Page Index Toggle Pages: 1 [2] 3 
Topic Tools
Very Hot Topic (More than 25 Replies) Program to assess playing strength? (Read 18617 times)
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #16 - 11/13/13 at 05:32:31
Post Tools
I think both Regan and Guid/Bratko tweak their formulas to avoid penalizing "suboptimal" but still winning moves in already winning positions. Don't know the technical details though.

Nice story about the Taimanov book. An underappreciated player who contribuded a lot to White's treatment of the Nimzo-Indian, King's Indian and Benoni. I've got Alekhine's and Smyslov's game collections lined up for study in few months' time; I'll try not to rely too much on silicon.
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
ErictheRed
God Member
*****
Offline


USCF National Master

Posts: 2534
Location: USA
Joined: 10/02/05
Re: Program to assess playing strength?
Reply #15 - 11/13/13 at 04:03:42
Post Tools
In general, one problem I have with a lot of this stuff is in evaluating "technique."  If the game is very complex I'm sure that comparing moves to a computer engine's output is a pretty good way of evaluating the strength of the move.  But what about those pawn-up rook endings with pawns on both sides of the board, etc?  Or endings where you give up material but head for a theoretical win 20 moves away?  It's outside of the computer's horizon, and maybe the computer sees a faster win, but would you really punish someone for playing the easiest win?

That's where a lot of subjectivity will come in, IMO.  So it'd be harder to automate than with football, but if it's just something you want to do on your own with human input for a limited number of games, it's probably pretty accurate.  

This reminds me: way back in 1999 I bought my first PC, and I picked up a copy of Fritz 6 as well as Taimanov's Selected Games.  I fired up Fritz 6 and went through the first game in the book with it, thinking that it'd be an incredible experience, that I'd understand the game more deeply, etc. 

Instead, it just disagreed with all of Taimanov's evaluations, exclams, question marks, etc.  Eventually I realized that I wasn't learning anything from looking at Fritz's output and just read the book.

Engine's are much better now, and I don't mean that as any kind of retort to your efforts, Stigma.  This conversation just reminded me of that experience.
  
Back to top
 
IP Logged
 
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #14 - 11/13/13 at 03:53:56
Post Tools
Of course it would be nice to automate this and apply it to lots of games and players. But that's just not what I'm trying to do, I'm content with analyzing myself and maybe a student or two.

The big issue is finding a way to usefully measure complexity/difficulty (not sure if these should be considered the same or two different factors) since playing complex chess is a good strategy against other humans, even at some cost to objective move strength. Chess-db calculates players' "exchanging tendency", which is related but not the same thing.
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
ErictheRed
God Member
*****
Offline


USCF National Master

Posts: 2534
Location: USA
Joined: 10/02/05
Re: Program to assess playing strength?
Reply #13 - 11/13/13 at 03:41:09
Post Tools
Well okay Stigma you're probably right, and of course it's similar to what's been done at chess-db.com.  But as a scientist/engineer I'd want to see lots of data that can predict future performance, etc., without knowledge of ratings.  I'm probably just being a bit obtuse because of my background.
  
Back to top
 
IP Logged
 
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #12 - 11/12/13 at 22:36:48
Post Tools
Well I'm not doing a scientific research project here, just trying to find out how well I play in different games and different kinds of positions, and what aspects of my play are improving (if any). Why can't I subjectively assess game complexity on a 10-point scale and use that in my calculations? That's what I already do with tiredness, and it's not "a lot" of human input, just a single number.

There may well be some objective meausere of complexity and/or difficulty too; I think I saw difference in evaluation between the top candidate moves used as a criterion in some paper.

Certainly if someone manages to play either long games or very complex games (or both) and at the same time not depart too much from the computers' evaluations, that translates to strong play (and will lead to a high rating over time).
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
ErictheRed
God Member
*****
Offline


USCF National Master

Posts: 2534
Location: USA
Joined: 10/02/05
Re: Program to assess playing strength?
Reply #11 - 11/12/13 at 18:32:28
Post Tools
Stigma wrote on 11/11/13 at 19:38:02:
ErictheRed, aren't you making it more complicated than it needs to be? 

I'm of course not saying that you can realistically evaluate a player or even an individual game on this % metric and the moves that deviate from "best play" alone. But if you factor in game length, a subjective measure of game complexity plus (when available) knowledge of the player's clock times, opening preparation and recent training work, aren't we getting somewhere?


You just said "subjective," so I'm not sure that we're getting somewhere or not.  Perhaps the answer is yes, but it would still take a ton of human judgment.  It's not the kind of thing that you can enter the box score of a game--like in football--and have it all automated. 

So I just don't see how useful that would be; it's still very subjective and needs a lot of human input.
  
Back to top
 
IP Logged
 
Jupp53
God Member
*****
Offline


be

Posts: 988
Location: Frankfurt/Main
Joined: 01/04/09
Gender: Male
Re: Program to assess playing strength?
Reply #10 - 11/12/13 at 00:05:19
Post Tools
You could use SCID. There's an autoanalyze option were you can set a treshhold for the difference between your move and the engine move. And SCID allows to autoanalyze a complete file of games. This is a feature I didn't find neither in ChessBase nor in Aquarium or ChessAssistant.
  

Medical textbooks say I should be dead since April 2002.
Dum spiro spero. Smiley
Narcissm is the humans primary disease.
Back to top
 
IP Logged
 
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #9 - 11/11/13 at 19:41:51
Post Tools
P.S. I've also made habit of always registering how tired i felt during a game since I've had some sleep problems in recent years. Both wakefulness and tactical sharpness are important short-term factors that affect my results.
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #8 - 11/11/13 at 19:38:02
Post Tools
ErictheRed, aren't you making it more complicated than it needs to be? 

I'm of course not saying that you can realistically evaluate a player or even an individual game on this % metric and the moves that deviate from "best play" alone. But if you factor in game length, a subjective measure of game complexity plus (when available) knowledge of the player's clock times, opening preparation and recent training work, aren't we getting somewhere?
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
ErictheRed
God Member
*****
Offline


USCF National Master

Posts: 2534
Location: USA
Joined: 10/02/05
Re: Program to assess playing strength?
Reply #7 - 11/11/13 at 16:23:24
Post Tools
I got interested in a lot of advanced metrics in football (American), for instance F/+: http://www.footballoutsiders.com/stats/feiplus. ;

This works in American Football because you (basically) have two teams that alternate doing the same thing over and over for 60 minutes.  There are a lot of ways to evaluate how efficiently each team is doing that same thing (trying to move the ball down field and score).

I just don't see how something like that is possible in chess, where you don't do the same thing over and over.  Some games you sacrifice all your material to force mate, other times you win a pawn and play "boringly" for 40 moves but the outcome of the game was never in doubt, etc.  

I'm just not sure how to evaluate chess players' moves absolutely without a TON of subjective human input, other than what has already been done (comparing moves to engine output).
  
Back to top
 
IP Logged
 
Vass
God Member
*****
Offline


I Love ChessPublishing!

Posts: 1105
Joined: 06/22/11
Re: Program to assess playing strength?
Reply #6 - 11/11/13 at 09:21:21
Post Tools
Stigma wrote on 11/11/13 at 06:31:57:
...
But then I noticed one player at 94% - guess who... http://chess-db.com/public/analyzedgames.jsp?id=2903741

Heh! Our hero.. The usual suspect.  Cool
  
Back to top
 
IP Logged
 
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #5 - 11/11/13 at 06:31:57
Post Tools
I thought the world's top ranked players would have the highest % scores on chess-db. Solid players like Carlsen, Kramnik and Leko are around 90% and Nakamura above 91% - very impressive for such a sharp player (this method will typically overvalue careful, solid players and deep opening preparation).

But then I noticed one player at 94% - guess who... http://chess-db.com/public/analyzedgames.jsp?id=2903741
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #4 - 11/11/13 at 06:17:23
Post Tools
I forgot to mention it, but I did find a website that does essentially what I wanted to do:

http://chess-db.com/public/index.jsp

It displays games (analyzed with Stockfish I believe) with a % score (rather than a rating) for both sides' play. The graphic showing which moves departed from best play, and by how much, are also very useful (though similar to the graphs in Fritz and other playing GUIs).

I would still prefer to be able to do this myself in a program, but this is not a bad substitute. Though not that many of my games are in their database yet. They really should expand it.
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
brabo
God Member
*****
Offline


Welcome chessfriend

Posts: 1073
Joined: 02/02/07
Re: Program to assess playing strength?
Reply #3 - 10/09/12 at 12:32:49
Post Tools
Stigma wrote on 10/09/12 at 12:18:40:

I could imagine some program like Fritz, Rybka or Chess Assistant analyzing the moves of a game as input (with an engine of my choice), and then giving as output the estimated rating of the players. And maybe even set it to work on all the games of a player from a tournament, as a series. 

If there really is no such feature available, I will just have to learn some programming!

No such feature is commercially available. At least I am not aware and I am pretty acquainted with all kind of engines and programs.

Quote:

Ken Regan's method does this with large numbers of games at the same time, but I don't expect that to be commercially available.

I have read his work. I understood that you need a lot of HW for that, a lot of time and he uses a self written non commercial program. So only for private use.

Quote:

The practical use I'm thinking of is simply to determine if a player (me or a student) is really improving, i.e. if training is having the desired effect on playing strength. Performance ratings alone may depend too much on chance for this use, at least if I'm stuck with a small sample.

Personally I think it is wrong to check after each tournament if you improved. You should take longer periods (e.g. 6 months) to make serious conclusions. It is like you check every day your weight when you are on diet. That is also not recommended as the weightdifference between 2 days doesn't say anything about the success of the diet.
  
Back to top
WWW  
IP Logged
 
Stigma
God Member
*****
Offline


There is a crack in everything.

Posts: 3277
Joined: 11/07/06
Gender: Male
Re: Program to assess playing strength?
Reply #2 - 10/09/12 at 12:18:40
Post Tools
brabo wrote on 10/09/12 at 10:51:36:

I doubt very much that this is available at least not as public software. I also don't see the reason to make this as it is impracticable to employ. To know somebodies strength in a tournament is only relevant immediately after the tournament and you can't achieve that by creating a GUI/engine.


I could imagine some program like Fritz, Rybka or Chess Assistant analyzing the moves of a game as input (with an engine of my choice), and then giving as output the estimated rating of the players. And maybe even set it to work on all the games of a player from a tournament, as a series. 

If there really is no such feature available, I will just have to learn some programming! Ken Regan's method does this with large numbers of games at the same time, but I don't expect that to be commercially available.

The practical use I'm thinking of is simply to determine if a player (me or a student) is really improving, i.e. if training is having the desired effect on playing strength. Performance ratings alone may depend too much on chance for this use, at least if I'm stuck with a small sample.
  

Improvement begins at the edge of your comfort zone. -Jonathan Rowson
Back to top
 
IP Logged
 
Page Index Toggle Pages: 1 [2] 3 
Topic Tools
Bookmarks: del.icio.us Digg Facebook Google Google+ Linked in reddit StumbleUpon Twitter Yahoo