are there already programs/projects to calculate a rating
by looking at the moves of a particular game ?

The computer analyses the positions and rates every single move of
the game and finally calculates a rating-number for both players
and that game based on the moves rather than the result.

The program will be optimised so that the overall rating
difference of a players reflect the expectation value
of his points in a tournament.

(Sterten) wrote in rec.games.chess.misc:

are there already programs/projects to calculate a rating
by looking at the moves of a particular game ?

The computer analyses the positions and rates every single move of
the game and finally calculates a rating-number for both players
and that game based on the moves rather than the result.

The program will be optimised so that the overall rating
difference of a players reflect the expectation value
of his points in a tournament.


Rating is not an absolute value, it's establishing differences in
strenght, or even better, predicting win/lose scores.
The value depends on the pool of players, not on the strenght of
individual moves.
A 2100 player in pool one can be as strong as a 1900 player in pool two.

Programs that use rating features use personalities or values that have
been tested against players with a certain rating and see how much the
results differ from their actual rating. Sometimes ratings come close to
your own OTB ratings, but more than often they don't, because they're
tested against players from other pools.


--
CeeBee


Uxbridge: "By God, sir, I've lost my leg!"
Wellington: "By God, sir, so you have!"


Google CeeBee @ www.geocities.com/ceebee_2

are there already programs/projects to calculate a rating
by looking at the moves of a particular game ?

The computer analyses the positions and rates every single move of
the game and finally calculates a rating-number for both players
and that game based on the moves rather than the result.

The program will be optimised so that the overall rating
difference of a players reflect the expectation value
of his points in a tournament.


Rating is not an absolute value, it's establishing differences in
strenght, or even better, predicting win/lose scores.
The value depends on the pool of players, not on the strenght of
individual moves.

with this proposed move-rating system, rating
becomes more absolute.
You can even rate two players who both play
their very first game.

(Sterten) wrote in rec.games.chess.misc:


with this proposed move-rating system, rating
becomes more absolute.
You can even rate two players who both play
their very first game.

And how do you get to valueing those individual moves? As soon as you
state that "solving this move is equivalent to solving it by a GM of Elo
2550." As soon as you have done that, you have started to work with the
elementary thing about ratings: differences.

What you propose is like determining the top speed of a car relative to a
lot of other cars we know the performance of. The result again is not an
absolute value, but a difference in speed.

You are certainly not the first, and will certainly not be the last person
who wants to "benchmark" individual player's strength, independent from
tournament and OTB performance.

In what way you use a rating system like you propose, you always use a
rating system based on differences. It will always come back to your
calculations.



--
CeeBee


Uxbridge: "By God, sir, I've lost my leg!"
Wellington: "By God, sir, so you have!"


Google CeeBee @ www.geocities.com/ceebee_2

(Sterten) wrote in rec.games.chess.misc:


with this proposed move-rating system, rating
becomes more absolute.
You can even rate two players who both play
their very first game.

And how do you get to valueing those individual moves? As soon as you

by a strong chess-computer. Sorry if this was unclear.

state that "solving this move is equivalent to solving it by a GM of Elo
2550." As soon as you have done that, you have started to work with the
elementary thing about ratings: differences.

What you propose is like determining the top speed of a car relative to a
lot of other cars we know the performance of. The result again is not an
absolute value, but a difference in speed.

You are certainly not the first, and will certainly not be the last person
who wants to "benchmark" individual player's strength, independent from
tournament and OTB performance.

it's the first time in history, that we have the tools , i.e. strong
and fast chesscomputers, available now

In what way you use a rating system like you propose, you always use a
rating system based on differences. It will always come back to your
calculations.

right. But you fix the base _once_ .
Then you can compare ancient players,computers,players from
separated islands, ET-players etc. all with one system.
You no longer have to care about disconnected isolated
components in the multigraph of "chess-players".

Also the process to improve this sort of rating will probably
also lead to better chess-computers and their better understanding
of human moves.

"Sterten" wrote in message
...
(Sterten) wrote in rec.games.chess.misc:


with this proposed move-rating system, rating
becomes more absolute.
You can even rate two players who both play
their very first game.

And how do you get to valueing those individual moves? As soon as you

by a strong chess-computer. Sorry if this was unclear.

state that "solving this move is equivalent to solving it by a GM of Elo
2550." As soon as you have done that, you have started to work with the
elementary thing about ratings: differences.

What you propose is like determining the top speed of a car relative to
a
lot of other cars we know the performance of. The result again is not an
absolute value, but a difference in speed.

You are certainly not the first, and will certainly not be the last
person
who wants to "benchmark" individual player's strength, independent from
tournament and OTB performance.

it's the first time in history, that we have the tools , i.e. strong
and fast chesscomputers, available now

In what way you use a rating system like you propose, you always use a
rating system based on differences. It will always come back to your
calculations.

right. But you fix the base _once_ .
Then you can compare ancient players,computers,players from
separated islands, ET-players etc. all with one system.
You no longer have to care about disconnected isolated
components in the multigraph of "chess-players".

Also the process to improve this sort of rating will probably
also lead to better chess-computers and their better understanding
of human moves.

You seem unaware of the strengths and weaknesses of computers
playing chess.

Their main strength is their ability to calculate tactically - ie look
ahead
and count the pieces.
The weaknesses are long range planning, positional play and endgames.

So if you try to rate human play on a move by move basis the only feature
you can possibly rate is their tactical strength. Computers are very weak
on all the other features.

Regards

Terry

(Sterten) wrote in rec.games.chess.misc:


by a strong chess-computer. Sorry if this was unclear.

snip

right. But you fix the base _once_ .
Then you can compare ancient players,computers,players from
separated islands, ET-players etc. all with one system.
You no longer have to care about disconnected isolated
components in the multigraph of "chess-players".

Also the process to improve this sort of rating will probably
also lead to better chess-computers and their better understanding
of human moves.


Regardless of the limited value of the test results you will get (mostly
tactical ability) you suggest this sytem would be an improvement.
You state that benchmarking against a standard tells you more about
differences in strenght between two players than actual play between two
players.

However this is not the case. You simply mimic a real life system in a
rather crude way. Your results will _not_ be better than real life
ratings, simply because you fail to see that a rating is based on
differences, not on benchmarks.

--
CeeBee


Uxbridge: "By God, sir, I've lost my leg!"
Wellington: "By God, sir, so you have!"


Google CeeBee @ www.geocities.com/ceebee_2

"CeeBee" wrote in message
. 6.83...
(Sterten) wrote in rec.games.chess.misc:


by a strong chess-computer. Sorry if this was unclear.

snip

right. But you fix the base _once_ .
Then you can compare ancient players,computers,players from
separated islands, ET-players etc. all with one system.
You no longer have to care about disconnected isolated
components in the multigraph of "chess-players".

Also the process to improve this sort of rating will probably
also lead to better chess-computers and their better understanding
of human moves.


Regardless of the limited value of the test results you will get (mostly
tactical ability) you suggest this sytem would be an improvement.
You state that benchmarking against a standard tells you more about
differences in strenght between two players than actual play between two
players.

However this is not the case. You simply mimic a real life system in a
rather crude way. Your results will _not_ be better than real life
ratings, simply because you fail to see that a rating is based on
differences, not on benchmarks.

--
CeeBee


I have to agree with your views on this.

Regards

Terry

..
It is ohhh so easy to shoot-down an idea which may not be perfect in
practice!




But the idea of objectively measuring, not merely final results of games, but
overall accuracy of moves, is a legitimate one.

I would caution against, say, using the endgame tablebases to judge the
accuracy of a player's execution of the final phase, for example. Humans are
not always interested in following the fastest winning line, and any attempt to
judge them which incorporates such an assumption is faulty.
As for the openings, even the strongest chess computers of today rely upon
"book," making them fallible. And again, humans may on occasion,
*deliberately* choose what they know to be inferior moves, for the distinct
purpose of exploration of the unknown, or exploitation of a particular
opponent's perceived weaknesses. Or perhaps, even to keep their "best" lines
hidden, to be used another day, or against a more dangerous rival.
As for the middlegame, or simply tactics, where no book or tablebase can
provide the "correct" answers, a computer may very well be able to rate the
caliber of individual moves, and sum them up even better than the actual
outcome of any particular game or few games.

The chief advantage of such measurements would, of course, be the ability to
objectively compare the play of players from different generations. If say,
Mongredien, required two pawns up and a hundred moves to force a win against
poor defense, while Kasparov consistently wins against poor defense with no
advantage whatever, then we may have learned something about the relative
strengths of these two players. If Mongredien misses two-move combo's half the
time, while Kasparov misses them just two percent of the time, we may be able
to affix a meaningful description of this difference; similar to the ratings of
actual games, which have the problem that many, many players did not play
enough against their contemporaries to validate results.



Unlike the flying ceebee and terrybean, I am unafraid of such an *objective*
approach to my own games. I do not consider my actual results to have been the
result of luck, and I think an objective evaluation like this of my individual
moves would only serve to confirm, rather than call into doubt, my miserable
actual results OTB. :-) But again I must stress, that a few of my inferior
moves were selected *deliberately*, and so should not count against me the same
as my oversights.




1.h4! (tick, tick)

"NoMoreChess" wrote in message
...
.
It is ohhh so easy to shoot-down an idea which may not be perfect in
practice!

But the idea of objectively measuring, not merely final results of
games, but
overall accuracy of moves, is a legitimate one.

The idea is sound enough but there is no way of doing it. Computers cannot
be used.


I would caution against, say, using the endgame tablebases to judge the
accuracy of a player's execution of the final phase, for example. Humans
are
not always interested in following the fastest winning line, and any
attempt to
judge them which incorporates such an assumption is faulty.

Agreed. Remember that computer databases only evaluate endgames with very
few pieces on the board.

As for the openings, even the strongest chess computers of today rely
upon
"book," making them fallible. And again, humans may on occasion,
*deliberately* choose what they know to be inferior moves, for the
distinct
purpose of exploration of the unknown, or exploitation of a particular
opponent's perceived weaknesses. Or perhaps, even to keep their "best"
lines
hidden, to be used another day, or against a more dangerous rival.

Agreed.

As for the middlegame, or simply tactics, where no book or tablebase can
provide the "correct" answers, a computer may very well be able to rate
the
caliber of individual moves, and sum them up even better than the actual
outcome of any particular game or few games.

Disagree. Computers can only assess the tactics in a position. Assessing
positional moves and long term planning is too difficult for computers.


The chief advantage of such measurements would, of course, be the
ability to
objectively compare the play of players from different generations. If
say,
Mongredien, required two pawns up and a hundred moves to force a win
against
poor defense, while Kasparov consistently wins against poor defense with
no
advantage whatever, then we may have learned something about the relative
strengths of these two players. If Mongredien misses two-move combo's
half the
time, while Kasparov misses them just two percent of the time, we may be
able
to affix a meaningful description of this difference; similar to the
ratings of
actual games, which have the problem that many, many players did not play
enough against their contemporaries to validate results.

Once again you are only assessing tactics. The stronger players will take
the simplist way to victory not the quickest.




Unlike the flying ceebee and terrybean, I am unafraid of such an
*objective*
approach to my own games. I do not consider my actual results to have
been the
result of luck, and I think an objective evaluation like this of my
individual
moves would only serve to confirm, rather than call into doubt, my
miserable
actual results OTB. :-) But again I must stress, that a few of my
inferior
moves were selected *deliberately*, and so should not count against me the
same
as my oversights.


I dont think ceebee or myself are afraid of objective analysis of our
games - but the tools do not exist.

Ceebee also makes the good point that a rating is based on
differences, not on benchmarks. A 1500 rating in one rating pool is not the
same as 1500 in another rating pool. If you take 100 USCF 1500 players and
get
them to play each other with the rating starting from scratch. Some will
finish up with 2200
ratings. These 2200 players will not compare very well with FIDE 2200
players.

So - what is a true 2200 player ?. There isnt one - it depends on the other
players
in the same pool.

The original poster didn't understand rating systems or computer chess.
He/she assumed that computers were good at all aspects of the game and
this simply isnt true.

Regards