| |
Main
Date: 09 Aug 2007 19:03:25
From: SAT W-7
Subject: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
Thank you ..
They played two games , what pawns were missing from Rybka and what
color did they play.
Did Joel put up a fight ?
|
|
| |
Date: 19 Aug 2007 21:24:04
From: Guy Macon
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
help bot wrote:
>
>David Richerby wrote:
>
>> Fred wrote:
>>
>>> Could you explain the "contempt factor", please?
>>
>> Suppose a chess engine evaluates drawn positions as zero, which seems
>> natural enough. This means that, in a position where the evaluation
>> is, say, -0.02 (against the computer), the computer would accept a
>> draw by repetition if it saw nothing else better than -0.02. In
>> practice, this isn't a good idea as it means that the engine will play
>> for a draw as soon as some random little feature of its evaluation
>> function causes the evaluation to go ever so slightly negative.
>>
>> So, what the engines do is score a draw as something a little below
>> zero; say -0.20. This means that the engine will only play for a draw
>> if it is at a noticeable disadvantage. Of course, the difference
>> between -0.22 (play for a draw if there is one) and -0.19 (play for
>> the win!) is also essentially noise in the evaluation function. But
>> at least the engine can be sure it has some kind of disadvantage at
>> both scores and that this disadvantage isn't overwhelming.
>
> In normal play (no odds), even a score of -0.2 would be a poor
>choice, for many grandmasters would have little trouble obtaining
>this small an advantage, especially as White. For that matter,
>many lesser players (incl. chess programs) could easily luck into
>such an edge during the opening before the real show begins, so
>Rybka would effectively cheat itself by steering for draws here.
>
> As the world's strongest program, it would seem better to use a
>contempt factor of, say, 0.8, except when facing super-GMs or
>programs like ZapChess. Heck, against patzers, a contempt
>factor of 3 or 4 might even do no harm!
On the other hand, Rybka could reason that it being down 2 points
is a pretty good indicator that it was wrong about the opponenet
being a Patzer...
--
Guy Macon
<http://www.guymacon.com/ >
|
| | |
Date: 19 Aug 2007 19:19:25
From: Kenneth Sloan
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
Guy Macon wrote:
>
> On the other hand, Rybka could reason that it being down 2 points
> is a pretty good indicator that it was wrong about the opponenet
> being a Patzer...
>
Or, Rybka could reason that, since the opponent is a patzer it is better
to navigate a path THROUGH objectively lost positions in order to reach
a won position - counting on the opponent to NOT see the chances offered
to him. This might lead to a win, where playing *perfect* moves will
lead only to a draw.
Chess, like many games, is played on many levels. A player who decides
that the game is objectively drawn, and as a result forces play down
sterile, drawn lines because "it's best", will never be highly rated.
Consider the junk heap of abandoned opening lines. Some lines are
abandoned because they are inferior. Others are abandoned because they
are too well analyzed. They may, in some narrow sense, be "optimal"
- but they are not good practical weapons (for the grandmaster).
And then again - they actually ARE good practical weapons for the patzer.
Sam Sloan's opening repertoire seems to be based on a very high
"contempt factor". He enters objectively horrible positions - on the
theory that "there are traps here, and my opponent will surely blunder".
Sometimes, this contempt is justified. Here, for example, his
opponent misses a mate in 5 AND a mate in 4. Still, perhaps the
contempt factor was set just a tad too high...
http://www.cis.uab.edu/sloan/ChessGames/SloanVsSloan/
--
Kenneth Sloan KennethRSloan@gmail.com
Computer and Information Sciences +1-205-932-2213
University of Alabama at Birmingham FAX +1-205-934-5473
Birmingham, AL 35294-1170 http://www.cis.uab.edu/sloan/
|
| |
Date: 19 Aug 2007 06:10:37
From: help bot
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
On Aug 18, 2:13 pm, David Richerby <dav...@chiark.greenend.org.uk >
wrote:
> Fred <F...@google.com> wrote:
> > Could you explain the "contempt factor", please?
>
> Suppose a chess engine evaluates drawn positions as zero, which seems
> natural enough. This means that, in a position where the evaluation
> is, say, -0.02 (against the computer), the computer would accept a
> draw by repetition if it saw nothing else better than -0.02. In
> practice, this isn't a good idea as it means that the engine will play
> for a draw as soon as some random little feature of its evaluation
> function causes the evaluation to go ever so slightly negative.
>
> So, what the engines do is score a draw as something a little below
> zero; say -0.20. This means that the engine will only play for a draw
> if it is at a noticeable disadvantage. Of course, the difference
> between -0.22 (play for a draw if there is one) and -0.19 (play for
> the win!) is also essentially noise in the evaluation function. But
> at least the engine can be sure it has some kind of disadvantage at
> both scores and that this disadvantage isn't overwhelming.
In normal play (no odds), even a score of -0.2 would be a poor
choice, for many grandmasters would have little trouble obtaining
this small an advantage, especially as White. For that matter,
many lesser players (incl. chess programs) could easily luck into
such an edge during the opening before the real show begins, so
Rybka would effectively cheat itself by steering for draws here.
As the world's strongest program, it would seem better to use a
contempt factor of, say, 0.8, except when facing super-GMs or
programs like ZapChess. Heck, against patzers, a contempt
factor of 3 or 4 might even do no harm!
I read something on the Rybka Web site about a version which
they claimed would likely do better in odds games, but I don't
recall the details. There is a Larry Kaufman version, for instance,
which handles things a bit differently. IMO, the way Rybka
handled the pawn-odds games against GM Benjamin was far
from impressive; specifically, the opening phase went the wrong
direction, giving the human GM the types of positions where the
missing pawn was crucial, rather than largely irrelevant. In
essence, the fact that Rybka won the match anyway is a
testament to its superior middle game ability, along with the
fact that when ahead, Rybka did not wimp out and give away
free draws.
When I play over my own games with Fritz 5.32, I often find
that I missed better (by a half-point) moves many, many times
during the course of play, but so did my opponents. Most of
these sub-optimal errors just balance out, but when they don't,
the player who blundered or fell behind on optimality while his
opponent "hit" several best moves in a row simply loses. I
believe that at the lower levels, Rybka could probably save
dead lost positions (provided they are very complex) as well
as winning drawn ones (provided they are very complex), often
as not. Humans make a lot of mistakes, especially when the
time control approaches or when their own King is being
chased around the board.
I'm thinking the Web site mentioned a "gambit" version of
Rybka, designed specifically for the sort of play involved in
the pawn-odds matches. I'm thinking the Rybka team simply
substituted a tweaked openings book for the match with GM
Benjamin, but this strategy essentially failed.
-- help bot
|
| |
Date: 16 Aug 2007 21:27:01
From: help bot
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
On Aug 16, 4:22 pm, Fred <F...@google.com > wrote:
> > Game one: programmers claim problem with contempt factor
> >caused bad moves; GM Benjamin is awarded a win and the
>
> Could you explain the "contempt factor", please?
>
> Thanks.
In a normal game with no odds, the contempt factor tells
the program to "think optimistically", to go for wins even if
it has a tiny disadvantage and objectively should be content
to draw. Because humans are so prone to error, this can
enhance the program's results by avoiding repetitions and
dragging the games out until the human (or inferior program)
makes a mistake or a series of sub-optimal moves.
However, in this match with GM Benjamin, telling the
program to despise his extra material could have had dire
consequences, had he not so often and so willingly agreed
to repeat the position when he stood better. When you look
at what poor positions Rybka got out of the opening phase,
it becomes clear that almost any contempt at all would
have the opposite of the desired effect: it would prevent the
program from escaping with a draw where it is objectively
worse.
After playing over a good number of computer vs. computer
games involving Rybka, my impression was that it had a
better grasp or positional judgment than its opponents, and
this often translated into the win of a pawn, for instance.
But against GM Benjamin, the missing pawn nearly always
gave him a clear advantage in the opening, and the program
was unimpressive. I expect that a human super-GM would
have done much better at steering toward positions where
the missing pawn was not such a clear handicap, toward
gambit play where the material balance is not nearly so
important as time and weak squares, for instance.
In fact, it would be an interesting match to give Rybka the
pawn odds and let a super-GM try and "outsmart" the
program, knowing it has the technique to convert. Would
we get a Marshallesque 0-7 score, or would the super-GM
somehow manage to assert the vastly superior intelligence
of the human species, I wonder?
-- help bot
|
| | |
Date: 17 Aug 2007 10:31:02
From: Fred
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
On Thu, 16 Aug 2007 21:27:01 -0700, help bot <nomorechess@hotmail.com >
wrote:
>On Aug 16, 4:22 pm, Fred <F...@google.com> wrote:
>
>> > Game one: programmers claim problem with contempt factor
>> >caused bad moves; GM Benjamin is awarded a win and the
>>
>> Could you explain the "contempt factor", please?
>>
>> Thanks.
>
> In a normal game with no odds, the contempt factor tells
>the program to "think optimistically", to go for wins even if
>it has a tiny disadvantage and objectively should be content
>to draw. Because humans are so prone to error, this can
>enhance the program's results by avoiding repetitions and
>dragging the games out until the human (or inferior program)
>makes a mistake or a series of sub-optimal moves.
>
> However, in this match with GM Benjamin, telling the
>program to despise his extra material could have had dire
>consequences, had he not so often and so willingly agreed
>to repeat the position when he stood better. When you look
>at what poor positions Rybka got out of the opening phase,
>it becomes clear that almost any contempt at all would
>have the opposite of the desired effect: it would prevent the
>program from escaping with a draw where it is objectively
>worse.
>
> After playing over a good number of computer vs. computer
>games involving Rybka, my impression was that it had a
>better grasp or positional judgment than its opponents, and
>this often translated into the win of a pawn, for instance.
>But against GM Benjamin, the missing pawn nearly always
>gave him a clear advantage in the opening, and the program
>was unimpressive. I expect that a human super-GM would
>have done much better at steering toward positions where
>the missing pawn was not such a clear handicap, toward
>gambit play where the material balance is not nearly so
>important as time and weak squares, for instance.
>
> In fact, it would be an interesting match to give Rybka the
>pawn odds and let a super-GM try and "outsmart" the
>program, knowing it has the technique to convert. Would
>we get a Marshallesque 0-7 score, or would the super-GM
>somehow manage to assert the vastly superior intelligence
>of the human species, I wonder?
>
>
> -- help bot
Thanks.
|
| |
Date: 10 Aug 2007 20:51:22
From: SAT W-7
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
Thank you for the break downs ..
will look there games up in a few days when i have more time...
|
| |
Date: 10 Aug 2007 01:55:26
From: help bot
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
On Aug 9, 9:03 pm, Sin...@webtv.net (SAT W-7) wrote:
> Thank you ..
>
> They played two games , what pawns were missing from Rybka and what
> color did they play.
> Did Joel put up a fight ?
(Rybka wins two, GM Benjamin wins one, the rest are
drawn.)
Now that I have completed playing over all these games,
I can say that generally, GM Benjamin took draws when he
had the advantage, not wanting to take risks. This allowed
Rybka to win the match, since it had no such aversion to
"risky play" and in fact many of the draws were the result
of Rybka and GM Benjamin repeating the position, the
former because it stood worse, and the latter because he
was chicken or had no clue how to make headway.
Perhaps the time controls had something to do with this.
GM Benjamin seemed to handle his pawn advantage in
the openings very well, often steering into lines which
closely resemble "book" openings but where his opponent
was missing a *key* pawn. But in the middle game the
computer reigned supreme, keeping the tactics well under
control while occasionally prying loose a pawn. In both its
wins Rybka showed remarkable "cleverness"; once it used
every piece left on the board to entrap the enemy Queen,
netting an exchange; and in the other win, the program
neatly exploited tactics to win what might well have been
a drawable ending, these tactics aside.
Many years ago, a commercial chess program was
accused of "cheating" because its programmer(s) had
deliberately steered it in the openings to those types of
positions which humans play relatively poorly; this gave
it an edge over other programs in computer vs. human
ratings, but it did so by dragging down the play of the
opponents, not raising the level of the computer's own
play.
Now in several of these odds games between Rybka
and GM Benjamin you will notice that Rybka played
moves like ...Nc6 (in front of the c-pawn, blocking in
its own Queen), ...f5, ...e6, landing it a backward
pawn or a very queer-looking position; it was deja vu
all over again! But the truth is, GM Benjamin is no
stranger to strange-looking chess; he in fact wrote a
book or two on unusual chess openings, analyzing
oddball lines in considerable detail and sorting them
out into different categories, so I think it was just that
the time control was arranged to allow this match to
be completed fairly quickly, despite it requiring a
total of eight games, one for each gambit pawn.
If you look at the ratings given for each of the two
contestants, you may find GM Benjamin listed
somewhere in the 2500s, while I saw the number
"3100" given for Rybka! (I'll buy 3100 for the games
I replayed in which Rybka convincingly defeated
Shredder, but these games were of a far lower
quality, for whatever reason.) In one game, GM
Benjamin had a nice position as the result of the
opening (and Rybka's missing *key* pawn), then
he went and lined up his Queen and Bishop on h7,
as if he were playing a rank beginner!
I think it is not entirely clear whether JB is really
a stronger opponent than his rating would suggest
for the top programs when you consider his lack
of "the will to win", which cost him many draws.
The most difficult opponents for Rybka would be
those who can successfully steer toward closed
positions, and then outmaneuver the program to
go on to (occasionally) win; not someone who can
successfully steer into repetitions of positions, but
who sometimes hangs material and loses. I'm
thinking Boris Spassky in his prime, Or Tigran
Petrosian in his prime. Someone who can either
shut down all counterplay, or else who can make
all the opponent's work count for nothing, because
his King has been mated the moment he was
about to promote a pawn on the other wing.
Perhaps we will learn more about the software
problem which allegedly caused Rybka to muck
up so badly. As far as I could see, the program
was *not* particularly adept at playing openings a
pawn down, and GM Benjamin made it look rather
worse than in any of the computer vs. computer
match ups with no odds; in those games, Rybka
quite often emerged from the openings with some
substantial edge, as though its judgment of
position were somehow superior to the other
programs'.
-- help bot
|
| |
Date: 09 Aug 2007 22:00:42
From: help bot
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
On Aug 9, 9:03 pm, Sin...@webtv.net (SAT W-7) wrote:
> They played two games , what pawns were missing from Rybka and what
> color did they play.
> Did Joel put up a fight ?
Game one: programmers claim problem with contempt factor
caused bad moves; GM Benjamin is awarded a win and the
rest of the game is played out after a "fix", but why, nobody
knows; the grandmaster did not even try to win it a second
time. Rybka was White and gave the d2-pawn as odds.
Game two: game was drawn after some really boring play
in which GM Benjamin had a clear advantage, but was too
chicken to try and make something of it. Rybka was Black
and gave the c7-pawn.
Game three: GM Benjamin hangs back his extra pawn,
but simplifies to hold a draw quite easily. Rybka had White
and gave the b2-pawn as odds. (Opening choice? The super--
ultra--accelerated fianchetto.)
Game four: GM Benjamin is thoroughly outplayed in typical
computer style; Rybka: "my backward pawn? no problem.
Your extra pawns? no problem. Your trapped Queen? Big
problem!" Rybka had Black and gave the a7-pawn as odds.
-- help bot
|
| | |
Date: 16 Aug 2007 17:22:13
From: Fred
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
> Game one: programmers claim problem with contempt factor
>caused bad moves; GM Benjamin is awarded a win and the
Could you explain the "contempt factor", please?
Thanks.
|
| | | |
Date: 18 Aug 2007 20:13:07
From: David Richerby
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
Fred <Fred@google.com > wrote:
> Could you explain the "contempt factor", please?
Suppose a chess engine evaluates drawn positions as zero, which seems
natural enough. This means that, in a position where the evaluation
is, say, -0.02 (against the computer), the computer would accept a
draw by repetition if it saw nothing else better than -0.02. In
practice, this isn't a good idea as it means that the engine will play
for a draw as soon as some random little feature of its evaluation
function causes the evaluation to go ever so slightly negative.
So, what the engines do is score a draw as something a little below
zero; say -0.20. This means that the engine will only play for a draw
if it is at a noticeable disadvantage. Of course, the difference
between -0.22 (play for a draw if there is one) and -0.19 (play for
the win!) is also essentially noise in the evaluation function. But
at least the engine can be sure it has some kind of disadvantage at
both scores and that this disadvantage isn't overwhelming.
Dave.
--
David Richerby Expensive Simple Drink (TM): it's like
www.chiark.greenend.org.uk/~davidr/ a refreshing juice beverage but it
has no moving parts and it'll break
the bank!
|
| |
Date: 09 Aug 2007 20:47:29
From: help bot
Subject: Re: GM Joel Benjamin vs Rybka today.Give us your break down Helpbot
|
I am just now replaying these games, and it looks like there
were some "issues" with the contempt factor or something
else which seriously mucked up Rybka's play.
Witness game one, where Rybka got the patzeriffic idea
of N-g5 and Nxf7, losing of course. And game two, where
as Black with the c7-pawn gambit odds, after 1.e4!, Rybka
responded with ...Q-a5 -- another duffer's move.
Before I finish replaying that game, let me state that I
believe the contempt factor ought to have been set so the
program could try to escape any inferior positions by
seeking draws; this is because, unlike in normal games,
a grandmaster with an extra pawn at move one is a very
dangerous opponent!
Witness, for example, GM Benjamin's clever choice in
the first game, where White is missing the d-pawn, of
1.e4 *Nc6* -- a move normally punishable by shoving the
d-pawn and knocking the Knight around. Without the
d-pawn, White just looks at this Knight and thinks: hey,
you: get outta my face!
I hope the rest of these games are not severely marred
by any tweaking the programmers did to "help" Rybka
in this particular match, because it looks like their kind
of help is no real help at all.
-- help bot
|
|
