Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old May 12th 04, 06:54 AM
Sidney
 
Posts: n/a
Default Freeware rating evaluator?

I'm looking for a freeware program which can evaluate the approx. USCF or FIDE
Elo rating of a player by looking at their moves in a PGN.

Is there any such program?


  #2   Report Post  
Old May 12th 04, 09:43 AM
David Richerby
 
Posts: n/a
Default Freeware rating evaluator?

Sidney wrote:
I'm looking for a freeware program which can evaluate the approx. USCF
or FIDE Elo rating of a player by looking at their moves in a PGN.

Is there any such program?


I very much doubt it. Producing such a program would involve a massive
statistical analysis of games played by all reasonable combinations of
strengths of players and it's not at all clear to me that one would be
able to estimate the rating to within reasonable margins of error just
from looking at moves.


Dave.

--
David Richerby Devil Tree (TM): it's like a tree
www.chiark.greenend.org.uk/~davidr/ that's possessed by Satan!
  #3   Report Post  
Old May 14th 04, 07:23 AM
Akorps666
 
Posts: n/a
Default Freeware rating evaluator?

It would be a good project to make one though

  #4   Report Post  
Old May 14th 04, 10:05 AM
David Richerby
 
Posts: n/a
Default Freeware rating evaluator?

Akorps666 wrote:
It would be a good project to make one though


Only if the idea is sound, which I don't think it is. I can play, shall
we say, the first three moves as well as any Grandmaster alive or dead;
likewise various trivial endgames such as KRRk. This shows that there is
no clear link between individual moves and rating. Of course, there must
be some link but it is a very subtle and subtle correlations require a lot
of data to acquire any kind of certainty. What would be the point of
writing this software and finding that you need to feed it a thousand
games to get a rating more accurate than +/-250 points?


Dave.

--
David Richerby Old-Fashioned Permanent Shack (TM):
www.chiark.greenend.org.uk/~davidr/ it's like a house in the woods but
it'll be there for ever and it's
perfect for your grandparents!
  #5   Report Post  
Old May 14th 04, 10:13 AM
yves
 
Posts: n/a
Default Freeware rating evaluator?

Sidney wrote:

Is there any such program?


Not that I am aware off.

Chessbase does include some test suites to mesure your performance by
solving specific problems.

Some programs also include some guidance to trace blunders. So they
know when performace is bad.

For chess programs, there exist a series of tests to measure their
performance. All of these are based on specific positions and finding
of the single best move in these positions. None are based on selecting
a good move among a list of possible good moves.


  #6   Report Post  
Old May 14th 04, 05:28 PM
David Kane
 
Posts: n/a
Default Freeware rating evaluator?


"David Richerby" wrote in message
...
Akorps666 wrote:
It would be a good project to make one though


Only if the idea is sound, which I don't think it is. I can play, shall
we say, the first three moves as well as any Grandmaster alive or dead;
likewise various trivial endgames such as KRRk. This shows that there is
no clear link between individual moves and rating.


Every move you make contains some information about your strength.
While KRRk wouldn't distinguish you from a grandmaster, it would give
information about your strength compared to someone rated, eg.500.
And if you dropped a pawn at 6-ply, that kind of error would be rare
for a GM, so the rating estimate would become more accurate with
each move evaluated.


Of course, there must
be some link but it is a very subtle and subtle correlations require a lot
of data to acquire any kind of certainty. What would be the point of
writing this software and finding that you need to feed it a thousand
games to get a rating more accurate than +/-250 points?


The point is that the moves of a game (along with time control) contain
much more information about your strength than the mere result, so
that you should be able to obtain a more accurate rating in *fewer*
games than it would take looking only at the result.

Determining some theoretically optimal way to convert moves played
to winning probability could take a lifetime of work. But the threshold
for being useful is only that it does so better than result rating - a
very low target.

DK


  #7   Report Post  
Old May 14th 04, 11:13 PM
Akorps666
 
Posts: n/a
Default Freeware rating evaluator?

The only practical metric I can think of would be frequency of tactical
blunders. Nunn did a comparison of Carlsbad 1911 with a modern tournament and
found the play in the older tournament was much weaker.

So the only idea I have, which may not work, would be to run a bunch of games
of 2800 players through a test to see how frequently they blunder, run a bunch
of games of 2700 players through a test to see how frequently the blunder ...
and hope that some kind of curve correlating frequency of blunders and ratings
comes out and is useful. There are all kinds of problems that might turn up
that could sink this approach. But if some useful correlation does appear, then
one could go back in time and by frequency of blunders try to estimate the
ratings of players from the olden days.

Some obvious problems a

false positives (moves which the engines say are blunders but which aren't.
Examining Tal's games, sometimes it takes quite awhile before one can prove
that what the engines think is a blunder is actually a fantastically deep
correct sacrifice.) It may take a strong human player examining the game to
catch some of these, though some of the engines, such as Shredder 8, appear to
be getting very strong.

player style: sharp players may have more blunders than stodgy players of the
same rating simply due to taking more risk. A possible way around this is to
work in the frequency of draws in a player's games into the statistics.

strategic understanding: it takes much more computer time to decide on the best
strategic moves, if that is even possible. So far Shredder 8 and Hiarcs 9 seem
to be best at finding such moves.

tablebases: there aren't many games which hit the tablebases unfortunately,
otherwise we would have a completely objective metric for some positions. It is
interesting to go back though, I found some errors Capablanca made in a rook
ending using the tablebases.

  #8   Report Post  
Old May 15th 04, 08:44 AM
Solomon Nuffert
 
Posts: n/a
Default Freeware rating evaluator?

Hi!

I think to create the rough, imperfect tool for definition of chess strength
rather simply, modifying an option " the analysis of game "
Let's assume, we have chess program FX, which plays valid 2.500
We give FX to analyze game.
In a position Y best move from the point of view of the program FX conduct
to an evaluation = 0.00
If the player has made this move, to him the strength 2.500 is given. If he
has made a move, then the position is estimated in -0.10 his strength
2.400 - 0.20 = 2.300 - 1.00 = 1.500 and so on.
The move after move FX estimated all game.
Strength are summarized and is deduced average arithmetic(or geometric). The
more games, the more precisely evaluation of chess strength.

--
Regards
SoloMoon



Sidney пишет в
t...
I'm looking for a freeware program which can evaluate the approx. USCF or

FIDE
Elo rating of a player by looking at their moves in a PGN.

Is there any such program?






  #9   Report Post  
Old May 15th 04, 12:27 PM
Terry
 
Posts: n/a
Default Freeware rating evaluator?


"Solomon Nuffert" wrote in message
...
Hi!

I think to create the rough, imperfect tool for definition of chess

strength
rather simply, modifying an option " the analysis of game "
Let's assume, we have chess program FX, which plays valid 2.500
We give FX to analyze game.
In a position Y best move from the point of view of the program FX conduct
to an evaluation = 0.00
If the player has made this move, to him the strength 2.500 is given. If

he
has made a move, then the position is estimated in -0.10 his strength
2.400 - 0.20 = 2.300 - 1.00 = 1.500 and so on.
The move after move FX estimated all game.
Strength are summarized and is deduced average arithmetic(or geometric).

The
more games, the more precisely evaluation of chess strength.

--
Regards
SoloMoon



Sidney пишет в
t...
I'm looking for a freeware program which can evaluate the approx. USCF

or
FIDE
Elo rating of a player by looking at their moves in a PGN.

Is there any such program?




The only part of chess computers can rate is tactics only.
They are hopeless at positional play. If a comp rates your move
at 0.10 instead of 0.0 - so what ?

Regards


  #10   Report Post  
Old May 15th 04, 04:21 PM
David Kane
 
Posts: n/a
Default Freeware rating evaluator?


"Akorps666" wrote in message
...
The only practical metric I can think of would be frequency of tactical
blunders. Nunn did a comparison of Carlsbad 1911 with a modern tournament

and
found the play in the older tournament was much weaker.

So the only idea I have, which may not work, would be to run a bunch of

games
of 2800 players through a test to see how frequently they blunder, run a

bunch
of games of 2700 players through a test to see how frequently the blunder

....
and hope that some kind of curve correlating frequency of blunders and

ratings
comes out and is useful.


I think you would need to quantify the blunders. E.g. determine the blunder
rate
vs. ply depth for each class of player and then fit player data to it. (Time
control
would have to be a variable of course) Another idea, using all moves (not
just
blunders), would be to have a number of algorithms simulating play at each
class,
and then see which fits the moves best.

DK


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
Suggestion for Blitzin or similar - hide opponents rating. Dr. David Kirkby rec.games.chess.computer (Computer Chess) 3 May 3rd 04 03:21 AM
USCF Tournament rating fees Ivan rec.games.chess.analysis (Chess Analysis) 17 February 25th 04 02:47 AM
Can FIDE really rate players down to 1001 ?? Sam Sloan rec.games.chess.computer (Computer Chess) 9 January 28th 04 09:02 PM
number of players at each FIDE rating level gec rec.games.chess.analysis (Chess Analysis) 10 November 13th 03 12:06 AM
USCF rating floors... Howard Goldowsky rec.games.chess.analysis (Chess Analysis) 2 August 19th 03 04:49 PM


All times are GMT +1. The time now is 11:12 AM.

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