Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old February 12th 04, 02:20 AM
Mike N.
 
Posts: n/a
Default CM9000 Analysis

If you print out a movelist with auto-annotations after performing an
auto-analysis in the gameroom, the top of the printout shows some statistics
including "Total Error" and "Relevant Error". I can't find descriptions of
either of these in the documentation. Does anyone know what they mean?

The "Total Error" and "Relevant Error" can be 0.00, even when CM9000 does
not agree with 100% of your moves.

Thanks.


Mike N.


  #2   Report Post  
Old February 12th 04, 02:56 PM
Gregory Topov
 
Posts: n/a
Default CM9000 Analysis

"Mike N." wrote in message
news:[email protected]_s04...
If you print out a movelist with auto-annotations after performing an
auto-analysis in the gameroom, the top of the printout shows some

statistics
including "Total Error" and "Relevant Error". I can't find descriptions

of
either of these in the documentation. Does anyone know what they mean?

The "Total Error" and "Relevant Error" can be 0.00, even when CM9000 does
not agree with 100% of your moves.


I couldn't find it in the documentation anywhere either, nor in the
readme.txt files that came with CM9K and with the patches. I'm also curious
to know what exactly it means. For instance, in a recent "blunderful" game,
CM9K gave me the following: (results listed in order of white/black):
CM9000 Agrees - W:42 B:43
CM9000 Disagrees - W:8 B:7
Agreement Pct - W:84% B:86%
Total Error - W:15.29 B:13.34
Relevant Error - W:6.48 B:6.43

The agreement percentage is clearly based on the amount CM9000 agrees with
42/50=84% and 43/50=86%.

But I also don't know what "Total Error" and "Relevant Error" refer to. I
presume that somehow they are related to evaluation results, and loss of
material? e.g. if I blunder away a piece by hanging a knight, that might
contribute 3 points towards "total error". But if my opponent misses the
opportunity to get my hanging, that adds 3 points to his "Total Error". But
my "Relevant Error" stays on 0, because my hanging knight didn't lose any
material in the game due to him missing it. Just guessing here, but could
that be how it works?

Like Mike, I'd also welcome any explanation here.

--
Gregory Topov
---------------------------------------------------------------------
"I don't necessarily agree with everything I say." - Marshall McLuhan


  #3   Report Post  
Old February 12th 04, 10:28 PM
John Merlino
 
Posts: n/a
Default CM9000 Analysis

"Gregory Topov" wrote in message ...
"Mike N." wrote in message
news:[email protected]_s04...
If you print out a movelist with auto-annotations after performing an
auto-analysis in the gameroom, the top of the printout shows some

statistics
including "Total Error" and "Relevant Error". I can't find descriptions

of
either of these in the documentation. Does anyone know what they mean?

The "Total Error" and "Relevant Error" can be 0.00, even when CM9000 does
not agree with 100% of your moves.


I couldn't find it in the documentation anywhere either, nor in the
readme.txt files that came with CM9K and with the patches. I'm also curious
to know what exactly it means. For instance, in a recent "blunderful" game,
CM9K gave me the following: (results listed in order of white/black):
CM9000 Agrees - W:42 B:43
CM9000 Disagrees - W:8 B:7
Agreement Pct - W:84% B:86%
Total Error - W:15.29 B:13.34
Relevant Error - W:6.48 B:6.43

The agreement percentage is clearly based on the amount CM9000 agrees with
42/50=84% and 43/50=86%.

But I also don't know what "Total Error" and "Relevant Error" refer to. I
presume that somehow they are related to evaluation results, and loss of
material? e.g. if I blunder away a piece by hanging a knight, that might
contribute 3 points towards "total error". But if my opponent misses the
opportunity to get my hanging, that adds 3 points to his "Total Error". But
my "Relevant Error" stays on 0, because my hanging knight didn't lose any
material in the game due to him missing it. Just guessing here, but could
that be how it works?

Like Mike, I'd also welcome any explanation here.


Here is what the person who wrote that function says:
---------------
The total error is the sum of the deltas (1.0 = pawn) between the
player's move and Chessmaster's recommended move. However, there's
also a small threshold (can't remember how big) below which the delta
is not considered interesting enough. This is how the Chessmaster can
disagree with a move, but the difference in score is not counted
toward the total error.

Relevant error, which is always less than or equal to total error,
doesn't count errors made by either side once the game is considered
out of reach. The game is considered out of reach when one player gets
ahead by a certain amount of material (5.0?) and never gives up the
big lead.
---------------

There you go,

jm
  #4   Report Post  
Old February 13th 04, 03:06 PM
Gregory Topov
 
Posts: n/a
Default CM9000 Analysis

"John Merlino" wrote:
The total error is the sum of the deltas (1.0 = pawn) between the
player's move and Chessmaster's recommended move. However, there's
also a small threshold (can't remember how big) below which the delta
is not considered interesting enough. This is how the Chessmaster can
disagree with a move, but the difference in score is not counted
toward the total error.

Relevant error, which is always less than or equal to total error,
doesn't count errors made by either side once the game is considered
out of reach. The game is considered out of reach when one player gets
ahead by a certain amount of material (5.0?) and never gives up the
big lead.


Thanks for the explanation John, although I'm not totally sure I quite
understand it. For instance, take the game I used as an example. The
results we
Total Error - White:15.29 Black:13.34
Relevant Error - White:6.48 Black:6.43

Your explanation suggests that errors made by either side once the game is
considered out of reach aren't part of the "relevant error". However, in
this game, the game was never quite out of reach - it was a draw! First
white got a huge lead of around +12. Then white blundered badly, and black
was able to equalize, and force a draw with even material. So would it work
like this:
1. As black blunders, he contributes 6 points to total error and relevant
error. At this point the computer considers the game out of reach, and so
as black keeps blundering, the next 7 points of blunders aren't included
under relevant errors. Hence total errors for black: 13, relevant error for
black: 6.
2. But now in the middle game, white begins blundering and losing the
advantage. The first 7 points of blunder don't count as relevant error,
because the game is still outside black's reach. But once white blunders
even more and allows the game to get within black's reach, all blunders from
now on count as relevant error. So after blundering away 7 points, all
blunders from now on do count as relevant error, and the next 6 points of
blunders count as relevant error. Hence the final result.
One thing interesting about the scores: it shows that both players made
significant errors, and so a draw was a fair result!

Another thing I don't quite understand: Chessmaster can "disagree" with a
move and yet not have that move count to errors? I thought Chessmaster only
"disagreed" with moves if they errors? In this particular game, Chessmaster
would have played differently than most moves for both sides. Yet of the
50+ moves in the game, the final auto-annotation said it only "disagreed"
with 7 or 8 from each player, and agreed with about 85% of the moves in the
game.

Here's the game, so you can auto-analyze it in Chessmaster yourself if
desired. Be warned, there's some horrible chess here - it's a game between
beginners! (but some interesting tactical positions and possibilities!)

[Site "www.chess21.com"]
[White "NN"]
[Black "me"]
[Result "1/2-1/2"]
[WhiteElo "1350"]
[BlackElo "1389"]
[TimeControl "15/0"]

1.f4 d5 2.Nf3 Nc6 3.d4 g6 4.c3 Bg7 5.Be3 Nf6 6.Nbd2 O-O 7.h3 h6 8.g4 Ne4
9.Nxe4 dxe4 10.Nd2 Qd6 11.Nxe4 Qe6 12.Bg2 b6 13.d5 Bxc3+ 14.bxc3 Rd8 15.dxe6
Rxd1+ 16.Rxd1 Bxe6 17.a3 f5 18.gxf5 Bxf5 19.Ng3 Bc2 20.Bxc6 Bxd1 21.Bxa8 Ba4
22.Ne4 c6 23. Bxc6 Bxc6 24.Rg1 Bxe4 25.Rxg6+ Bxg6 26.Kd2 Kf7 27.c4 Bf5 28.h4
Kf6 29.c5 bxc5 30.Bxc5 a6 31.Kc3 Ke6 32.Kd4 Kd7 33.e4 Bg4 34.f5 Be2 35.Kd5
h5 36.e5 Bg4 37.f6 exf6 38.exf6 Ke8 39.Ke5 Kf7 40.Bd4 Kg6 41.Kd6 Kf7 42.Kc6
Be2 43. Kb6 Ke6 44.Kc5 Kf7 45.Kd6 Bg4 46.Be5 Ke8 47.Kc7 Kf7 48.Kb6 Be2
49.Kc6 Bg4 50.Kd6 Ke8 51.Kd5 Kf7 52.Kd6 {draw by repetition}
1/2-1/2

Analysis:
10.Qd6? was a genuine mouseslip (intended was 10.Qd5), resulting in the loss
of a pawn.
12.b6? Allows white to fork the queen and the knight with a pawn after
13.d5!
13.Bxc3? A foolish sacrifice that doesn't resolve the fork problem. Better
was ...Qd7, giving up the knight.
14.Rd8? Leads to a queen and rook exchange, but black is still worse off
than after ...Qd7.
16.Bxe6? Black needed to ignore the threatening pawn, and instead protect
the knight on c6 which will be pinned on the g2-a8 diagonal after white
moves the knight off e4.
17.a3?! Missing 17.Ng3! to eventually win the bishop on c6.
17.f5?! Missing 17.Rd8, preventing further material losses of either the
knight or the rook.
19.Ng3! At last. Computer evaluation puts white ahead by a score of more
than 12!
23.Bxc6? Fearing a trapped bishop and wanting to exchange it for a pawn, but
the loss is more costly than white thinks, because the knight on e4 will be
helplessly pinned.
24.Rg1? Now black wins the knight for free. Less costly for white was
24.Nf2, giving up the rook and winning the bishop.
25.Rxg6? Obviously not seeing that the rook is hanging. After 25.Bxg6, white
has an extra pawn, but by playing carefully black is able to secure a draw.

Summary:
The blunders I made in the beginning of the game led to great losses, but
weren't quite as foolish as merely hanging my pieces, and were the result of
less immediately obvious tactical shots. Although I was down by an
evaluation of more than +12, my opponent made three consecutive blunders to
throw away the game and allow a draw.
--
Gregory Topov
---------------------------------------------------------------------
"I don't necessarily agree with everything I say." - Marshall McLuhan


  #5   Report Post  
Old February 13th 04, 08:54 PM
John Merlino
 
Posts: n/a
Default CM9000 Analysis

"Gregory Topov" wrote in message ...
"John Merlino" wrote:
The total error is the sum of the deltas (1.0 = pawn) between the
player's move and Chessmaster's recommended move. However, there's
also a small threshold (can't remember how big) below which the delta
is not considered interesting enough. This is how the Chessmaster can
disagree with a move, but the difference in score is not counted
toward the total error.

Relevant error, which is always less than or equal to total error,
doesn't count errors made by either side once the game is considered
out of reach. The game is considered out of reach when one player gets
ahead by a certain amount of material (5.0?) and never gives up the
big lead.


Thanks for the explanation John, although I'm not totally sure I quite
understand it. For instance, take the game I used as an example. The
results we
Total Error - White:15.29 Black:13.34
Relevant Error - White:6.48 Black:6.43

Your explanation suggests that errors made by either side once the game is
considered out of reach aren't part of the "relevant error". However, in
this game, the game was never quite out of reach - it was a draw! First
white got a huge lead of around +12. Then white blundered badly, and black
was able to equalize, and force a draw with even material. So would it work
like this:
1. As black blunders, he contributes 6 points to total error and relevant
error. At this point the computer considers the game out of reach, and so
as black keeps blundering, the next 7 points of blunders aren't included
under relevant errors. Hence total errors for black: 13, relevant error for
black: 6.
2. But now in the middle game, white begins blundering and losing the
advantage. The first 7 points of blunder don't count as relevant error,
because the game is still outside black's reach. But once white blunders
even more and allows the game to get within black's reach, all blunders from
now on count as relevant error. So after blundering away 7 points, all
blunders from now on do count as relevant error, and the next 6 points of
blunders count as relevant error. Hence the final result.
One thing interesting about the scores: it shows that both players made
significant errors, and so a draw was a fair result!


Your assessment of how the process works is correct. The analysis
engine only looks at the current evaluation of the position to
determine if the game is "out of reach", because it always assumes
"best play". Even though the game was a draw, analysis of a position
must always make the assumption that both sides are going to make the
best moves.

Another thing I don't quite understand: Chessmaster can "disagree" with a
move and yet not have that move count to errors? I thought Chessmaster only
"disagreed" with moves if they errors? In this particular game, Chessmaster
would have played differently than most moves for both sides. Yet of the
50+ moves in the game, the final auto-annotation said it only "disagreed"
with 7 or 8 from each player, and agreed with about 85% of the moves in the
game.


Even though Chessmaster may "disagree" with a move, all that TRULY
means is that the engine would have made a different move. However,
sometimes the difference between two (or more) moves is just a few
centipawns of evaluation. For example, in your game, on Black's 21st
move, the difference between Ba4, Bb3 and Bb2 is less than 0.10 of
score. The analysis engine assumes that any move that is within a
small evaluation of what is considered the "best" move is probably
"just as good". Another example of this is 30...a6, which is literally
0.01 worse than 30...a5, according to the engine. This is simply not
significant enough to report, because it might be "just as good" in
the long run.

This results in a big difference (and some confusion) in the way the
analysis engine works than just agreeing or disagreeing with moves.
There are many instances throughout the analysis of this game where CM
would have played a different move, and the above examples are just
two of them from this one game.

But I think I falsely created your misunderstanding above by phrasing
something poorly in the explanation of how the scoring works. So, let
me rewrite the first paragraph as follows:

--------------
The total error is the sum of the deltas (1.0 = pawn) between the
player's move and Chessmaster's recommended move. However, there's
also a small threshold (can't remember how big) below which the delta
is not considered interesting enough. This is how the Chessmaster can
POSSIBLY PLAY A DIFFERENT move (this used to read "disagree with a
move"), but the difference in score is not counted toward the total
error.
--------------

The problem is terminology. "Disagreeing" with a move can mean two
things:

1) Chessmaster would have played a different move, and
2) The analysis engine tells you that Chessmaster disagreed with the
move.

It is possible for 1) to happen without 2) occurring, because of that
small threshold. Does this make sense?

Here's the game, so you can auto-analyze it in Chessmaster yourself if
desired. Be warned, there's some horrible chess here - it's a game between
beginners! (but some interesting tactical positions and possibilities!)

[Site "www.chess21.com"]
[White "NN"]
[Black "me"]
[Result "1/2-1/2"]
[WhiteElo "1350"]
[BlackElo "1389"]
[TimeControl "15/0"]

1.f4 d5 2.Nf3 Nc6 3.d4 g6 4.c3 Bg7 5.Be3 Nf6 6.Nbd2 O-O 7.h3 h6 8.g4 Ne4
9.Nxe4 dxe4 10.Nd2 Qd6 11.Nxe4 Qe6 12.Bg2 b6 13.d5 Bxc3+ 14.bxc3 Rd8 15.dxe6
Rxd1+ 16.Rxd1 Bxe6 17.a3 f5 18.gxf5 Bxf5 19.Ng3 Bc2 20.Bxc6 Bxd1 21.Bxa8 Ba4
22.Ne4 c6 23. Bxc6 Bxc6 24.Rg1 Bxe4 25.Rxg6+ Bxg6 26.Kd2 Kf7 27.c4 Bf5 28.h4
Kf6 29.c5 bxc5 30.Bxc5 a6 31.Kc3 Ke6 32.Kd4 Kd7 33.e4 Bg4 34.f5 Be2 35.Kd5
h5 36.e5 Bg4 37.f6 exf6 38.exf6 Ke8 39.Ke5 Kf7 40.Bd4 Kg6 41.Kd6 Kf7 42.Kc6
Be2 43. Kb6 Ke6 44.Kc5 Kf7 45.Kd6 Bg4 46.Be5 Ke8 47.Kc7 Kf7 48.Kb6 Be2
49.Kc6 Bg4 50.Kd6 Ke8 51.Kd5 Kf7 52.Kd6 {draw by repetition}
1/2-1/2


Amusing game, as it definitely brings back memories of how well (or
badly?) I play. :-)

I analyzed the game with a slightly different personality, and at 60
seconds per move on a P4-2.4. Obviously, the more time you give the
engine to think about each move (and the faster the processor), the
better the analysis will be.

My summary came up as follows:

White Black
Book Move 2 1
Leave Book 0 1
CM9000 Agrees 44 42
CM9000 Disagrees 6 8
Agreement Pct. 88% 84%
Total Error 16.64 14.39
Relevant Error 6.69 7.12
Missed Mate 0 0
Moved Into Mate 0 0

This is similar to your results.

jm


  #6   Report Post  
Old February 14th 04, 02:24 AM
Gregory Topov
 
Posts: n/a
Default CM9000 Analysis

"John Merlino" wrote in message

But I think I falsely created your misunderstanding above by phrasing
something poorly in the explanation of how the scoring works. So, let
me rewrite the first paragraph as follows:
--------------
The total error is the sum of the deltas (1.0 = pawn) between the
player's move and Chessmaster's recommended move. However, there's
also a small threshold (can't remember how big) below which the delta
is not considered interesting enough. This is how the Chessmaster can
POSSIBLY PLAY A DIFFERENT move (this used to read "disagree with a
move"), but the difference in score is not counted toward the total
error.


That explains it perfectly. I was indeed confusing the "CM9000 Disagrees"
figure that the auto-annotation produces with the "possibly plays a
different move." Now all is clear. Your explanation is greatly
appreciated!

Amusing game, as it definitely brings back memories of how well (or
badly?) I play. :-)


And hey, that's after Chessmaster has improved my game...you should have
seen what my playing was like before Chessmaster came into my life! grin

--
Gregory Topov
---------------------------------------------------------------------
"I don't necessarily agree with everything I say." - Marshall McLuhan


  #7   Report Post  
Old February 14th 04, 02:31 AM
Gregory Topov
 
Posts: n/a
Default CM9000 Analysis

"John Merlino" wrote in message:

CM9000 Agrees 44 42
CM9000 Disagrees 6 8
Agreement Pct. 88% 84%
Total Error 16.64 14.39
Relevant Error 6.69 7.12


Just to confirm that I understand you correctly, using your auto-analysis
results:
1. So these results mean that the 6 white moves that CM9000 "Disagreed"
with, happened to have a total error value of 16.64. CM9000 may have
selected a different move for the other 44 moves that white played , but the
difference in evaluation was so small it wasn't counted as a move to
"Disagree" with, nor towards Total Error.
2. Similarly, the 14.39 Total Error for black, was produced by the
evaluation difference for the 8 moves that CM9000 Disagreed with.
3. Of this 14.39, 14.39-7.12=7.28 were errors made when CM9000 regarded the
game as out of reach for black, and so they didn't count towards Relevant
Error.

Do I have it right?

--
Gregory Topov
---------------------------------------------------------------------
"I don't necessarily agree with everything I say." - Marshall McLuhan


  #8   Report Post  
Old February 14th 04, 03:49 AM
Mike N.
 
Posts: n/a
Default CM9000 Analysis

Thanks, John & Gregory.

Mike



"John Merlino" wrote in message
om...
"Gregory Topov" wrote in message

...
"John Merlino" wrote:
The total error is the sum of the deltas (1.0 = pawn) between the
player's move and Chessmaster's recommended move. However, there's
also a small threshold (can't remember how big) below which the delta
is not considered interesting enough. This is how the Chessmaster can
disagree with a move, but the difference in score is not counted
toward the total error.

Relevant error, which is always less than or equal to total error,
doesn't count errors made by either side once the game is considered
out of reach. The game is considered out of reach when one player gets
ahead by a certain amount of material (5.0?) and never gives up the
big lead.


Thanks for the explanation John, although I'm not totally sure I quite
understand it. For instance, take the game I used as an example. The
results we
Total Error - White:15.29 Black:13.34
Relevant Error - White:6.48 Black:6.43

Your explanation suggests that errors made by either side once the game

is
considered out of reach aren't part of the "relevant error". However,

in
this game, the game was never quite out of reach - it was a draw! First
white got a huge lead of around +12. Then white blundered badly, and

black
was able to equalize, and force a draw with even material. So would it

work
like this:
1. As black blunders, he contributes 6 points to total error and

relevant
error. At this point the computer considers the game out of reach, and

so
as black keeps blundering, the next 7 points of blunders aren't included
under relevant errors. Hence total errors for black: 13, relevant error

for
black: 6.
2. But now in the middle game, white begins blundering and losing the
advantage. The first 7 points of blunder don't count as relevant error,
because the game is still outside black's reach. But once white

blunders
even more and allows the game to get within black's reach, all blunders

from
now on count as relevant error. So after blundering away 7 points, all
blunders from now on do count as relevant error, and the next 6 points

of
blunders count as relevant error. Hence the final result.
One thing interesting about the scores: it shows that both players made
significant errors, and so a draw was a fair result!


Your assessment of how the process works is correct. The analysis
engine only looks at the current evaluation of the position to
determine if the game is "out of reach", because it always assumes
"best play". Even though the game was a draw, analysis of a position
must always make the assumption that both sides are going to make the
best moves.

Another thing I don't quite understand: Chessmaster can "disagree" with

a
move and yet not have that move count to errors? I thought Chessmaster

only
"disagreed" with moves if they errors? In this particular game,

Chessmaster
would have played differently than most moves for both sides. Yet of

the
50+ moves in the game, the final auto-annotation said it only

"disagreed"
with 7 or 8 from each player, and agreed with about 85% of the moves in

the
game.


Even though Chessmaster may "disagree" with a move, all that TRULY
means is that the engine would have made a different move. However,
sometimes the difference between two (or more) moves is just a few
centipawns of evaluation. For example, in your game, on Black's 21st
move, the difference between Ba4, Bb3 and Bb2 is less than 0.10 of
score. The analysis engine assumes that any move that is within a
small evaluation of what is considered the "best" move is probably
"just as good". Another example of this is 30...a6, which is literally
0.01 worse than 30...a5, according to the engine. This is simply not
significant enough to report, because it might be "just as good" in
the long run.

This results in a big difference (and some confusion) in the way the
analysis engine works than just agreeing or disagreeing with moves.
There are many instances throughout the analysis of this game where CM
would have played a different move, and the above examples are just
two of them from this one game.

But I think I falsely created your misunderstanding above by phrasing
something poorly in the explanation of how the scoring works. So, let
me rewrite the first paragraph as follows:

--------------
The total error is the sum of the deltas (1.0 = pawn) between the
player's move and Chessmaster's recommended move. However, there's
also a small threshold (can't remember how big) below which the delta
is not considered interesting enough. This is how the Chessmaster can
POSSIBLY PLAY A DIFFERENT move (this used to read "disagree with a
move"), but the difference in score is not counted toward the total
error.
--------------

The problem is terminology. "Disagreeing" with a move can mean two
things:

1) Chessmaster would have played a different move, and
2) The analysis engine tells you that Chessmaster disagreed with the
move.

It is possible for 1) to happen without 2) occurring, because of that
small threshold. Does this make sense?

Here's the game, so you can auto-analyze it in Chessmaster yourself if
desired. Be warned, there's some horrible chess here - it's a game

between
beginners! (but some interesting tactical positions and possibilities!)

[Site "www.chess21.com"]
[White "NN"]
[Black "me"]
[Result "1/2-1/2"]
[WhiteElo "1350"]
[BlackElo "1389"]
[TimeControl "15/0"]

1.f4 d5 2.Nf3 Nc6 3.d4 g6 4.c3 Bg7 5.Be3 Nf6 6.Nbd2 O-O 7.h3 h6 8.g4 Ne4
9.Nxe4 dxe4 10.Nd2 Qd6 11.Nxe4 Qe6 12.Bg2 b6 13.d5 Bxc3+ 14.bxc3 Rd8

15.dxe6
Rxd1+ 16.Rxd1 Bxe6 17.a3 f5 18.gxf5 Bxf5 19.Ng3 Bc2 20.Bxc6 Bxd1 21.Bxa8

Ba4
22.Ne4 c6 23. Bxc6 Bxc6 24.Rg1 Bxe4 25.Rxg6+ Bxg6 26.Kd2 Kf7 27.c4 Bf5

28.h4
Kf6 29.c5 bxc5 30.Bxc5 a6 31.Kc3 Ke6 32.Kd4 Kd7 33.e4 Bg4 34.f5 Be2

35.Kd5
h5 36.e5 Bg4 37.f6 exf6 38.exf6 Ke8 39.Ke5 Kf7 40.Bd4 Kg6 41.Kd6 Kf7

42.Kc6
Be2 43. Kb6 Ke6 44.Kc5 Kf7 45.Kd6 Bg4 46.Be5 Ke8 47.Kc7 Kf7 48.Kb6 Be2
49.Kc6 Bg4 50.Kd6 Ke8 51.Kd5 Kf7 52.Kd6 {draw by repetition}
1/2-1/2


Amusing game, as it definitely brings back memories of how well (or
badly?) I play. :-)

I analyzed the game with a slightly different personality, and at 60
seconds per move on a P4-2.4. Obviously, the more time you give the
engine to think about each move (and the faster the processor), the
better the analysis will be.

My summary came up as follows:

White Black
Book Move 2 1
Leave Book 0 1
CM9000 Agrees 44 42
CM9000 Disagrees 6 8
Agreement Pct. 88% 84%
Total Error 16.64 14.39
Relevant Error 6.69 7.12
Missed Mate 0 0
Moved Into Mate 0 0

This is similar to your results.

jm



  #9   Report Post  
Old February 14th 04, 05:43 PM
John Merlino
 
Posts: n/a
Default CM9000 Analysis

"Gregory Topov" wrote in message ...
"John Merlino" wrote in message:

CM9000 Agrees 44 42
CM9000 Disagrees 6 8
Agreement Pct. 88% 84%
Total Error 16.64 14.39
Relevant Error 6.69 7.12


Just to confirm that I understand you correctly, using your auto-analysis
results:
1. So these results mean that the 6 white moves that CM9000 "Disagreed"
with, happened to have a total error value of 16.64. CM9000 may have
selected a different move for the other 44 moves that white played , but the
difference in evaluation was so small it wasn't counted as a move to
"Disagree" with, nor towards Total Error.
2. Similarly, the 14.39 Total Error for black, was produced by the
evaluation difference for the 8 moves that CM9000 Disagreed with.
3. Of this 14.39, 14.39-7.12=7.28 were errors made when CM9000 regarded the
game as out of reach for black, and so they didn't count towards Relevant
Error.

Do I have it right?


That is absolutely correct.

jm
Reply
Thread Tools
Display Modes

Posting Rules

Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Teaching young children chess (also: CM9000 vs Fritz) Jestrada rec.games.chess.analysis (Chess Analysis) 0 January 8th 04 03:53 PM
Teaching young children chess (also: CM9000 vs Fritz) Jestrada rec.games.chess.computer (Computer Chess) 0 January 8th 04 03:53 PM
Analysis with Fritz 8 Marco rec.games.chess.computer (Computer Chess) 7 November 25th 03 11:57 AM
Chess Assistant 7 Interactive Analysis broken Alberich rec.games.chess.computer (Computer Chess) 3 September 16th 03 07:42 PM
Lasker-Steinitz analysis by Kasparov, Dvoretzky and Fritz Henri H. Arsenault rec.games.chess.analysis (Chess Analysis) 8 August 16th 03 01:42 PM


All times are GMT +1. The time now is 01:50 PM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Copyright 2004-2019 ChessBanter.
The comments are property of their posters.
 

About Us

"It's about Chess"

 

Copyright © 2017