Deep Thought /WCCC Round 5 vs Fritz.
Robert Hyatt
In the 5th round game, which went as follows:
1. e4 c5 2. Nf3 Nc6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 e5 6. Ndb5 d6
7. Bg5 a6 8. Na3 b5 9. Bxf6 gxf6 10. Nd5 f5 11. Bd3 Be6 12. Qh5
f4 13. O-O
The question was asked about this move for white (13. O-O)
It's not in any of my opening books, so I tried it on Crafty, which
has an opening book made up of well over 50,000 GM games as well as
lots of things like MCO, etc.
Blacks 12. f4 had only one response in Craftys opening book, 13. c3.
However, it was only played in one game, so Crafty would not play it
had it reached this position. I did let it search for a short while,
and it preferred O-O-O from depth=1 until I stopped it. This is
admittedly short analysis, but until f5 for black, there were many
options for black and white. I suspect either (a) black was taken
out of book by 12. Qh5, or else 12. ... f5 was entered into the book
for reasons unknown (perhaps as preparation to counteract Qh5 if
DT had played it in the past.)
I may (for fun) let Crafty "stew" over the first 25 moves or so and
post (or email) the output if anyone is interested. It would be
interesting to see where the game became hopeless (maybe with 13. O-O
from the looks of things).
Bob
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Peter Kappler
>
>In the 5th round game, which went as follows:
>
>1. e4 c5 2. Nf3 Nc6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 e5 6. Ndb5 d6
>7. Bg5 a6 8. Na3 b5 9. Bxf6 gxf6 10. Nd5 f5 11. Bd3 Be6 12. Qh5
>f4 13. O-O
>
>
>The question was asked about this move for white (13. O-O)
>
>It's not in any of my opening books, so I tried it on Crafty, which
>has an opening book made up of well over 50,000 GM games as well as
>lots of things like MCO, etc.
>
>Blacks 12. f4 had only one response in Craftys opening book, 13. c3.
(deletion)...
>
>Bob
>
>--
>Robert Hyatt Computer and Information Sciences
>hy...@cis.uab.edu University of Alabama at Birmingham
>(205) 934-2213 115A Campbell Hall, UAB Station
>(205) 934-5473 FAX Birmingham, AL 35294-1170
I don't have Sveshnikov's book, "The Sicilian Pelikan", here, or I'd just
quote his analysis after 12.Qh5 f4. However, I'm certain that White
should not be castling kingside before Black has placed a bishop on g7.
By castling early, White simply invites Black to do away with Bg7 and
play Rg8 instead, with excellent attacking chances.
I doubt that either program had any clue on move 13 that things would get
dangerous for white.
Crafty's 13.c3 looks like a much better move. I'll check Sveshnikov's
book tonight, if I find otherwise, I'll repost.
Peter Kappler (USCF 2140)
Robert Hyatt
Note that 13. c3 is *not* "crafty's" move. It was played in exactly one
of the GM games that makes up crafty's opening book. Sorry, but I can not
attribute the source yet. Crafty sucks in PGN, but tosses out the who/what/
when/where/etc stuff at present. I plan on putting this in a second file
that would be used when someone playing crafty could say something like
"book?" and get the *entire* book line with citation. In the case of
multiple choices along the way, it would probably simply revert to the
most popular path, or else wait until past the branch before trying to
produce the citation.
In any case, since the move 13. c3 was only played once, Crafty is
"suspicious" of the move and won't play it. This solves a problem I
had early on in using GM games to make up the opening book. Many of
the GM games have a GM on one side and a patzer on the other. If you
don't weed out the "chaff" you get killed. Some of you ICC'ers will
remember a month or two back when I started using the current book
approach that we immediately saw games where crafty played 1. g4, 1. b4,
and the like! I waded through the games and found some really odd things
played against Tal, Petrosian, etc. This "fix" has stopped *most* of
that.
From my brief analysis, Crafty would play 13. O-O-O in this position. It
does that because of the previously discussed "look left" and "look right"
algorithm that decides that queenside looks safer (not difficult in this
position.) I find that this particular problem is a difficult one to
handle. If you don't have a bonus for castling, the program will likely
wait until it's too late, if you do, it will likely castle into an attack.
It'll be solved one day of course.
BTW, I have gone through everything I have now, and I can't find this
line anywhere. Of course, I don't have any "specific" Sicilian books,
but was surprised to not find it in older MCO's. Unfortunately the older
MCO's I have are all English descriptive, yet I made the mental conversion
to algebraic about 15 years ago and can "barely" stumble through the
descriptive. I have real problems with things like "KN-Qb3" as I don't
remember which was the KN. In any case, if I had a dime for every game
Crafty has lost in such positions due to its not understanding long-term
attacking possibilities, I would go buy a Corvette ZR-1. (maybe... :^) )
Robert Hyatt
Fabian Maeser <mae...@inf.ethz.ch> wrote:
>In article <3qkfv5$f...@pelham.cis.uab.edu> Robert Hyatt,
>hy...@willis.cis.uab.edu writes:
>>>I don't have Sveshnikov's book, "The Sicilian Pelikan", here, or I'd just
>>>quote his analysis after 12.Qh5 f4. However, I'm certain that White
>>>should not be castling kingside before Black has placed a bishop on g7.
>>>By castling early, White simply invites Black to do away with Bg7 and
>>>play Rg8 instead, with excellent attacking chances.
>
>I think I remember a game Short-Sax (match?) where this move might have
>been played for the first time. It went something like 13.g3 Rg8 and
>ended in a draw.
>
>The move 12...Rg8 is much more popular (than 12...f4) - it somewhat saved
>the variation's reputation as 12...Bg7 13.0-0 f4 14.c4 is considered
>favourable to White. Top GM's like Shirov and Salov played Rg8 on several
>occasions. The point about the 13.0-0 move in the Deep Blue - Fritz game
>is that it is only a transposition to this line. I can't remember any (GM)
>game going 12...Rg8 13.0-0?, but in comments to 12...Rg8 (Informant) it is
>stated that 13.0-0? f4! gives Black a dangerous attack and a clear
>advantage. ^^^
>
>I never had the ocaasion to test this line in one of my own tournament
>games,
>but in a few blitz games I've played the Black attack soon decided the
>game. Usually the white queen has to go back to d1 (like in the DB-Fritz
>game)
>and Black strengthens the attack by bringing in his queen (to h4 or g5)
>and
>knight (d4) and of course his f-pawn. BTW, has anyone found a win for
>Black
>yet after 16.f3? (Deep Blue played 16.c4 allowing the f-pawn to f3...)
>What
>do you say, Crafty?!
>
First, Crafty is *not* going to see something Deep Thought did not. At least
not in my lifetime, unless it had a Cray to use. However, the difference in
positional knowledge might well make crafty play better moves positionally, if
only these moves don't lose to some deep tactic that it can't see. Personally,
I believe from what I saw that white was lost after castling. If DT couldn't
see the roof falling in at that point with its sophisticated search, then no
program of mine (or anybody else's either...) is going to see anything. That
said, O-O probably does not appear bad to a *very* deep search, so then it
boils down to chess knowledge and positional understanding of the attacking
potential for the three possible king positions (O-O, O-O-O, and "sit") As
you can see, there *is* hope for knowledge playing a part, regardless of how deep
the engine can search.
Fabian Maeser
Well, I think the reason of Deep Blue's unfortunate opening was that
13.0-0?
has never been played by GM's, so it didn't make it to the opening book,
although it is known (from annotations by GM's) to be a bad move.
Fabian
___/ ___/ _ \ / Fabian Maeser, ETH Zurich, Switzerland
/ / / _/ / e-mail: mae...@inf.ethz.ch
__/ __/ _ \ / http://nobi.ethz.ch/febi/febi.html
/ / / / / "It's all in a days work for...
_/ ___/ ____/ __/ BICYCLE REPAIR MAN!"
Tim Mirabile
>In the 5th round game, which went as follows:
>1. e4 c5 2. Nf3 Nc6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 e5 6. Ndb5 d6
>7. Bg5 a6 8. Na3 b5 9. Bxf6 gxf6 10. Nd5 f5 11. Bd3 Be6 12. Qh5
>f4 13. O-O
>The question was asked about this move for white (13. O-O)
>It's not in any of my opening books, so I tried it on Crafty, which
>has an opening book made up of well over 50,000 GM games as well as
>lots of things like MCO, etc.
>Blacks 12. f4 had only one response in Craftys opening book, 13. c3.
>However, it was only played in one game, so Crafty would not play it
>had it reached this position. I did let it search for a short while,
>and it preferred O-O-O from depth=1 until I stopped it. This is
>admittedly short analysis, but until f5 for black, there were many
>options for black and white. I suspect either (a) black was taken
>out of book by 12. Qh5, or else 12. ... f5 was entered into the book
>for reasons unknown (perhaps as preparation to counteract Qh5 if
>DT had played it in the past.)
Sveshnikov, in his book, _The Sicilian Pelikan_, gives the following:
"12...f4?! is premature because of 13.g3!; for example: 13...f3 14.c3 Bg7
15.Nc2 Ne7 16.Nce3 Nd5 17.ed5 Bd7 18.Qf3 and Black does not have compenastion
for the pawn (Blatny-Radoicevic, Sombor 1966)."
He also gives, after 12.Qh5 Rg8:
"13.O-O is dangerous because of 13...f4! [...] for example 14.h3 h6?!
(clearly stronger is 14...Rg6!) [...] (Bilunova-Gaprindashvili, Yaitse
1981)
Perhaps (a) is true, and Fritz3 played 12...f4 on it's own.
If anyone has Fritz3, perhaps they could enter the first 12 moves up to
12.Qh5, put it on infinite mode, turn off the opening book, watch the
analysis and see if 12...f4 is considered. Remember to let it think a
lot longer if you don't have a Pentium.
--
Tim
Tim Mirabile
>>In the 5th round game, which went as follows:
>>1. e4 c5 2. Nf3 Nc6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 e5 6. Ndb5 d6
>>7. Bg5 a6 8. Na3 b5 9. Bxf6 gxf6 10. Nd5 f5 11. Bd3 Be6 12. Qh5
>>f4 13. O-O
>>The question was asked about this move for white (13. O-O)
>>It's not in any of my opening books, so I tried it on Crafty, which
>>has an opening book made up of well over 50,000 GM games as well as
>>lots of things like MCO, etc.
I scanned my database (500000 games), and found:
(6) 12.Qh5 Rg8 13.O-O f4 Black won 5 and lost 1
(6) 12.Qh5 f4 13.g3 Rg8 14.gf4 Bg4 White won all 6 - most lines involve a queen
sac by White.
(4) 12.Qh5 f4 13.g3 Rg8 14.c3 Bg4 Black won one, White won one, two draws
(1) 12.Qh5 f4 13.c3 Rg8 14.g3 Bg4 Short-Sax, St John, 1988 - informant 45/189.
Sax gives 12...f4! N
(1) 12.Qh5 f4 13.g3 Rg8 14.Qh7 Black managed to win after 14...Rg6 15.Qh4 Qh4 16.gh4
(1) 12.Qh5 f4 13.g3 Bg7 Not particularly relevant
--
Tim
Vincent Diepeveen
>First, Crafty is *not* going to see something Deep Thought did not. At least
>not in my lifetime, unless it had a Cray to use. However, the difference in
>positional knowledge might well make crafty play better moves positionally, if
>only these moves don't lose to some deep tactic that it can't see. Personally,
>I believe from what I saw that white was lost after castling. If DT couldn't
>see the roof falling in at that point with its sophisticated search, then no
>program of mine (or anybody else's either...) is going to see anything. That
>said, O-O probably does not appear bad to a *very* deep search, so then it
>boils down to chess knowledge and positional understanding of the attacking
>potential for the three possible king positions (O-O, O-O-O, and "sit") As
>you can see, there *is* hope for knowledge playing a part, regardless of how deep
>the engine can search.
King safety in evaluating procedure is, i guess, the most difficult thing
to programm; you can never have a 100% good kingsafety evaluation,
even not 99%. If you increase it, then too many pieces are placed
with king, if you decrease it, then the programm will go mate...
Didn't these games on the World championship show that a computer-chess-
programm looses when it plays worse moves than its opponent programm does,
and doesn't win because it plays better moves?
Fritz % is climbing, assuming it has a better openingsbook...
It is gonna be a hard time against Kasparov 1996. I put my money on
Kasparov, especially when he begins to play with 1.a2-a4??, leading to
a lost position for white, but a won position against computers... ^8
Now serious: if a gap in openingsbook is almost surely loosing against
a micro-programm, then how does a computer want to win from Kasparov,
when you are sure that Kasparov gets hisself a won openingsposition?
The only problem for Kasparov is to win a won position. As soon as he has
won one game, then he surely wins the match: making draws is very easy
against a certain machine; it cannot even detect it at a certain depth?
Vincent.
vdie...@cs.ruu.nl
Vincent Diepeveen
Manual to translate the Swedish rating of Fritz?
...add 500 computerelopoints to Fritz,
when playing for the world title (openingsbook, you know?????),
add 300 c.e.points when handled by a Fritz-expert,
decrease c.e.p. 300 when handled by a Swedish guy,
decrease c.e.p. 1500, when playing humans... ^8
Swedish-Rating List(or something) represents a fair rating,
based on old hardware, SURPRISING RESULTS (Pentium 586 - 386 notebook),
and EQUAL CHANCES: old openingsbook - new openingsbook (known to crush that
old everywhere sold old-openingsbook).
--
+--------------------------------------+
|| email : vdie...@cs.ruu.nl ||
|| Vincent Diepeveen ||
+======================================+