A Chess forum. ChessBanter

If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.

Go Back   Home » ChessBanter forum » Chess Newsgroups » rec.games.chess.computer (Computer Chess)
Site Map Home Register Authors List Search Today's Posts Mark Forums Read Web Partners

Way to id games where result doesn't match position



 
 
Thread Tools Display Modes
  #1  
Old August 12th 03, 10:29 PM
Mhoulsby
external usenet poster
 
Posts: n/a
Default Way to id games where result doesn't match position

From: (Mike Ogush)
Date: 12/08/03 21:49 GMT Daylight Time
Message-id:

I would like to identify those games where the result was either
misreported or one side lost on time in a winning or drawn position
where the evaluation of the final game position does not match the
reported result.

I can see how to do this manually for each game by either:
1) loading the game into a chess engine
2) going to the last position,
3) having the engine analyze for a while and
4) me comparing the engines evaluation to the result
or,
1) converting the last position into EPD
2) loading the EPD into a chess engine
and doing steps 3 and 4above.

What I was wondering is if there is any utility/chess engine that
would do this automatically?


AFAIK there isn't. I'm pretty sure that if there was, then Knut Neven would
know about it, and use it. Knut told me that he spends an inordinate amount of
his time, iteration after iteration after iteration, doing manually precisely
what you describe. This for a database of around 3,000,000 games, poor guy!

http://members.shaw.ca/kneven/chess/databasing.html


I would just give it a set of games and
after a while it would return the subset of games, which had results
not matching the evaluation of the final position. I am OK with the
fact that there might be false positives because a computer might not
understand that the result is really accurate (due to the horizon
effect or lack of positional knowledge). I would like the analysis to
be done by as strong an engine as possible.

One extension that would be even better is a utility that would go
through every position in each game and let me know the first (if any)
position during the game that belied the game result.

Any pointers you could give me would be greatly appreciated.


I, too, am interested in doing this. However, I confess to harbouring doubts
about the existence of an effective strategy for achieving this goal. For
example: how, I wonder, does one tell when a result is *correct*,
*notwithstanding* that the final position contradicts the result (perhaps
because of a flagfall, or because Ivanchuk was one of the players... there is,
I believe, more than one example of his resigning a perfectly level postion,
simply because he made a mistake analysing the position in his head [I seem to
recall his having done this vs. Anand] or refusing to accept his opponent's
resignation on the grounds that the postion is level [he has done this at least
once, I don't remember who his opponent was on that occasion]). These sorts of
circumstance must occur frequently in rating-limited/junior tournaments and the
like, too....

This could be your life's work!

Enjoy!

Mark
  #2  
Old August 13th 03, 03:28 PM
Anders Thulin
external usenet poster
 
Posts: n/a
Default Way to id games where result doesn't match position



Mike Ogush wrote:

What I was wondering is if there is any utility/chess engine that
would do this automatically?


Don't know if anything lets you do it straight off the shelf.
But it shouldn't be too difficult to do this programmatically.

You need to get at the final recorded position as well as the result.
That should be possible to get at fairly easily -- I'd probably
look at Edwards SAN kit first before rolling my own. As long
as I don't have to verify that the moves are legal it's easy.
Then, spit out the EPD followed by result -- and possibly any other
identifying info.

Next, feed that (with suitable modifications into Crafty),
and record output. Any problems here? Don't think so.

Then go over both files, and correlate results. Report
discrepancies.

Choose another engine, and it will depend entirely on how
scriptable it is.

On a Unix system, I'd guess about a week's job (a few hours a day,
not a weeks 24/7 coding marathon?


--
Anders Thulin http://www.algonet.se/~ath

  #3  
Old August 13th 03, 10:42 PM
Mike Ogush
external usenet poster
 
Posts: n/a
Default Way to id games where result doesn't match position

On 13 Aug 2003 12:39:45 GMT, (NoKetch) wrote:

SNIPPED



Hi Mike
At
http://www.geocities.com/lyapko/winboard.htm
there is a program called lgpgnver.exe that checks pgn files for invalid result
claims. I think it works very good.

Dave

Dave


Hi Dave,

Your utility works well for situations when the final position does
not need to be analyzed to find a problem with the game result!
So I added it to my library of utilities for examining games.

However, the program I am looking for would do analysis of the final
position and notice any conflicts between the analysis results and the
game result. As an example I include a game of mine, where I I
offered a draw and it was accepted. The tool I want should notice
that Black can force checkmate from the final position and so the
position result (-+ or 0:1) conflicts with the game result (1/2-1/2).

Whipple (1535) - Ogush,M [D35]
Golden Gate Open San Francisco (1), 02.07.1978

1.d4 Nf6 2.c4 e6 3.Nc3 d5 4.Bg5 Nbd7 5.Nf3 c6 6.cxd5 exd5 7.e3 Bd6
8.Bd3 0-0 9.0-0 Re8 10.Re1 Nf8 11.h3 Ng6 12.Qc2 h6 13.Bxg6 hxg5 14.Bd3
g4 15.Ng5 gxh3 16.gxh3 Ne4 17.Ncxe4 dxe4 18.Nxe4 Bxh3 19.f3 Bb4 20.Re2
Re6 21.Rh2 Qh4 22.Kh1 Qh5 23.Qf2 Qh8 24.Rg1 Rh6 25.Qg3 Be6 26.Nf6+ Kf8
27.Nh7+ Ke7 28.Rxh6 gxh6 29.Qh2 Bd6 30.Qh5 Kd7 31.Rg2 Bxa2 32.Qxh6 Kc7
33.Qg5 Qd8 34.Qa5+ b6 35.Qxa2 Qh4+ 36.Kg1 Qe1+ 37.Bf1 Qxe3+ 38.Rf2
Rg8+ {Draw agreed} -

[BTW: There is an ammusing story regarding the game. When I was
analyzing it at home after the tournament I was berating myself for
having missed the winning move: 38...Bh2+ 39.Kxh2 Qxf2+ 40.Bg2 Qh4+
41.Kg1 Qxd4+ 42.Kf1 Qd3+ 43.Kf2 Qxh7 and Black is up the exchange and
two pawns Imagine my surprise when years later I first had a
computer analyze the game and it showed me the forced mate after
38...Rg8+. That sure let me know I continue to need to work on
tactics.]

Mike Ogush
  #4  
Old August 13th 03, 11:25 PM
Mark Houlsby
external usenet poster
 
Posts: n/a
Default Way to id games where result doesn't match position

Anders Thulin wrote in message ...
Mike Ogush wrote:

What I was wondering is if there is any utility/chess engine that
would do this automatically?


Don't know if anything lets you do it straight off the shelf.
But it shouldn't be too difficult to do this programmatically.

You need to get at the final recorded position as well as the result.
That should be possible to get at fairly easily -- I'd probably
look at Edwards SAN kit first before rolling my own. As long
as I don't have to verify that the moves are legal it's easy.
Then, spit out the EPD followed by result -- and possibly any other
identifying info.

Next, feed that (with suitable modifications into Crafty),
and record output. Any problems here? Don't think so.

Then go over both files, and correlate results. Report
discrepancies.

Choose another engine, and it will depend entirely on how
scriptable it is.


I'm sorry, I must be dumb or something, but I still don't see how this solves
the problem of games with respect to which the result *looks wrong* but is, in
fact, *correct*....

On a Unix system, I'd guess about a week's job (a few hours a day,
not a weeks 24/7 coding marathon?


For a database of around three million games?! Really?!

Interesting post, thanks.

Mark
  #5  
Old August 14th 03, 01:42 AM
Mike Ogush
external usenet poster
 
Posts: n/a
Default Way to id games where result doesn't match position

On 13 Aug 2003 15:25:50 -0700, (Mark
Houlsby) wrote:

Anders Thulin wrote in message ...
Mike Ogush wrote:

What I was wondering is if there is any utility/chess engine that
would do this automatically?


Don't know if anything lets you do it straight off the shelf.
But it shouldn't be too difficult to do this programmatically.

You need to get at the final recorded position as well as the result.
That should be possible to get at fairly easily -- I'd probably
look at Edwards SAN kit first before rolling my own. As long
as I don't have to verify that the moves are legal it's easy.
Then, spit out the EPD followed by result -- and possibly any other
identifying info.

Next, feed that (with suitable modifications into Crafty),
and record output. Any problems here? Don't think so.

Then go over both files, and correlate results. Report
discrepancies.

Choose another engine, and it will depend entirely on how
scriptable it is.


I'm sorry, I must be dumb or something, but I still don't see how this solves
the problem of games with respect to which the result *looks wrong* but is, in
fact, *correct*....


Anders solution is one that would solve the problem I laid out. I was
just hpoing that it already existed so I didn't have to do the
programming.

I don't actually care if the game result is "correct" or not. What I
want this utility for is to identify those games where the game result
distorts the evaluation of a particular opening line (either in the
statistics that a database might produce for all games in that line or
in the backsolving done by BookUp or BookBuilder) when the actual
positon at game end would not distort the evaluation. I hope this
clarifies what I am asking for.


On a Unix system, I'd guess about a week's job (a few hours a day,
not a weeks 24/7 coding marathon?


For a database of around three million games?! Really?!


I think Anders was referring to the time to code the script/program
NOT the time to run it over a large game collection. If I limited the
comparison to only the last position in the game it would take a
little under 2 years per million games, assuming the end postion of
each game was analyzed for 1 minute. I could analyze the position
for a shorter time, but the engine would be giving poorer evaluations
as I shortedned the time.

As you might guess I was not planning to do this for every game in my
collection. I was only going to run it over a few thousand games in
the opening lines that I am studying.

Interesting post, thanks.

Mark


Mike
  #6  
Old August 15th 03, 09:36 PM
Anders Thulin
external usenet poster
 
Posts: n/a
Default Way to id games where result doesn't match position



Mike Ogush wrote:

I don't actually care if the game result is "correct" or not. What I
want this utility for is to identify those games where the game result
distorts the evaluation of a particular opening line (either in the
statistics that a database might produce for all games in that line or
in the backsolving done by BookUp or BookBuilder) when the actual
positon at game end would not distort the evaluation. I hope this
clarifies what I am asking for.


That would probably speed up the actual process of evaluating the
positions quite considerably, as you don't have to bother about
positions far outside the openings database you're working with.

There's a problem, though: stupid mistakes may produce games
in which the end result is fully correct, but where a minimum of
dispassionate analysis a few moves earlier would have produced
a different outcome. That kind of bias would probably also
distort your openings ...

--
Anders Thulin http://www.algonet.se/~ath

 




Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


All times are GMT +1. The time now is 09:47 AM.


Powered by vBulletin® Version 3.6.4
Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.Content Relevant URLs by vBSEO 2.4.0
Copyright 2004-2017 ChessBanter.
The comments are property of their posters.