| |
Main
Date: 06 Oct 2006 09:24:54
From: Wilma
Subject: Capa cheated with CM9k in 1918!
|
I posted this in the wrong group. Meant to post it in here.
Chessmaster 9000 analyzed the classic game Capablanca v shall, New
York 1918. It was in this game that shall introduced the shall
Gambit.
I used the factory settings for CM, including a hash table of 4MB. The
analysis was done by the Chessmaster 9000 personality. I set the time
alloted to analyze a move at 180 seconds to simulate classical time
control. My machine has 256M memory and runs at 1.66gH.
The first 11 moves for both sides were book moves. The engine, The King
3.23, began analyzing with white's 12th move. The game ended with
shall's resignation on his 36th move, so Capa had 25 moves analyzed,
and shall had 24 moves analyzed.
=======================================================
Using zero tolerance, the results were as follows:
Capablanca: 72% of his moves were CM9k's best move.
shall: 58% of his moves were CM9k's best move.
=======================================================
I got these numbers by actually going through the analysis move by move.
After each players move, CM gives the best move for the other player's
next move. If the player made that move, I counted it as a hit. That's
the only way I saw to get a zero tolerance count. CM provides stats for
"CM agree" which apparently uses something other than zero tolerance; I
couldn't find out exactly how it decides that. It gives Capablance 100%
agreement and shall 83% agreement with CM. Somebody else will have to
tell us what "agreement" means -- JM, You reading?
Here is the game:
[Event "New York"]
[Date "1918"]
[White "Capablanca, J.R."]
[Black "shall, Frank J."]
[Result "1-0"]
1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.O-O Be7 6.Re1 b5 7.Bb3 O-O 8.c3
d5 9.exd5 Nxd5 10.Nxe5 Nxe5 11.Rxe5 Nf6 12.Re1 Bd6 13.h3 Ng4 14.Qf3 Qh4
15.d4 Nxf2 16.Re2 Bg4 17.hxg4 Bh2+ 18.Kf1 Bg3 19.Rxf2 Qh1+ 20.Ke2 Bxf2
21.Bd2 Bh4 22.Qh3 Rae8+ 23.Kd3 Qf1+ 24.Kc2 Bf2 25.Qf3 Qg1 26.Bd5 c5
27.dxc5 Bxc5 28.b4 Bd6 29.a4 a5 30.axb5 axb4 31.Ra6 bxc3 32.Nxc3 Bb4
33.b6 Bxc3 34.Bxc3 h6 35.b7 Re3 36.Bxf7+ 1-0
Wilma
|
|
| |
Date: 06 Oct 2006 18:48:06
From:
Subject: Re: Capa cheated with CM9k in 1918!
|
> > Indeed, "agreement" means that the player either chose the same move
> as
> > CM, or the score of the move chosen was within a small range of that
> > move (I'm not sure of the range -- I can look it up tonight) so that
> it
> > is probably "just as good".
> >
> > jm
> >
> >
Wilma wrote:
> Wow. So CM9k agrees with Capa's moves 100% of the time in this game and
> the exactly same move 72% of the time. Seems pretty obvious that no one
> can reasonably cite such data about Kramnik as evidence that he cheated.
>
> Wilma
I looked it up. If the score of a move is less than 0.5 worse than the
move that CM thinks is best, it will still "agree" with the move. This
threshold is priily because the average CM user tends to perform
short analysis (10 seconds per move, on average) on low-end machines.
Therefore, a perfectly reasonable move in long analysis may not look as
good in short analysis.
So, to avoid confusion/arguments, we stuck with the 0.5 threshold.
jm
|
| | |
Date: 16 Oct 2006 10:27:03
From: David Richerby
Subject: Re: Capa cheated with CM9k in 1918!
|
<[email protected] > wrote:
>Wilma wrote:
>> <[email protected]> wrote:
>>> Indeed, "agreement" means that the player either chose the same
>>> move as CM, or the score of the move chosen was within a small
>>> range of that move
>>
>> Wow. So CM9k agrees with Capa's moves 100% of the time in this game
>> and the exactly same move 72% of the time.
>
> I looked it up. If the score of a move is less than 0.5 worse than
> the move that CM thinks is best, it will still "agree" with the
> move.
Hardly surprising that Capablanca should always be within half a pawn
of any decent chess engine!
> This threshold is priily because the average CM user tends to
> perform short analysis (10 seconds per move, on average) on low-end
> machines. Therefore, a perfectly reasonable move in long analysis
> may not look as good in short analysis.
Wouldn't it have been better to allow that parameter to be tuned? As
it is, Chessmaster will `agree' with almost any move that doesn't drop
material: that's useful for finding blunders but not much good for
serious analysis.
Dave.
--
David Richerby Miniature Unholy Car (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ high-performance luxury car but it's
also a crime against nature and you
can hold in it your hand!
|
| | | |
Date: 16 Oct 2006 18:08:15
From: Wilma
Subject: Re: Capa cheated with CM9k in 1918!
|
I didn't use the CM agreement percentage CM calculates. I went through
the analysis move by move myself and counted as a hit only the moves CM
said were the BEST move. That is, I used a zero tolerance instead of
CM's .5 tolerance, which was too liberal for my purposes.
BTW, the .5 parameter can indeed be tuned, Merlino said. I've tried a
few different settings in the method he gave me. This setting,
[annotator]
slightly=5
better=25
much=50,
gives me the same agreement percentage I get by going through the
analusis move by move, but it's too strict for other features of the
analysis. Nearly every move that isn't the best is called the worst.
Setting the parameters at 25, 50, 100 is better overall, but I have to
go through move by move to get a zero tolerance measure.
So, I'm leaving the parameter at the default of .5 and going through the
analysis move by move to get the percentage of best move hits. It only
takes a minute to do that, so it not worth exausting the possibilities
looking for the best parameter settings.
Wilma
David Richerby <[email protected] > wrote in
news:3Ml*[email protected]:
> <[email protected]> wrote:
>>Wilma wrote:
>>> <[email protected]> wrote:
>>>> Indeed, "agreement" means that the player either chose the same
>>>> move as CM, or the score of the move chosen was within a small
>>>> range of that move
>>>
>>> Wow. So CM9k agrees with Capa's moves 100% of the time in this game
>>> and the exactly same move 72% of the time.
>>
>> I looked it up. If the score of a move is less than 0.5 worse than
>> the move that CM thinks is best, it will still "agree" with the
>> move.
>
> Hardly surprising that Capablanca should always be within half a pawn
> of any decent chess engine!
>
>
>> This threshold is priily because the average CM user tends to
>> perform short analysis (10 seconds per move, on average) on low-end
>> machines. Therefore, a perfectly reasonable move in long analysis
>> may not look as good in short analysis.
>
> Wouldn't it have been better to allow that parameter to be tuned? As
> it is, Chessmaster will `agree' with almost any move that doesn't drop
> material: that's useful for finding blunders but not much good for
> serious analysis.
>
>
> Dave.
>
|
| | |
Date: 07 Oct 2006 04:49:38
From: Wilma
Subject: Re: Capa cheated with CM9k in 1918!
|
Thanks for the information, jm.
Wilma
[email protected] wrote in news:1160185686.768391.147410
@c28g2000cwb.googlegroups.com:
>> > Indeed, "agreement" means that the player either chose the same
move
>> as
>> > CM, or the score of the move chosen was within a small range of
that
>> > move (I'm not sure of the range -- I can look it up tonight) so
that
>> it
>> > is probably "just as good".
>> >
>> > jm
>> >
>> >
> Wilma wrote:
>> Wow. So CM9k agrees with Capa's moves 100% of the time in this game
and
>> the exactly same move 72% of the time. Seems pretty obvious that no
one
>> can reasonably cite such data about Kramnik as evidence that he
cheated.
>>
>> Wilma
>
> I looked it up. If the score of a move is less than 0.5 worse than the
> move that CM thinks is best, it will still "agree" with the move. This
> threshold is priily because the average CM user tends to perform
> short analysis (10 seconds per move, on average) on low-end machines.
> Therefore, a perfectly reasonable move in long analysis may not look
as
> good in short analysis.
>
> So, to avoid confusion/arguments, we stuck with the 0.5 threshold.
>
> jm
>
>
|
| |
Date: 06 Oct 2006 16:02:45
From:
Subject: Re: Capa cheated with CM9k in 1918!
|
Wilma wrote:
> I posted this in the wrong group. Meant to post it in here.
>
> Chessmaster 9000 analyzed the classic game Capablanca v shall, New
> York 1918. It was in this game that shall introduced the shall
> Gambit.
>
> I used the factory settings for CM, including a hash table of 4MB. The
> analysis was done by the Chessmaster 9000 personality. I set the time
> alloted to analyze a move at 180 seconds to simulate classical time
> control. My machine has 256M memory and runs at 1.66gH.
>
> The first 11 moves for both sides were book moves. The engine, The King
> 3.23, began analyzing with white's 12th move. The game ended with
> shall's resignation on his 36th move, so Capa had 25 moves analyzed,
> and shall had 24 moves analyzed.
>
> =======================================================
> Using zero tolerance, the results were as follows:
>
> Capablanca: 72% of his moves were CM9k's best move.
>
> shall: 58% of his moves were CM9k's best move.
> =======================================================
>
> I got these numbers by actually going through the analysis move by move.
> After each players move, CM gives the best move for the other player's
> next move. If the player made that move, I counted it as a hit. That's
> the only way I saw to get a zero tolerance count. CM provides stats for
> "CM agree" which apparently uses something other than zero tolerance; I
> couldn't find out exactly how it decides that. It gives Capablance 100%
> agreement and shall 83% agreement with CM. Somebody else will have to
> tell us what "agreement" means -- JM, You reading?
Indeed, "agreement" means that the player either chose the same move as
CM, or the score of the move chosen was within a small range of that
move (I'm not sure of the range -- I can look it up tonight) so that it
is probably "just as good".
jm
|
| | |
Date: 06 Oct 2006 23:13:24
From: Wilma
Subject: Re: Capa cheated with CM9k in 1918!
|
Wow. So CM9k agrees with Capa's moves 100% of the time in this game and
the exactly same move 72% of the time. Seems pretty obvious that no one
can reasonably cite such data about Kramnik as evidence that he cheated.
Wilma
[email protected] wrote in news:1160175765.888439.119480
@k70g2000cwa.googlegroups.com:
>
> Wilma wrote:
{snip}
>
>> =======================================================
>> Using zero tolerance, the results were as follows:
>>
>> Capablanca: 72% of his moves were CM9k's best move.
>>
>> shall: 58% of his moves were CM9k's best move.
>> =======================================================
>>
>> I got these numbers by actually going through the analysis move by
move.
>> After each players move, CM gives the best move for the other
player's
>> next move. If the player made that move, I counted it as a hit.
That's
>> the only way I saw to get a zero tolerance count. CM provides stats
for
>> "CM agree" which apparently uses something other than zero tolerance;
I
>> couldn't find out exactly how it decides that. It gives Capablance
100%
>> agreement and shall 83% agreement with CM. Somebody else will have
to
>> tell us what "agreement" means -- JM, You reading?
>
> Indeed, "agreement" means that the player either chose the same move
as
> CM, or the score of the move chosen was within a small range of that
> move (I'm not sure of the range -- I can look it up tonight) so that
it
> is probably "just as good".
>
> jm
>
>
|
| |
Date: 06 Oct 2006 11:24:14
From: John A Swartz
Subject: Re: Capa cheated with CM9k in 1918!
|
The didn't call Capa "The Chess Machine" for nothing... ;-)
Wilma wrote:
> I posted this in the wrong group. Meant to post it in here.
>
> Chessmaster 9000 analyzed the classic game Capablanca v shall, New
> York 1918. It was in this game that shall introduced the shall
> Gambit.
>
> I used the factory settings for CM, including a hash table of 4MB. The
> analysis was done by the Chessmaster 9000 personality. I set the time
> alloted to analyze a move at 180 seconds to simulate classical time
> control. My machine has 256M memory and runs at 1.66gH.
>
> The first 11 moves for both sides were book moves. The engine, The King
> 3.23, began analyzing with white's 12th move. The game ended with
> shall's resignation on his 36th move, so Capa had 25 moves analyzed,
> and shall had 24 moves analyzed.
>
> =======================================================
> Using zero tolerance, the results were as follows:
>
> Capablanca: 72% of his moves were CM9k's best move.
>
> shall: 58% of his moves were CM9k's best move.
> =======================================================
>
> I got these numbers by actually going through the analysis move by move.
> After each players move, CM gives the best move for the other player's
> next move. If the player made that move, I counted it as a hit. That's
> the only way I saw to get a zero tolerance count. CM provides stats for
> "CM agree" which apparently uses something other than zero tolerance; I
> couldn't find out exactly how it decides that. It gives Capablance 100%
> agreement and shall 83% agreement with CM. Somebody else will have to
> tell us what "agreement" means -- JM, You reading?
>
> Here is the game:
>
> [Event "New York"]
> [Date "1918"]
> [White "Capablanca, J.R."]
> [Black "shall, Frank J."]
> [Result "1-0"]
>
> 1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.O-O Be7 6.Re1 b5 7.Bb3 O-O 8.c3
> d5 9.exd5 Nxd5 10.Nxe5 Nxe5 11.Rxe5 Nf6 12.Re1 Bd6 13.h3 Ng4 14.Qf3 Qh4
> 15.d4 Nxf2 16.Re2 Bg4 17.hxg4 Bh2+ 18.Kf1 Bg3 19.Rxf2 Qh1+ 20.Ke2 Bxf2
> 21.Bd2 Bh4 22.Qh3 Rae8+ 23.Kd3 Qf1+ 24.Kc2 Bf2 25.Qf3 Qg1 26.Bd5 c5
> 27.dxc5 Bxc5 28.b4 Bd6 29.a4 a5 30.axb5 axb4 31.Ra6 bxc3 32.Nxc3 Bb4
> 33.b6 Bxc3 34.Bxc3 h6 35.b7 Re3 36.Bxf7+ 1-0
>
> Wilma
>
|
| |
Date: 06 Oct 2006 14:47:38
From: Simon Krahnke
Subject: Re: Capa cheated with CM9k in 1918!
|
* Wilma <[email protected] > (11:24) schrieb:
> It gives Capablance 100% agreement and shall 83% agreement with CM.
> Somebody else will have to tell us what "agreement" means -- JM, You
> reading?
I think this means that the move made isn't much worse than the best
move, that its value is inside a fixed gin of the best moves value.
The values of two moves may actually be identical, the engine will pick
one move as move, but the other really isn't worse.
mfg, simon .... l
|
| | |
Date: 06 Oct 2006 23:07:35
From: Wilma
Subject: Re: Capa cheated with CM9k in 1918!
|
That sounds right.
Wilma
Simon Krahnke <[email protected] > wrote in news:[email protected]:
> * Wilma <[email protected]> (11:24) schrieb:
>
>> It gives Capablance 100% agreement and shall 83% agreement with CM.
>> Somebody else will have to tell us what "agreement" means -- JM, You
>> reading?
>
> I think this means that the move made isn't much worse than the best
> move, that its value is inside a fixed gin of the best moves value.
>
> The values of two moves may actually be identical, the engine will pick
> one move as move, but the other really isn't worse.
>
> mfg, simon .... l
>
|
| |
Date: 06 Oct 2006 11:44:56
From: David Richerby
Subject: Re: Capa cheated with CM9k in 1918!
|
In article <[email protected] >,
> I used the factory settings for CM, including a hash table of 4MB.
Why such a small hash table with 256M of RAM? You'll get the
strongest analysis by making the hash table as big as you can, so long
as everything fits in RAM at the same time (a hash table in your swap
file is worth than no hash table at all).
> =======================================================
> Using zero tolerance, the results were as follows:
>
> Capablanca: 72% of his moves were CM9k's best move.
>
> shall: 58% of his moves were CM9k's best move.
> =======================================================
LOL. Obviously, Capa was able to refute the gambit over the board
because he had hidden a time machine in the toilets. :-)
Dave.
--
David Richerby Hilarious Pickled Pants (TM): it's
www.chiark.greenend.org.uk/~davidr/ like a well-tailored pair of trousers
but it's preserved in vinegar and a
bundle of laughs!
|
| | |
Date: 06 Oct 2006 11:30:39
From: Wilma
Subject: Re: Capa cheated with CM9k in 1918!
|
I wanted a conservative test of the proposition.
Wilma
David Richerby <[email protected] > wrote in
news:yqp*[email protected]:
> In article <[email protected]>,
>> I used the factory settings for CM, including a hash table of 4MB.
>
> Why such a small hash table with 256M of RAM? You'll get the
> strongest analysis by making the hash table as big as you can, so long
> as everything fits in RAM at the same time (a hash table in your swap
> file is worth than no hash table at all).
>
>
>> =======================================================
>> Using zero tolerance, the results were as follows:
>>
>> Capablanca: 72% of his moves were CM9k's best move.
>>
>> shall: 58% of his moves were CM9k's best move.
>> =======================================================
>
> LOL. Obviously, Capa was able to refute the gambit over the board
> because he had hidden a time machine in the toilets. :-)
>
>
> Dave.
>
|
|
