|
Main
Date: 04 Apr 2005 01:13:10
From: Bark!
Subject: Rating from moves??
|
Is it posible to somehow get unrated player's ELO from the moves he plays? To be more concrete, I'd like to get my opponent's rating based on game(s) that we played against each other so I can review my play against specific sort of opponents (~1200, ~1600.. etc.) Thanx in advance
|
|
|
Date: 09 Apr 2005 08:57:10
From: tom_h
Subject: Re: Rating from moves??
|
David Kane wrote: > "CeeBee" <[email protected]> wrote in message > news:[email protected]... > > "David Kane" <[email protected]> wrote in > > rec.games.chess.analysis: > > > > > > > So you're back to the extreme claim that knowing the > > > moves has zero predictive value? > > > > I can't go back to a claim I never made, neither normal, nor > > extreme. > > > > We discussed the question "is a move based rating system better than > > a result based system". > > > > > No. This is a clear cut case of moves being better than results. > > > > So you basically don't have to prove it because it is a clear cut > > case. > > > > I see. > > > > What is your answer? In the scenario I gave, > who will do a better job of handicapping the > next games, the person with the moves of the > earlier games or the person with the results > of the games? > > Your extreme defense of Elo rating > seems to stem from a misunderstanding > of what result-based rating is all about. I apologize if it looks like I am inserting myself into this conversation out of nowhere... and if it looks like that it is probably because I am guilty as charged (-: However, due to not having a life and trying to do something to distract myself from studying, I spent the last 30 min or so reading most of this thread, and I wanted to add my 2 cents... 1) You guys - CeeBee and David - probably agree with each other more than you disagree. I suspect that although David is taking the theoretical position that the moves contain more information than the results, he might concede that there are so many variables as to how to rank a chess move as to make this very hard to accomplish in practice. For example, if somebody wins a game because of a theoretically bad but practically advantageous move, then the software that we are using to generate ratings should reflect the practical advantages of that move... however this is very hard to do. I am sure that CeeBee, for his part, understands that good moves correlate with good results to at least some degree (-: He just thinks that there are other off-the-board aspects that somehow are not reflected in the record of the moves, but which do contribute to the result. I am just trying to paraphrase in a way that emphasizes that the viewpoints are not completely different... feel free to correct me if I am wrong. 2) I tend to agree with David if we are talking about pure theory... you should be able to devise a more accurate rating for someone by including *all* of the information available, such as results, moves, maybe even the person's demeanor at the board. David has given a few examples already that show that, but I wanted to add another one from a sort of different angle. Suppose you are at a casino and you win on the slots twice... these are the only games you ever played. You still should not expect to win again a third time, assuming the machine is working properly, etc. There are no moves to talk about, but it makes the point that past results are not necessarily the best predictors of future results... in this case, a probabilistic analysis of the odds for that machine would be the best predictor. 3) BUT,... many chess software programmers, I am sure, have tried to make a rating for the human based on moves, because it would be a very desirable feature for, say, the Chessmaster audience. However, I have not heard of a software that actually does this well (or even at all - but correct me if I am wrong). I suppose there are going to be practical obstacles for a while if not forever because chess is a complicated game. Given that we have a system that works pretty well for active players just by using results, this is going to be very hard to improve on. 4) I once read in a book by Suba, I think, an anecdote about this kind of project. Basically someone approached him and said that they had solved chess, and that they had devised a system that would rank all the moves and then predict the winner of the game according to which player had the higher score. The system worked perfectly, and Suba was impressed, until he discovered the `catch' - the moves checkmate and resigns were given fantastically high positive or negative scores respectively, reflecting the fact that they were more important than all of the other moves combined... (-: It was a long time ago and I don't have the book, so I can't cite that, but it is a pretty nice story I think. Tom
|
| |
Date: 10 Apr 2005 13:56:05
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], tom_h <[email protected] > mukotrpnim radom iznade ovu umotvorinu: > cut > David has given a few examples already that show that, but I wanted to > add another one from a sort of different angle. Suppose you are at a > casino and you win on the slots twice... these are the only games you > ever played. You still should not expect to win again a third time, > assuming the machine is working properly, etc. There are no moves to > talk about, but it makes the point that past results are not > necessarily the best predictors of future results... in this case, a > probabilistic analysis of the odds for that machine would be the best > predictor. > cut Sorry, I have to comment this example. It's bad because you can't influence the outcome of the slot game. There's no factor as your knowledge which you can win slot games with. It's just luck. And chess is not about luck.
|
|
Date: 07 Apr 2005 06:04:02
From:
Subject: Re: Rating from moves??
|
David Kane wrote: > No I've actually demonstrated clear cut examples where > move-based evaluation is superior to result-based rating > (which you conveniently avoid responding to) That demonstrates > the potential of the concept and *should* shift the argument > to practical matters - I'm coming into this thread late, so perhaps someone has has already made this point, but it seems to me that practicality is by far the biggest obstacle to a move-based rating system, whatever its merits otherwise. A system based on quality of moves would require extensive analysis for every move of every game submitted for rating. The impracticality of this, for even the best-funded of chess organizations, is obvious. > A move-based rating system, even if *inferior* to a result > rating system, would still be extremely useful for the 99+% > of the world's chess players who don't have result-based > ratings. Such players can use the many "solitaire chess" books, which provide exactly the sort of move-based rating you propose. This sort of thing has been around since the 1940s, if not before. Analogous computer programs may now also be available.
|
| |
Date: 07 Apr 2005 09:05:22
From: David Kane
Subject: Re: Rating from moves??
|
<[email protected] > wrote in message news:[email protected]... > > David Kane wrote: > > No I've actually demonstrated clear cut examples where > > move-based evaluation is superior to result-based rating > > (which you conveniently avoid responding to) That demonstrates > > the potential of the concept and *should* shift the argument > > to practical matters - > > I'm coming into this thread late, so perhaps someone has has already > made this point, but it seems to me that practicality is by far the > biggest obstacle to a move-based rating system, whatever its merits > otherwise. A system based on quality of moves would require extensive > analysis for every move of every game submitted for rating. The > impracticality of this, for even the best-funded of chess > organizations, is obvious. I agree with this comment, and the administrative simplicity of rating results is undeniable. But, if we had a rating calculator that was as or more accurate than result rating, I'd bet it would get a lot of use. Keep in mind that there are millions of chess players who don't even belong to organizations that do rating, yet they still play chess. This thread was started by someone asking for such a tool, and such requests pop up all the time.
|
|
Date: 04 Apr 2005 04:58:14
From: Sanny
Subject: Re: Rating from moves??
|
You play with Computer Online at http://www.softtanks.com/chessgame/Chess.html It gives you scores based on level you defeated. If you can win higher levels it means you have a high rating. Bye Sanjay.
|
| |
Date: 05 Apr 2005 01:53:59
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], Sanny <[email protected] > mukotrpnim radom iznade ovu umotvorinu: > You play with Computer Online at > > http://www.softtanks.com/chessgame/Chess.html > > It gives you scores based on level you defeated. If you can win higher > levels it means you have a high rating. > > Bye > Sanjay. No, no... I KNOW my rating is about 1800 but I want my opponents rating from one or more games and that's the tricky part. Thanks anyway.
|
| | |
Date: 05 Apr 2005 07:42:08
From: Terry
Subject: Re: Rating from moves??
|
"Bark!" <[email protected] > wrote in message news:[email protected]... > Na njuzima:[email protected], > Sanny <[email protected]> mukotrpnim radom iznade ovu umotvorinu: > >> You play with Computer Online at >> >> http://www.softtanks.com/chessgame/Chess.html >> >> It gives you scores based on level you defeated. If you can win higher >> levels it means you have a high rating. >> >> Bye >> Sanjay. > > No, no... I KNOW my rating is about 1800 but I want my opponents rating > from > one or more games and that's the tricky part. Thanks anyway. > > It cant be done. Full stop. Regards
|
| | | |
Date: 06 Apr 2005 01:44:27
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], Terry <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > "Bark!" <[email protected]> wrote in message > news:[email protected]... >> Na njuzima:[email protected], >> Sanny <[email protected]> mukotrpnim radom iznade ovu umotvorinu: >> >>> You play with Computer Online at >>> >>> http://www.softtanks.com/chessgame/Chess.html >>> >>> It gives you scores based on level you defeated. If you can win >>> higher levels it means you have a high rating. >>> >>> Bye >>> Sanjay. >> >> No, no... I KNOW my rating is about 1800 but I want my opponents >> rating from >> one or more games and that's the tricky part. Thanks anyway. >> >> > It cant be done. Full stop. > > Regards Fair enough :)
|
| |
Date: 04 Apr 2005 22:28:22
From: anthony mee
Subject: Re: Rating from moves??
|
Rating isn't a measure of the strength of a player but only of his results - ie it would be theoretically possible to play rotten chess and have a high rating through many lucky results. The opposite case is also possible - I'm sure we all believe ourselves in this position. A der S "Sanny" <[email protected] > wrote in message news:[email protected]... > You play with Computer Online at > > http://www.softtanks.com/chessgame/Chess.html > > It gives you scores based on level you defeated. If you can win higher > levels it means you have a high rating. > > Bye > Sanjay. >
|
| | |
Date: 04 Apr 2005 14:34:37
From: David Kane
Subject: Re: Rating from moves??
|
"anthony mee" <[email protected] > wrote in message news:[email protected]... > Rating isn't a measure of the strength of a player but only of his results - > ie it would be theoretically possible to play rotten chess and have a high > rating through many lucky results. The opposite case is also possible - I'm > sure we all believe ourselves in this position. Actually, I've yet to hear any plausible theory that would cause somebody to consistently have better results than someone who plays poorer moves. The closest I can think of something relating to personal hygiene or appearance - e.g. a person with revolting hygiene induces his opponent to play poorly, or an attractive female distracts her male opponents. My guess is that these aren't statistically significant, and that a move-based rating system could be superior to a result-based one. (Almost by definition a system that used both results and moves could be made superior to one relying on results alone)
|
| | | |
Date: 04 Apr 2005 22:38:18
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > Actually, I've yet to hear any plausible theory that would cause > somebody to consistently have better results than someone who > plays poorer moves. <snip > > and that a move-based rating system could be > superior to a result-based one. (Almost by definition a system > that used both results and moves could be made superior to one > relying on results alone) The problem is not so much some theory that has to be proved with facts, but the very definition of rating system. A rating system - Elo, USCF or whatever chess rating system - is not some sort of unproven theory, but a system based on calculation of results in a pool of players, not of "strenght". The basis of that is the fact that chess is a game played between people, not some abstract puzzle solving game. Rating predicts win/lose chances; it tells you something about strength: that is, strength relative to other players. And this is determined by who wins the game, not who makes the most beautiful moves. -- CeeBee "Eppur si muove!"
|
| | | | |
Date: 04 Apr 2005 16:41:31
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > > > Actually, I've yet to hear any plausible theory that would cause > > somebody to consistently have better results than someone who > > plays poorer moves. > > <snip> > > > and that a move-based rating system could be > > superior to a result-based one. (Almost by definition a system > > that used both results and moves could be made superior to one > > relying on results alone) > > The problem is not so much some theory that has to be proved with > facts, but the very definition of rating system. > > A rating system - Elo, USCF or whatever chess rating system - is > not some sort of unproven theory, but a system based on > calculation of results in a pool of players, not of "strenght". > > The basis of that is the fact that chess is a game played between > people, not some abstract puzzle solving game. > > Rating predicts win/lose chances; it tells you something about > strength: that is, strength relative to other players. And this is > determined by who wins the game, not who makes the most beautiful > moves. Ratings predict wins and losses. The point is that result-based ratings aren't the only way, nor the best way to make predictions. The strength of the ELO system is that they allow predictions to be made without knowing anything at all about the underlying game, as well as having certain administrative advantages. However, given that people do understand things about the underlying games, better predictors can be constructed. For example, if I tell you that A beat B and C beat B (in ice hockey), the W-L result would not allow you to make a prediction about A vs. C. If I am a knowledgeable hockey observer and know something else about the games, e.g. A beat B 47-0 and C beat B 6-5, then I would be able to come up with a superior prediction than you could. There is certainly no theoretical reason that I can think of why we couldn't improve our predictive beyond that of the weak ELO method.
|
| | | | | |
Date: 05 Apr 2005 00:26:13
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > There is certainly no theoretical reason that I can think of why > we couldn't improve our predictive beyond that of the weak ELO > method. I know a practical reason why it is quite useless trying to improve, and why the Elo rating method isn't weak at all: everyday's _practice_. The Elo rating predicts very well, right into in the top hunderd of the world, regardless of the "underlying moves" in all those games. Proof, pudding, eating. -- CeeBee "Eppur si muove!"
|
| | | | | | |
Date: 04 Apr 2005 22:51:06
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > > > There is certainly no theoretical reason that I can think of why > > we couldn't improve our predictive beyond that of the weak ELO > > method. > > > I know a practical reason why it is quite useless trying to improve, > and why the Elo rating method isn't weak at all: everyday's > _practice_. The Elo rating predicts very well, right into in the top > hunderd of the world, regardless of the "underlying moves" in all > those games. And what about all those millions of chess players who haven't played rated games (or enough of them) for the Elo method to have any predictive value? How could any person with any knowledge of chess deny that making good chess moves is predictive of good chess results? Aren't computers able to play extremely well doing nothing other than making good moves?
|
| | | | | | | |
Date: 05 Apr 2005 16:08:32
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > And what about all those millions of chess players who haven't > played rated games (or enough of them) for the Elo method > to have any predictive value? What is the use of giving a rating to people who never use it? What is the use of forcing some predictive value on someone who in ten or twenty rated game does have a rating? Should we skip a complete rating system that works okay because people have provisionary ratings for a few games? > How could any person with any knowledge of chess > deny that making good chess moves is predictive of good chess > results? I can. Because you wrongly assume that a game is a collection of isolated moves, which isn't the case. You also use a definition which is not very usable. Notions like "good chess moves" are useless as long as you can't know for sure that the move will always lead to the desired results. Even for computers this is only possible in a minority of setups, like in positions covered by computer tablebases. And that's only the calculating bit. The essence of a chess game is to win, not to make "good moves", whatever that might mean. A move can be a good move because it threatens and scares your opponent, leading to mistakes, making him lose the game. It might be a bad or sub par move, but it did the job. A chess game in practice is always won because of the mistakes of the other player. How are you going to rate isolated chess moves taking into account that the opponent makes mistakes? > Aren't computers able to play extremely well doing > nothing > other than making good moves? No, they don't make "good moves", they calculate deeper and better than 99% of the humans. If their calculation skills are surpassed by better _understanding_ of the game, they lose the game as well. Detach yourself from your mechanical view of chess. It might be so for computers, but never for men. No matter what you claim, you still have the wrong perception of rating - as an indication of absolute strength. Rating will never be something you can measure with an instrument without taking into account what your opponents will do. And in that case it will become a tool for relative strenght again. It's very well possible that there will be better tools to determine ratings, but valueing individual moves will never be the solution. -- CeeBee "Eppur si muove!"
|
| | | | | | | | |
Date: 05 Apr 2005 09:43:55
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > "David Kane" <[email protected]> wrote in > > How could any person with any knowledge of chess > > deny that making good chess moves is predictive of good chess > > results? > > I can. Because you wrongly assume that a game is a collection of > isolated moves, which isn't the case. You've yet to come up with a coherent explanation of this point. Every chess game I've lost has been because my moves weren't as good as my opponents'. You may believe that there are some mysterious mystical elements affecting the results of chess games, but I don't. Even accepting the existence of such factors, it is unclear how they would affect the games *without* affecting the moves. There is an elegance to the Elo theory - it's neat that using results only can come up with something meaningful. But that doesn't eliminate the very real shortcomings of such ratings - there is no mechanism forcing the world's chess players to play each other periodically. It's a theory with no data for the vast majority of players. Hence the need for a better rating system that makes use of all of the information about games, not just the result.
|
| | | | | | | | | |
Date: 05 Apr 2005 20:57:22
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > You've yet to come up with a coherent explanation of > this point. No, I haven't. What you are asking me to do is to prove why 2 is bigger than 1 and 4 bigger than 3. It doesn't need proof, it's a definition, an agreement. So is the objective of the game of chess. The winning player is not the one making the best moves, but the one winning the game. > Every chess game I've lost has been because > my moves weren't as good as my opponents'. You have yet to come up with a coherent explanation why this is both true for you and both generally valid. Any game where a blunder directly ended the game is potentially disproving your assumption. > You may believe that there are some mysterious mystical > elements affecting the results of chess games, but I don't. The rules affect the result of a chess game. The one who is getting mated or resigns, loses the game. You are trying to discuss the validity of the chess rules, not of the rating system. > It's a theory > with no data for the vast majority of players. Hence the > need for a better rating system that makes use of all > of the information about games, not just the result. Again it is you who has yet to come up with a coherent explanation of this point. Elo is no theory but a calculation method predicting results. The outcome is not correct because the theory is correct, but because the method leads to those results. That's why your second point is invalid as well: it works for _all_ players who are included in the rating system. Any club, any competition on any level shows you that. Not because the Elo system is the best rating system, but because it gives the results it has to give - like 2 and 2 always will give you 4. It's not a case of theory, but of definition. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | |
Date: 05 Apr 2005 18:10:40
From: Morphy's ghost
Subject: Re: Rating from moves??
|
In the year of our Lord Tue, 5 Apr 2005 09:43:55 -0700, "David Kane" <[email protected] > wrote: > >"CeeBee" <[email protected]> wrote in message >news:[email protected]... >> "David Kane" <[email protected]> wrote in > >> > How could any person with any knowledge of chess >> > deny that making good chess moves is predictive of good chess >> > results? >> >> I can. Because you wrongly assume that a game is a collection of >> isolated moves, which isn't the case. > >You've yet to come up with a coherent explanation of >this point. Every chess game I've lost has been because >my moves weren't as good as my opponents'. Oh really? I've won plenty of games where my opponent was making better moves than I was except for that one move that let me sac a piece for a mating attack or trap the queen or simplify into a won endgame. Fascism should rightly be called Connorism because Fascist methods and ideology are best exemplified in the posts of Stewart Connor. For example, the blanket condemnations of classes of people such as Correctional Officers and Republicans which Stewie makes harken back to the yellow Stars of David and pink triangles that the Third Reich used in similar condemnations. The attempt to control the thoughts of the populace by telling them whose opinions they should listen to and whose opinions they should ignore is the moral equivalent of jack-booted thugs wrecking the offices of opposition newspapers in the chaos of Krystalnacht. And in the calm assumption that Stewie makes that his opinions and directives are neither to questioned or doubted, but merely accepted and obeyed, do we not hear the echo of a shrill-voiced Austrian ennunciating the theory of the Fuhrer principle? Furthermore, apparently Stewart Connor has invented his own religion (loosely based upon the works of Thomas Jefferson) in order to support his own ideological beliefs, in much the same way as Nazi Germany came up with a new religion to supplant Christianity. Finally, it should be no surprise to anyone that Stewie uses the word "Jew" as an insult. Fascism should rightly be called Connorism." -- Morphy's ghost
|
| | | | | | | | | |
Date: 05 Apr 2005 17:08:24
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > There is an elegance to the Elo theory - it's neat that using > results only can come up with something meaningful. > But that doesn't eliminate the very real shortcomings of > such ratings - there is no mechanism forcing the world's > chess players to play each other periodically. It's a theory > with no data for the vast majority of players. Hence the > need for a better rating system that makes use of all > of the information about games, not just the result. But the thing is -- let's say I make an unclear sacrifice. Heck, let's go further than that, let's say I make an unsound sacrifice. The move is absolutely refutable. But I go on to win, anyway. My opponent doesn't find the refutation. Am I a strong player or a patzer? The "objective" value of the move is somewhat low. And yet Mikhail Tal made it to the world championship playing a lot of unsound sacrifices. The ONLY way you can evaluate moves like that and make sense of them is to look at the result achieved. You wrote: > Every chess game I've lost has been because > my moves weren't as good as my opponents'. But this is essentially a tautology, and has nothing to do with the absolute value of either player's moves. Again -- look at what you're conceding here: the winner's moves are, in aggregate, better than the loser's moves. Look at the result.
|
| | | | | | | | | | |
Date: 05 Apr 2005 19:22:39
From: David Richerby
Subject: Re: Rating from moves??
|
Ron <[email protected] > wrote: > "David Kane" <[email protected]> wrote: >> Every chess game I've lost has been because my moves weren't as good as >> my opponents'. > > But this is essentially a tautology Actually, it isn't. I could play fifty `perfect' moves, while my opponent makes dubious replies that aren't quite bad enough to lose. I then play one spectacularly stupid move and get mated in one. So, my opponent has played one perfect move and fifty poor ones; I've played fifty perfect moves and a complete clanger. However you score the moves, taking the average will give me a better score per move but I still lost the game. Dave. -- David Richerby Unholy Tool (TM): it's like a www.chiark.greenend.org.uk/~davidr/ screwdriver but it's also a crime against nature!
|
| | | | | | | | | | | |
Date: 06 Apr 2005 01:40:52
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:RNe*[email protected], David Richerby <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > Ron <[email protected]> wrote: >> "David Kane" <[email protected]> wrote: >>> Every chess game I've lost has been because my moves weren't as >>> good as my opponents'. >> >> But this is essentially a tautology > > Actually, it isn't. I could play fifty `perfect' moves, while my > opponent makes dubious replies that aren't quite bad enough to lose. > I then play one spectacularly stupid move and get mated in one. So, > my opponent has played one perfect move and fifty poor ones; I've > played fifty perfect moves and a complete clanger. However you score > the moves, taking the average will give me a better score per move > but I still lost the game. > > > Dave. I must say i disagree to some extent. Good player won't play 99% of the game perfect against weaker opponent and then make a blunder that mates him next move. Even if you do that, that's very rare. And besides, consistency of moves increases just as rating increases. And your example is good. You ARE better then this guy that mated you because your play will prove to be far more better in the long term. The result of this one game is meaningless. The moves are that speak of your strenght. Correct me if I'm wrong.
|
| | | | | | | | | | | | |
Date: 06 Apr 2005 12:20:13
From: David Richerby
Subject: Re: Rating from moves??
|
Bark! <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> I could play fifty `perfect' moves, while my opponent makes dubious >> replies that aren't quite bad enough to lose. I then play one >> spectacularly stupid move and get mated in one. So, my opponent has >> played one perfect move and fifty poor ones; I've played fifty perfect >> moves and a complete clanger. However you score the moves, taking the >> average will give me a better score per move but I still lost the game. > > [...] your example is good. You ARE better then this guy that mated you > because your play will prove to be far more better in the long term. The > result of this one game is meaningless. The moves are that speak of > your strenght. Er, good point. All I've established is that move-based ratings are different from result-based ratings, which we already knew. Thankfully, Ron has come to my rescue elsewhere in the thread by pointing out that the aim of chess is not to play good moves but to win. Playing good moves is a means to an end, not an end in itself. Dave. -- David Richerby Gigantic Hat (TM): it's like a hat www.chiark.greenend.org.uk/~davidr/ but it's huge!
|
| | | | | | | | | | | | | |
Date: 07 Apr 2005 00:46:18
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:zBq*[email protected], David Richerby <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > Bark! <[email protected]> wrote: >> David Richerby <[email protected]> wrote: >>> I could play fifty `perfect' moves, while my opponent makes dubious >>> replies that aren't quite bad enough to lose. I then play one >>> spectacularly stupid move and get mated in one. So, my opponent has >>> played one perfect move and fifty poor ones; I've played fifty >>> perfect moves and a complete clanger. However you score the moves, >>> taking the average will give me a better score per move but I still >>> lost the game. >> >> [...] your example is good. You ARE better then this guy that mated >> you because your play will prove to be far more better in the long >> term. The result of this one game is meaningless. The moves are >> that speak of your strenght. > > Er, good point. All I've established is that move-based ratings are > different from result-based ratings, which we already knew. > > Thankfully, Ron has come to my rescue elsewhere in the thread by > pointing out that the aim of chess is not to play good moves but to > win. Playing good moves is a means to an end, not an end in itself. > > > Dave. Funny. Read my latest posts. We are all right. :) Yet, some of us almost argue. I never managed to provoke such a discussion on usenet :)
|
| | | | | | | | | | | |
Date: 05 Apr 2005 12:42:07
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:RNe*[email protected]... > Ron <[email protected]> wrote: > > "David Kane" <[email protected]> wrote: > >> Every chess game I've lost has been because my moves weren't as good as > >> my opponents'. > > > > But this is essentially a tautology > > Actually, it isn't. I could play fifty `perfect' moves, while my opponent > makes dubious replies that aren't quite bad enough to lose. I then play > one spectacularly stupid move and get mated in one. So, my opponent has > played one perfect move and fifty poor ones; I've played fifty perfect > moves and a complete clanger. However you score the moves, taking the > average will give me a better score per move but I still lost the game. > > You are saying this without any basis whatsoever. Allowing a mate in one could be evidence of your skill that outweighs the other moves. The point is that you could look at the moves of the guy who won, and still determine that he wasn't a very good player. The "result" has much, much less information about the player's skill than his moves, and hence much less predictive value. That is a limitation to Elo ratings that no amount of refinement can overcome.
|
| | | | | | | | | | | | |
Date: 05 Apr 2005 22:24:50
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> Ron <[email protected]> wrote: >>> David Kane <[email protected]> wrote: >>>> Every chess game I've lost has been because my moves weren't as >>>> good as my opponents'. >>> >>> But this is essentially a tautology >> >> Actually, it isn't. I could play fifty `perfect' moves, while my >> opponent makes dubious replies that aren't quite bad enough to lose. I >> then play one spectacularly stupid move and get mated in one. So, my >> opponent has played one perfect move and fifty poor ones; I've played >> fifty perfect moves and a complete clanger. However you score the >> moves, taking the average will give me a better score per move but I >> still lost the game. > > You are saying this without any basis whatsoever. It doesn't have any formal basis, no, since you haven't actually said how one might work. I'm assuming that you get some number of points for each move (up to some maximum corresponding to the theoretically perfect move for the position). Based on this assumption, I've demonstrated that you can lose a game of chess while still making moves that were, on average, better than your opponent's. This shows that your rek ``Every chess game I've lost has been because my moves weren't as good as my opponents' '' is not, as Ron claimed, a tautology. > Allowing a mate in one could be evidence of your skill that outweighs > the other moves. So moves that are more critical to the game should carry more weight? Fair enough. How do you propose to detect these critical moves? > The point is that you could look at the moves of the guy who won, and > still determine that he wasn't a very good player. The "result" has > much, much less information about the player's skill than his moves, > and hence much less predictive value. That `hence' only follows if you can extract the information from the moves. It's not at all clear that you can. Dave. -- David Richerby Nuclear Robot (TM): it's like a www.chiark.greenend.org.uk/~davidr/ high-tech robot that's made of atoms!
|
| | | | | | | | | | | | | |
Date: 05 Apr 2005 23:28:35
From: Ron
Subject: Re: Rating from moves??
|
In article <EhC*[email protected] >, David Richerby <[email protected] > wrote: > Based on this assumption, I've demonstrated that you can lose a game of > chess while still making moves that were, on average, better than your > opponent's. This shows that your rek ``Every chess game I've lost has > been because my moves weren't as good as my opponents' '' is not, as Ron > claimed, a tautology. Well, this is getting far afield, but a move which loses you the game is a bigger error than a move which gives your opponent a positional advantage. If you make a lot of slightly better moves than your opponent, and one move that's hideously worse than anything he's done, where does that leave you? Is your average move better or worse than his? It probably depends on what you mean by "average." Your median move score is better than his. Your mode move score is better than his. But you mean move score is almost certainly not better than his. (You can make 40 moves which are each .1 better than your opponent's, and your mean score will be worse if you follow them up with one move that hangs a rook for nothing.) -Ron
|
| | | | | | | | | | | | | | |
Date: 06 Apr 2005 12:05:32
From: David Richerby
Subject: Re: Rating from moves??
|
Ron <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> Based on this assumption, I've demonstrated that you can lose a game of >> chess while still making moves that were, on average, better than your >> opponent's. This shows that your rek ``Every chess game I've lost has >> been because my moves weren't as good as my opponents' '' is not, as Ron >> claimed, a tautology. > > Well, this is getting far afield, but a move which loses you the game is > a bigger error than a move which gives your opponent a positional > advantage. If you make a lot of slightly better moves than your > opponent, and one move that's hideously worse than anything he's done, > where does that leave you? Is your average move better or worse than > his? Well, if we're giving each moves ks out of some maximum, for all sufficiently large n, the average of n perfect moves followed by one shocker is higher than the average of n imperfect moves followed by one perfect move. > It probably depends on what you mean by "average." The modal interpretation for `average' is `arithmetic mean'. :-) Dave. -- David Richerby Mouldy.com (TM): it's like an www.chiark.greenend.org.uk/~davidr/ E-commerce portal but it's starting to grow mushrooms!
|
| | | | | | | | | | |
Date: 05 Apr 2005 10:35:31
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-54142C.10082205042005@newssvr13-ext.news.prodigy.com ... > In article <[email protected]>, > "David Kane" <[email protected]> wrote: > > > There is an elegance to the Elo theory - it's neat that using > > results only can come up with something meaningful. > > But that doesn't eliminate the very real shortcomings of > > such ratings - there is no mechanism forcing the world's > > chess players to play each other periodically. It's a theory > > with no data for the vast majority of players. Hence the > > need for a better rating system that makes use of all > > of the information about games, not just the result. > > But the thing is -- let's say I make an unclear sacrifice. Heck, let's > go further than that, let's say I make an unsound sacrifice. The move is > absolutely refutable. > > But I go on to win, anyway. My opponent doesn't find the refutation. > > Am I a strong player or a patzer? The "objective" value of the move is > somewhat low. And yet Mikhail Tal made it to the world championship > playing a lot of unsound sacrifices. > > The ONLY way you can evaluate moves like that and make sense of them is > to look at the result achieved. > Says who? Your sacrifice might have been the type of move that 1800 players play, but which 2400s would easily refute. You got a good position. But you missed a tactical combination at move 24 which would have won a pawn that most 2200s would have seen. You played the resulting rook ending much better than a 1400 player ever would, but made a few inaccurate moves, giving your opponent chances to draw. Every move is evidence of your skill. If you have the moves, the result is almost an afterthought. I concede that we have no tool to combine this information in a quantitatively precise way. I wouldn't even argue against the position that it would be practically difficult to create such a tool, even though my hunch is that it would be fairly straightforward. But the position that moves are completely meaningless is just plain ridiculous.
|
| | | | | | | | | | | |
Date: 05 Apr 2005 21:00:28
From: mrlount
Subject: Re: Rating from moves??
|
David Kane wrote: > > "Ron" <[email protected]> wrote in message > news:ronaldinho_m-54142C.10082205042005@newssvr13-ext.news.prodigy.com > ... >> In article <[email protected]>, >> "David Kane" <[email protected]> wrote: >> >> > There is an elegance to the Elo theory - it's neat that using >> > results only can come up with something meaningful. >> > But that doesn't eliminate the very real shortcomings of >> > such ratings - there is no mechanism forcing the world's >> > chess players to play each other periodically. It's a theory >> > with no data for the vast majority of players. Hence the >> > need for a better rating system that makes use of all >> > of the information about games, not just the result. >> >> But the thing is -- let's say I make an unclear sacrifice. Heck, > let's >> go further than that, let's say I make an unsound sacrifice. The > move is >> absolutely refutable. >> >> But I go on to win, anyway. My opponent doesn't find the refutation. >> >> Am I a strong player or a patzer? The "objective" value of the move > is >> somewhat low. And yet Mikhail Tal made it to the world championship >> playing a lot of unsound sacrifices. >> >> The ONLY way you can evaluate moves like that and make sense of them > is >> to look at the result achieved. >> > > Says who? Your sacrifice might have been the type of move > that 1800 players play, but which 2400s would easily refute. You > got a good position. But you missed a tactical combination at move > 24 which would have won a pawn that most 2200s > would have seen. You played the resulting rook ending much > better than a 1400 player ever would, but made a few > inaccurate moves, giving your opponent chances > to draw. Every move is evidence of your skill. If you have the > moves, the result is almost an afterthought. > > I concede that we have no tool to combine this information > in a quantitatively precise way. I wouldn't even argue against > the position that it would be practically difficult to create such > a tool, even though my hunch is that it would be fairly > straightforward. But the position that moves are completely > meaningless is just plain ridiculous. I think this all misses the point. The idea of the Elo is to give a statistical basis for predicting a result. This means that it must go on results, not the moves made. However good your moves are ( or aren't ) is irrelevant. The person who wins by hook or by crook should have the higher Elo, it's that simple, if you're more likely to win, then you're more likely to win. This is not chess mythology, this is sound mathematics that everyone learns in grade school.
|
| | | | | | | | | | | | |
Date: 05 Apr 2005 13:50:40
From: David Kane
Subject: Re: Rating from moves??
|
"mrlount" <[email protected] > wrote in message news:[email protected]... > David Kane wrote: > > > > I concede that we have no tool to combine this information > > in a quantitatively precise way. I wouldn't even argue against > > the position that it would be practically difficult to create such > > a tool, even though my hunch is that it would be fairly > > straightforward. But the position that moves are completely > > meaningless is just plain ridiculous. > > I think this all misses the point. The idea of the Elo is to give a > statistical basis for predicting a result. > > This means that it must go on results, not the moves made. However good your > moves are ( or aren't ) is irrelevant. The person who wins by hook or by > crook should have the higher Elo, it's that simple, if you're more likely > to win, then you're more likely to win. I think you are missing the point. Bad players make bad moves whether they win or lose. Better players make better moves whether they win or lose. I'm sure that my moves in my wins are objectively worse than a GMs moves in his losses. Elo ratings don't make use of that, but that is not to say that another rating method couldn't. The only way that move rating wouldn't work is if there is something *other* than the quality of my moves that determines the results of my games (on average). No one has suggested any such possibilities, though I'll concede the theoretical possibility. In any case, we need to develop the move-rating tool in order to figure out what those factors are.
|
| | | | | | | | | | | | | |
Date: 05 Apr 2005 23:25:28
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > The only way that move rating wouldn't work is if there is > something *other* than the quality of my moves that determines > the results of my games (on average). No one has suggested > any such possibilities, though I'll concede the theoretical > possibility. Actually, we have suggested a possibility, repeatedly: The quality of your opponents moves determines the outcome of your games as much as the quality of your moves does. Furthermore, I think it's impossible to judge a move on it's own. Moves exist as part of a plan. If I make the rights moves, but don't understand the correct plan, then those moves are not doing me very much good. Sooner or later I'll follow up incorrectly because I don't understand the correct plan. Now you could argue, well, those later moves are errors. And sure, you're right. But if you were somehow trying to evaluate moves objectively, you'd be giving me too much credit for the "good" moves I made leading up to those errors. Moves don't exist in a void. Nor is there an ideal move in any given situation. A player may choose move X over move Y not because move X is objectively better, but because he feels that move gives him better practical chances in the ensuing position. Heck -- that's why I play the Smith-Morra. I know it's not as good, objectively speaking, as the open or c3 sicilians. On the other hand, my practical results are a hell of a lot better with it. Somebody looking at my games trying to score them based on the moves would have to take points away for 2.d4 and 3.c3, and yet, in fact, those moves lead to wins for me. I also don't think you answered the problem of the unsound sacrifice. I make a move which results in me winning, but objectively speaking, it should result in me losing. Now, if I play those moves and win against 1500s, you can say I'm doing something wrong. Stronger players will punish me. But you can't say that about Tal playing against Botvinnik. Nevertheless, any objective evaluation of his moves would have to "take points away" -- they're BAD moves, absolutely refutable. That his opponents failed to refute them doesn't make them good moves. You may dock him points for his failure to spot the correct refutation -- but still, do I get points for making an unsound sacrifice which wins, do I lose them, or does it depend on the strength of my opponent? The object of a game of chess is to win, not neccesarily to make the best moves. This may seems counterintuitive but I think you'll see plenty of cases where strong players made intentionally inferior moves (Lasker is probably the pre-eminent example) because of their confidence in their ability to win from the resulting position. If the goal was only to make the best moves, such a strategy would be silly. -Ron
|
| | | | | | | | | | | | | | |
Date: 06 Apr 2005 11:56:47
From: David Richerby
Subject: Re: Rating from moves??
|
Ron <[email protected] > wrote: > The object of a game of chess is to win, not neccesarily to make the > best moves. *slaps forehead* Excellent point. I think that just about puts the nail in the coffin of the whole scheme. Dave. -- David Richerby Fluorescent Miniature Atom Bomb (TM): www.chiark.greenend.org.uk/~davidr/ it's like a weapon of mass destruction but you can hold in it your hand and it'll hurt your eyes!
|
| | | | | | | | | | | | | | | |
Date: 06 Apr 2005 09:53:37
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:Fli*[email protected]... > Ron <[email protected]> wrote: > > The object of a game of chess is to win, not neccesarily to make the > > best moves. > > *slaps forehead* Excellent point. I think that just about puts the nail > in the coffin of the whole scheme. > The objective in ice hockey is to win. Does knowing whether the final score was 47-0 or 6-5 have zero predictive value as to who will win a rematch? The objective in running is to win. Does knowing that the winner of athon A finished with a time of 3:49:51 while the winner of athon B finished with a time of 2:12:39 have no predictive value as to who would win if they raced each other? If Anand can clinch the 1st place prize by drawing his last round, and his opponent (rated 200 points below him) can also gain rating points and prize money with a draw, isn't it plausible that their quick draw indicates that something other than winning was their objective? No reasonable person would argue that a single pre-arranged draw invalidates the whole concept of Elo ratings, so why does some bizarre hypothetical scenario invalidate move rating?
|
| | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 17:02:07
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> Ron <[email protected]> wrote: >>> The object of a game of chess is to win, not neccesarily to make >>> the best moves. >> >> *slaps forehead* Excellent point. I think that just about puts the >> nail in the coffin of the whole scheme. > > The objective in ice hockey is to win. Does knowing whether the > final score was 47-0 or 6-5 have zero predictive value as to who > will win a rematch? I think that's different. I'd argue that the final score is as much a part of the result as the word `win' and obviously that has a lot of predictive value. Perhaps one could say that `won in 25 moves' is the result of a chess game -- that also carries more information. But to extend the idea of move-based ratings into ice hockey, it seems that a better analogy would be trying to work out which was the better team by analysing snapshots of the game taken every few seconds. That's not a perfect analogy but I think it's closer than just looking at the scoreline. The same applies to your athon example. Now, you may want to argue that the scoreline isn't part of the result, which is fair enough. However, it is dramatically easier to extract information from the scoreline than it is to extract it from the moves -- indeed, nobody has yet suggested a convincing way of doing that. > If Anand can clinch the 1st place prize by drawing his last round, and > his opponent (rated 200 points below him) can also gain rating points > and prize money with a draw, isn't it plausible that their quick draw > indicates that something other than winning was their objective? Good point. The laws of chess, though, are pretty clear that the aim is to win rather than to make pretty moves: Article 1.1.1 of the FIDE rules states that ``The objective of each player is to place the opponent's king `under attack' such that the opponent has no legal move which would avoid the `capture' of the king on the following move. The player who achieves this goal is said to have `checkmated' the opponent's king and to have won the game. The opponent whose king has been checkmated has lost the game.'' Of course, in some tournament or match situations, a draw is good enough. Good moves, though, are only good moves insofar as they bring the position closer to being won. Dave. -- David Richerby Mouldy Love Clock (TM): it's like a www.chiark.greenend.org.uk/~davidr/ clock that you can share with someone special but it's starting to grow mushrooms!
|
| | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 18:33:22
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > Does knowing whether the > final score was 47-0 or 6-5 have zero predictive value as to who > will win a rematch? I don't think anybody is claiming that the moves of a chess game have zero predictive value. You're setting up a straw man. You're alleging that it's somehow easier/better than result-based ratings, and these examples do nothing to further that point.
|
| | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 13:12:19
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-E18C1F.11332006042005@newssvr14-ext.news.prodigy.com ... > In article <[email protected]>, > "David Kane" <[email protected]> wrote: > > > Does knowing whether the > > final score was 47-0 or 6-5 have zero predictive value as to who > > will win a rematch? > > I don't think anybody is claiming that the moves of a chess game have > zero predictive value. You're setting up a straw man. No. You are arguing that this information fundamentally can't be used. I personally do think it is a ridiculous position, but I've done my best to try to see where you are coming from.
|
| | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 16:27:33
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > Ron <[email protected]> wrote: >> David Kane <[email protected]> wrote: >>> Does knowing whether the final score was 47-0 or 6-5 have zero >>> predictive value as to who will win a rematch? >> >> I don't think anybody is claiming that the moves of a chess game >> have zero predictive value. You're setting up a straw man. > > No. You are arguing that this information fundamentally can't be > used. I don't think I've said that. I've certainly said that the schemes suggested so far don't work, though. Dave. -- David Richerby Indelible Simple Whisky (TM): it's www.chiark.greenend.org.uk/~davidr/ like a single-malt whisky but it has no moving parts and it can't be erased!
|
| | | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 23:53:49
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > You are arguing that this information fundamentally can't be > used. I never said that it "fundamentally can't be used." I've even condeded that some of your measures have some predictive value -- which is the exact opposite of saying that they "fundamentally can't be used." -Ron
|
| | | | | | | | | | | | | | |
Date: 05 Apr 2005 16:58:49
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-1E25BD.16253005042005@newssvr13-ext.news.prodigy.com ... > In article <[email protected]>, > "David Kane" <[email protected]> wrote: > > > > The only way that move rating wouldn't work is if there is > > something *other* than the quality of my moves that determines > > the results of my games (on average). No one has suggested > > any such possibilities, though I'll concede the theoretical > > possibility. > > Actually, we have suggested a possibility, repeatedly: > > The quality of your opponents moves determines the outcome of your > games as much as the quality of your moves does. > For move-based ratings to be invalid there would have to be something about you, not reflected in the moves you play, which makes your opponents play worse against you than against someone else who made the same moves you do. So far, I am the only one to propose any such mechanism (the hygiene idea), and don't believe it is significant. Random fluctuations in performance exist whether you are talking about results or moves. Throwing away information contained in the moves can only hurt, not help, in making predictions.
|
| | | | | | | | | | | | | | | |
Date: 06 Apr 2005 00:51:28
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > For move-based ratings to be invalid there would > have to be something about you, not reflected in the moves > you play, which makes your opponents play worse against > you than against someone else who made the same moves > you do. So far, I am the only one to propose any such > mechanism (the hygiene idea), and don't believe it is > significant. I think there's very clearly a psychological element to a game of chess. Read Lasker's work, or Tal's account of his match with Botvinnik. These writings are chock-full of discussions of the "why" of particular moves -- including specific times when (for example) Tal concedes he chose a move that probably not best for purely psychological reasons. It's undeniable. Kasparov has been known to bluff (are you going to accept a sacrifice you can't see the end of against him?). It's part of the game. Heck, I myself have made moves entirely because I suspected that my opponent would be seduced by a check or a threat and therefore wouldn't notice that there was, in fact, a stronger continuation. I can think of two examples from my own play (both involving letting my opponent give check) where I saw stronger replies for my opponent but was absolutely certain they'd ignore them. How did I know this? In one of the instances (involving allowing a knight fork at c2, denying me castling rights) my opponent clearly had a better option. And yet... somehow, I knew. Anyway, you may write of my experience as irrelevant (I'm a pretty weak player) but you can't write off the comments of Tal and other top players. -Ron
|
| | | | | | | | | | | | | | | | |
Date: 05 Apr 2005 23:32:03
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-D5ECBD.17512805042005@newssvr14-ext.news.prodigy.com ... > Anyway, you may write of my experience as irrelevant (I'm a pretty weak > player) but you can't write off the comments of Tal and other top > players. > As long as the so-called psychological factors manifest themselves in the moves that are played, you support the possibility of move-based rating. Similarly, your "unsound sacrifice" argument is directed against a position I've never taken. You seem to think that I have argued that only bad players would play something that is theoretically unsound. Not so. All that I'm saying is that the moves you make (in aggregate) indicate the strength of your play.
|
| | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 08:17:40
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > "Ron" <[email protected]> wrote in message > news:ronaldinho_m-D5ECBD.17512805042005@newssvr14-ext.news.prodigy.com > ... > > > Anyway, you may write of my experience as irrelevant (I'm a pretty > weak > > player) but you can't write off the comments of Tal and other top > > players. > > > As long as the so-called psychological factors > manifest themselves in the moves that are played, > you support the possibility of move-based rating. Except that they would tend to indicate that a player was weaker than he actually was. This would be a source of error in move-based rating. > Similarly, your "unsound sacrifice" argument is directed > against a position I've never taken. You seem to think that > I have argued that only bad players would play something > that is theoretically unsound. Not so. All that I'm saying > is that the moves you make (in aggregate) indicate > the strength of your play. Well, the only methodology anyone has suggested has been comparing moves to a computer's moves -- but a computer will rate an unsound sacrifice as a blunder, and therefore rate a player who makes them too weakly. This is a systemic error which you haven't suggested a way to work around. You can't just claim that they somehow disappear "in aggregate" because it's an error that only works in one direction. -Ron
|
| | | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 03:55:39
From: David Kane
Subject: Re: Rating from moves??
|
> > Well, the only methodology anyone has suggested has been comparing > moves to a computer's moves -- but a computer will rate an unsound > sacrifice as a blunder, and therefore rate a player who makes them too > weakly. This is a systemic error which you haven't suggested a way to > work around. You can't just claim that they somehow disappear "in > aggregate" because it's an error that only works in one direction. > If it turned out that theoretically poorer moves correlate with higher ratings, then you would just program it that way. This is not a significant objection.
|
| | | | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 18:30:41
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > > Well, the only methodology anyone has suggested has been comparing > > moves to a computer's moves -- but a computer will rate an unsound > > sacrifice as a blunder, and therefore rate a player who makes them > too > > weakly. This is a systemic error which you haven't suggested a way > to > > work around. You can't just claim that they somehow disappear "in > > aggregate" because it's an error that only works in one direction. > > > > If it turned out that theoretically poorer moves > correlate with higher ratings, then you would > just program it that way. This is not a significant > objection. Except they don't correlate. Because some players play speculative sacrifices, and some don't. And you say, "just program it that way," but you say that like it's easy to do. My hole point is that it's not easy to do -- there are fundamental problems with the idea that you haven't addressed at a theoretical level, and you can't address them by just saying, "oh, just program it that way." -Ron
|
|
Date: 04 Apr 2005 05:14:59
From: Morphy's ghost
Subject: Re: Rating from moves??
|
In the year of our Lord Mon, 4 Apr 2005 01:13:10 +0200, "Bark!" <[email protected] > wrote: >Is it posible to somehow get unrated player's ELO from the moves he plays? >To be more concrete, I'd like to get my opponent's rating based on game(s) >that we played against each other so I can review my play against specific >sort of opponents (~1200, ~1600.. etc.) > >Thanx in advance > > If you have a fairly large sample of games with the opponent and if you know your own ELO, it would be possible to get a rough estimate from seeing what percentage of wins you score against him. Bray a fool in a morter with wheat. yet shall not his folly be beaten out of him. -- William Blake
|
|
Date: 04 Apr 2005 02:36:08
From: =?ISO-8859-1?Q?Claus-J=FCrgen_Heigl?=
Subject: Re: Rating from moves??
|
Bark! wrote: > Is it posible to somehow get unrated player's ELO from the moves he plays? > To be more concrete, I'd like to get my opponent's rating based on game(s) > that we played against each other so I can review my play against specific > sort of opponents (~1200, ~1600.. etc.) An experienced player can give you the general direction what your ratings would be. You can also calculate your performance rating given the results of your games. See here how this is done: http://chess.about.com/od/organizations/ss/aa04l04_7.htm Finally chess software like Fritz can do the calculations for you. Claus-Juergen
|
| |
Date: 05 Apr 2005 02:04:18
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], Claus-J�rgen Heigl <[email protected] > mukotrpnim radom iznade ovu umotvorinu: > Bark! wrote: >> Is it posible to somehow get unrated player's ELO from the moves he >> plays? To be more concrete, I'd like to get my opponent's rating >> based on game(s) that we played against each other so I can review >> my play against specific sort of opponents (~1200, ~1600.. etc.) > > An experienced player can give you the general direction what your > ratings would be. You can also calculate your performance rating given > the results of your games. See here how this is done: > http://chess.about.com/od/organizations/ss/aa04l04_7.htm > Finally chess software like Fritz can do the calculations for you. > > Claus-Juergen I know my rating (~1800) and I have Fritz. The question is how to estimate someone's rating based on moves played (my opponents). I know nothing about my opponent i.e. we played at Yahoo Chess, he's unrated, no games played... You get the picture. I beat him but still don't know his approximate rating. I think Peter Billam's reply is the closest answer I was looking for.
|
| | |
Date: 05 Apr 2005 04:08:17
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "Bark!" <[email protected]> wrote: > I know my rating (~1800) and I have Fritz. The question is how to estimate > someone's rating based on moves played (my opponents). I know nothing about > my opponent i.e. we played at Yahoo Chess, he's unrated, no games played... > You get the picture. I beat him but still don't know his approximate rating. > I think Peter Billam's reply is the closest answer I was looking for. You underestimate the practical difficulties of doing this. The problem with weaker players isn't so much that every move they make is just a tiny bit worse than every move a stronger player makes. A big part of why people get stuck is inconsistency -- they may outplay an expert in one game, and hang a piece to a two-move combination in the next -- against a 1400. The could play two dozen moves very well, and then hang a piece on move 25. I could point to a blunder that Anand made that's as bad as any I've made. Not subtle stuff -- hanging a piece in the first few moves. The reason why Anand's Anand and I'm me is that I make errors like that fairly often, and he makes them, well, next to never. But you look at the right game of mine and you'll see me slicing apart an opponent with a strong sacrificial attack, or sharply executing a rook ending. You look at the wrong game of his, and putting the two together, you think I'm the GM and he's the patzer. -Ron
|
| | | |
Date: 05 Apr 2005 08:31:58
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-1DFBDA.21081604042005@newssvr21-ext.news.prodigy.com ... > > I could point to a blunder that Anand made that's as bad as any I've > made. Not subtle stuff -- hanging a piece in the first few moves. > > The reason why Anand's Anand and I'm me is that I make errors like that > fairly often, and he makes them, well, next to never. But you look at > the right game of mine and you'll see me slicing apart an opponent with > a strong sacrificial attack, or sharply executing a rook ending. You > look at the wrong game of his, and putting the two together, you think > I'm the GM and he's the patzer. > If you calculated a result-based rating by selecting only Anand's losses and only your wins, you would also reach an incorrect conclusion about your playing strengths. The flaw is in the selection of data, not some fundamental shortcoming in looking at the moves. I have no doubt that looking at Anand's moves in his last few games and my moves in my last few games would easily resolve who the better player is. A move-based rating system does not have to be perfect to be useful. The target is imperfect Elo ratings, not perfection.
|
| | | | |
Date: 05 Apr 2005 19:25:50
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > A move-based rating system does not have to be perfect to be useful. > The target is imperfect Elo ratings, not perfection. However, a move-based rating system does have to *exist* in order to be useful. Nobody has given a plausible explanation of how such a system would work. Dave. -- David Richerby Poetic Peanut (TM): it's like a www.chiark.greenend.org.uk/~davidr/ roasted nut but it's in verse!
|
| | | | | |
Date: 05 Apr 2005 13:33:37
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:sHf*[email protected]... > David Kane <[email protected]> wrote: > > A move-based rating system does not have to be perfect to be useful. > > The target is imperfect Elo ratings, not perfection. > > However, a move-based rating system does have to *exist* in order to be > useful. Nobody has given a plausible explanation of how such a system > would work. > That's because I've been busy making arguments that shouldn't need to be made, namely that, in principle, the moves of chess games contain information about chessplaying skill that have predictive value. Anyone who has ever stopped to watch a chess game taking place in a public place will develop an opinion about the abilities of the players by watching the moves they make, especially if he is stronger than the players. No one would suggest that the only way to learn anything is for the players to play 20 games in a rating pool. Since computers can play much better than our hypothetical observer, as well as store information about how humans of all abilities play, the concept is quite simple. A computer would characterize the quality of the moves and then match that profile to those of other players (recognizing that time control is a factor of course). Now fire away, but remember that your argument must show why "results-only" Elo would make better predictions with the same amount of data.
|
| | | | | | |
Date: 05 Apr 2005 22:35:41
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> David Kane <[email protected]> wrote: >>> A move-based rating system does not have to be perfect to be >>> useful. The target is imperfect Elo ratings, not perfection. >> >> However, a move-based rating system does have to *exist* in order to >> be useful. Nobody has given a plausible explanation of how such a >> system would work. > > That's because I've been busy making arguments that shouldn't need to > be made, namely that, in principle, the moves of chess games contain > information about chessplaying skill that have predictive value. Fair enough. > Anyone who has ever stopped to watch a chess game taking place in a > public place will develop an opinion about the abilities of the players > by watching the moves they make, especially if he is stronger than the > players. [...] Fair enough. > A computer would characterize the quality of the moves and then match > that profile to those of other players (recognizing that time control > is a factor of course). I think there's an awful lot of complexity hidden by the words `would characterize'. Do you have any ideas about how such a characterization would be performed? > Now fire away, but remember that your argument must show why > "results-only" Elo would make better predictions with the same amount of > data. Well, as I said, Elo's rating system exists and makes predictions. The move-based system does not currently exist and currently makes no predictions. Dave. -- David Richerby Addictive Psychotic Radio (TM): it's www.chiark.greenend.org.uk/~davidr/ like a radio but it wants to kill you and you can never put it down!
|
| | | | | | | |
Date: 05 Apr 2005 17:01:52
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:1DF*[email protected]... > David Kane <[email protected]> wrote: > > > A computer would characterize the quality of the moves and then match > > that profile to those of other players (recognizing that time control > > is a factor of course). > > I think there's an awful lot of complexity hidden by the words `would > characterize'. Do you have any ideas about how such a characterization > would be performed? > Two ideas seem promising. A simple way that would almost certainly work well for club players, would be to calculate something like blunder rate vs. ply depth and fit that to empirical data. A more general algorithm would be to generate something like the probability that players of a certain strength would make a certain move in every position. (The details are left as an exercize for the reader) Then a best fit rating over all of the moves of all of the games would be calculated. > > Now fire away, but remember that your argument must show why > > "results-only" Elo would make better predictions with the same amount of > > data. > > Well, as I said, Elo's rating system exists and makes predictions. The > move-based system does not currently exist and currently makes no > predictions. > Fair enough. But consider running. At one time in history there was no way to measure time. If you wanted to find out who was the faster runner, the only way was to have people line up at the start and see who reaches the finish line first. The only way to measure yourself as a runner was how you finished in the races you ran. Find a recreational runner who has run a athon. Chances are pretty good that he can tell you the time it took him to finish to the nearest second. Ask him what place he finished and he probably has no idea. The point is that runners today rate themselves based upon their times. When Elo devised his result-based rating method, there was nothing but head-to-head results upon which to predict future results. Now we are in a different era. With the possible exception of high-level correspondence games and a few super-GMs at slow time controls, we have machines that can determine the quality of our games. I think eventually ordinary chess players will view their "move-rating" as their real rating, and understand their "result-rating" to be a less accurate approximation of it.
|
| | | | | | | | |
Date: 06 Apr 2005 11:47:08
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> David Kane <[email protected]> wrote: >>> A computer would characterize the quality of the moves and then >>> match that profile to those of other players >> >> I think there's an awful lot of complexity hidden by the words >> `would characterize'. Do you have any ideas about how such a >> characterization would be performed? > > Two ideas seem promising. A simple way that would almost certainly work > well for club players, would be to calculate something like blunder rate > vs. ply depth and fit that to empirical data. I'm not at all sure what you mean by `blunder rate vs ply depth'. Could you explain? Even without understanding what your statistic means, there's a huge problem with the method. What exactly is this `empirical data' you're going to fit to the statistic? It can't be any sort of existing rating because we're assuming a priori that move-based ratings are not directly correlated with result-based ratings -- if they were, it would be much easier to measure the result-based rating directly than to approximate it with a move-based rating. And I can't think of any empirical data on chess games that isn't a rating. > A more general algorithm would be to generate something like the > probability that players of a certain strength would make a certain move > in every position. That doesn't measure how good a player is: it measures how similar their play is to the reference engine. Why should my rating be a measure of how often I agree with FIDE SuperChess 5.3.8? What happens when somebody writes a better chess program? The feels incredibly arbitrary to me. > (The details are left as an exercize for the reader) *cough* I realise you're being light hearted but you don't have a rating scheme until these details are worked out or, at the very least, sketched. It's all very well omitting the details when they're tedious or obvious but, here, they're the very substance of the discussion. > When Elo devised his result-based rating method, there was nothing but > head-to-head results upon which to predict future results. Now we are in > a different era. Careful. We're not in the different era until we actually have this move- based rating system. Remember that ``computers are very powerful'' does not mean ``computers can do everything.'' > I think eventually ordinary chess players will view their "move- > rating" as their real rating, and understand their "result-rating" to be > a less accurate approximation of it. I've still not seen anything that convinces me that move-based ratings are well-defined or feasibly computable. It's premature to declare that they're the future of ratings. Dave. -- David Richerby Solar-Powered Pointy-Haired Tool www.chiark.greenend.org.uk/~davidr/ (TM): it's like a handy household tool that's completely clueless but it doesn't work in the dark!
|
| | | | | | | | | |
Date: 07 Apr 2005 00:33:21
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:uLd*[email protected], David Richerby <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > David Kane <[email protected]> wrote: >> David Richerby <[email protected]> wrote: >>> David Kane <[email protected]> wrote: >>>> A computer would characterize the quality of the moves and then >>>> match that profile to those of other players >>> >>> I think there's an awful lot of complexity hidden by the words >>> `would characterize'. Do you have any ideas about how such a >>> characterization would be performed? >> >> Two ideas seem promising. A simple way that would almost certainly >> work well for club players, would be to calculate something like >> blunder rate vs. ply depth and fit that to empirical data. > > I'm not at all sure what you mean by `blunder rate vs ply depth'. > Could you explain? > > Even without understanding what your statistic means, there's a huge > problem with the method. What exactly is this `empirical data' you're > going to fit to the statistic? It can't be any sort of existing > rating because we're assuming a priori that move-based ratings are > not directly correlated with result-based ratings -- if they were, it > would be much easier to measure the result-based rating directly than > to approximate it with a move-based rating. And I can't think of any > empirical data on chess games that isn't a rating. > > >> A more general algorithm would be to generate something like the >> probability that players of a certain strength would make a certain >> move in every position. > > That doesn't measure how good a player is: it measures how similar > their play is to the reference engine. Why should my rating be a > measure of how often I agree with FIDE SuperChess 5.3.8? What > happens when somebody writes a better chess program? The feels > incredibly arbitrary to me. > > >> (The details are left as an exercize for the reader) > > *cough* I realise you're being light hearted but you don't have a > rating scheme until these details are worked out or, at the very > least, sketched. It's all very well omitting the details when they're > tedious or obvious but, here, they're the very substance of the > discussion. > > >> When Elo devised his result-based rating method, there was nothing >> but head-to-head results upon which to predict future results. Now >> we are in a different era. > > Careful. We're not in the different era until we actually have this > move- based rating system. Remember that ``computers are very > powerful'' does not mean ``computers can do everything.'' > > >> I think eventually ordinary chess players will view their "move- >> rating" as their real rating, and understand their "result-rating" >> to be a less accurate approximation of it. > > I've still not seen anything that convinces me that move-based > ratings are well-defined or feasibly computable. It's premature to > declare that they're the future of ratings. > > > Dave. Great, guys. All of you have pluses. In my opinion, one game doesn't tell anything about a player. Ok, he's 2000 but has very bad day and plays as 1800 rated. So here comes "The ELO". It gives info about player's overall performance through longer period of time. But... (there are allways buts) Game moves still CAN indicate one's strength. For example, a player can play substandard moves but I think they are not far more weaker then the moves he would usually play. His moves still can tell us where he is. And as GM, you would probably make a good asumption how strong that player is. But the problem is if you play much stronger opponent. Your limited knowledge can't help you estimate stronger players ~ rating. Here should come in computers and help us value our opponents based on moves as we don't have his ELO.
|
| | | | | | | | | |
Date: 06 Apr 2005 09:37:54
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:uLd*[email protected]... > > Even without understanding what your statistic means, there's a huge > problem with the method. What exactly is this `empirical data' you're > going to fit to the statistic? It can't be any sort of existing rating > because we're assuming a priori that move-based ratings are not directly > correlated with result-based ratings -- if they were, it would be much > easier to measure the result-based rating directly than to approximate it > with a move-based rating. I'm talking about doing exactly this. Remember the "walks-by-a- game-in-progress-and-forms-an-opinion-about-the-players'-skill" idea? That makes use of comparing the moves as being played with our understanding of how players of certain strengths play. Quantify that and tell a computer how to do it, and you have move-based rating. The claim that result-based rating is "easier" is either outrageous or incredibly naive. If a huge percentage of the worlds' chess players don't have meaningful move-based ratings, that means the conditions to establish result-based ratings are difficult to meet. Just as Elo ratings have a way to convert from past results to future results, move based ratings also need a conversion algorithm from past moves played to future results. But you only need to figure it out once. Theoretical or practical difficulties are present in *both* systems. If I haven't played a rated game in 15 years, what is the value of my result-based rating? We don't know. I'm not meeting the condition of playing actively in my pool. Bureaucracies running the rating system might opt to purge me, they might opt to leave me in the system at my last rating. They might do a study of people in my situation and come up with something else to do. A move-based rating is potentially superior (i.e. more accurate ratings will be obtained with fewer games) and more universal (ratings won't be tied to a particular chess administration). But it *will* also have all the imperfections found in Elo ratings, no doubt about it.
|
| | | | | | | | | | |
Date: 07 Apr 2005 16:24:37
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> Even without understanding what your statistic means, there's a huge >> problem with the method. What exactly is this `empirical data' you're >> going to fit to the statistic? It can't be any sort of existing rating >> because we're assuming a priori that move-based ratings are not >> directly correlated with result-based ratings -- if they were, it would >> be much easier to measure the result-based rating directly than to >> approximate it with a move-based rating. > > I'm talking about doing exactly this. Remember the "walks-by-a- > game-in-progress-and-forms-an-opinion-about-the-players'-skill" > idea? That makes use of comparing the moves as being played with > our understanding of how players of certain strengths play. In that case, there's a definite inconsistency in your argument. At various points, you have argued (or seemed to me to have argued, at least) that a move-based ratings would be a more accurate measure of a player's strength than a result-based rating. If that is the case, you can't use result-based ratings to calibrate a move-based rating scheme because they measure different things. On the other hand, if they do measure the same thing, it is dramatically easier to calculate ratings from results than from moves so why not just do that? In this scenario, the only advantage of a move-based rating scheme is that you can calculate such a rating from fewer games. That would be useful but the world seems to get on quite happily having players be unrated in their first few tournaments. > Quantify that and tell a computer how to do it, and you have move-based > rating. Yes, you keep saying that. But nobody so far in this discussion has been able to come up with a way of quantifying it. This is a very significant point and you come across (perhaps unintentionally) as sweeping it under the carpet. > The claim that result-based rating is "easier" is either outrageous > or incredibly naive. I'm not at all sure what you mean by that sentence. Calculating a result- based rating is trivial. Calculating a move-based rating seems to be very, very difficult. > A move-based rating is potentially superior (i.e. more accurate ratings > will be obtained with fewer games) and more universal (ratings won't be > tied to a particular chess administration). That `more universal' point doesn't seem to work. The ratings would be tied to whoever built the computer system that calculated it. The system in question would be at least as complicated as a top-class chess engine so it's not feasible to hope that two organizations could build systems that would assign the same rating. Dave. -- David Richerby Swiss Carnivorous Beer (TM): it's like www.chiark.greenend.org.uk/~davidr/ a refreshing lager but it eats flesh and it's made in Switzerland!
|
| | | | | | | | | | | |
Date: 07 Apr 2005 09:26:27
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:nyg*[email protected]... > David Kane <[email protected]> wrote: > > David Richerby <[email protected]> wrote: > >> Even without understanding what your statistic means, there's a huge > >> problem with the method. What exactly is this `empirical data' you're > >> going to fit to the statistic? It can't be any sort of existing rating > >> because we're assuming a priori that move-based ratings are not > >> directly correlated with result-based ratings -- if they were, it would > >> be much easier to measure the result-based rating directly than to > >> approximate it with a move-based rating. > > > > I'm talking about doing exactly this. Remember the "walks-by-a- > > game-in-progress-and-forms-an-opinion-about-the-players'-skill" > > idea? That makes use of comparing the moves as being played with > > our understanding of how players of certain strengths play. > > In that case, there's a definite inconsistency in your argument. At > various points, you have argued (or seemed to me to have argued, at least) > that a move-based ratings would be a more accurate measure of a player's > strength than a result-based rating. If that is the case, you can't use > result-based ratings to calibrate a move-based rating scheme because they > measure different things. No. The ratings would have exactly the same objective: predicting future results. Because move based rating makes use of more information about the game than just the result, they are potentially more accurate. > > I'm not at all sure what you mean by that sentence. Calculating a result- > based rating is trivial. Calculating a move-based rating seems to be > very, very difficult. The calculation is trivial (although one could argue that collecting the information about results isn't quite trivial) But there is no mechanism forcing the world's chessplayers to play each other with a certain frequency, and no way to even contemplate arriving at that state. That is the inherent flaw in result rating: it uses only a small piece of information about each game and therefore needs lots of games. Developing a tool to do move-based rating would be very, very difficult. But once you had it, it would be pretty straightforward to run. > > A move-based rating is potentially superior (i.e. more accurate ratings > > will be obtained with fewer games) and more universal (ratings won't be > > tied to a particular chess administration). > > That `more universal' point doesn't seem to work. The ratings would be > tied to whoever built the computer system that calculated it. The system > in question would be at least as complicated as a top-class chess engine > so it's not feasible to hope that two organizations could build systems > that would assign the same rating. In theory, competing tools could be tested and the world could adopt the best one. In practice, once we have a single such tool that works better than result-rating, it will become the gold standard.
|
| | | | | | | | | | | | |
Date: 07 Apr 2005 18:57:20
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> In that case, there's a definite inconsistency in your argument. At >> various points, you have argued (or seemed to me to have argued, at >> least) that a move-based ratings would be a more accurate measure of a >> player's strength than a result-based rating. If that is the case, you >> can't use result-based ratings to calibrate a move-based rating scheme >> because they measure different things. > > No. The ratings would have exactly the same objective: predicting > future results. Because move based rating makes use of more information > about the game than just the result, they are potentially more accurate. But if you calibrate your move-based system by using the results-based system, then the move-based system is just an expensive way of calculating the result-based rating! We already have a cheap way of calculating that. >> I'm not at all sure what you mean by that sentence. Calculating a >> result-based rating is trivial. Calculating a move-based rating seems >> to be very, very difficult. > > The calculation is trivial (although one could argue that collecting > the information about results isn't quite trivial) The calculation is categorically not trivial! The input to a result-based rating system is a bunch of game results. You look at who beat whom and what their ratings were and calculate the new ratings using a formula that can be written on the back of a postage stamp. The input to a move-based rating system is a bunch of games and... well, nobody's said what the calculation is but it takes into account the moves that were made and how good they were and we don't even know how to calculate that. > Developing a tool to do move-based rating would be very, very > difficult. But once you had it, it would be pretty straightforward > to run. > > [...] > > In theory, competing tools could be tested and the world could adopt > the best one. Not if it's very, very difficult to do. And do you have any criteria in mind for choosing `the best one'? Dave. -- David Richerby Transparent Postman (TM): it's like www.chiark.greenend.org.uk/~davidr/ a man who delivers the mail but you can see right through it!
|
| | | | | | | | | | | | | |
Date: 08 Apr 2005 00:24:34
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:OWf*[email protected], David Richerby <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > David Kane <[email protected]> wrote: >> David Richerby <[email protected]> wrote: >>> In that case, there's a definite inconsistency in your argument. At >>> various points, you have argued (or seemed to me to have argued, at >>> least) that a move-based ratings would be a more accurate measure >>> of a player's strength than a result-based rating. If that is the >>> case, you can't use result-based ratings to calibrate a move-based >>> rating scheme because they measure different things. >> >> No. The ratings would have exactly the same objective: predicting >> future results. Because move based rating makes use of more >> information about the game than just the result, they are >> potentially more accurate. > > But if you calibrate your move-based system by using the results-based > system, then the move-based system is just an expensive way of > calculating the result-based rating! We already have a cheap way of > calculating that. > > >>> I'm not at all sure what you mean by that sentence. Calculating a >>> result-based rating is trivial. Calculating a move-based rating >>> seems to be very, very difficult. >> >> The calculation is trivial (although one could argue that collecting >> the information about results isn't quite trivial) > > The calculation is categorically not trivial! The input to a > result-based rating system is a bunch of game results. You look at > who beat whom and what their ratings were and calculate the new > ratings using a formula that can be written on the back of a postage > stamp. The input to a move-based rating system is a bunch of games > and... well, nobody's said what the calculation is but it takes into > account the moves that were made and how good they were and we don't > even know how to calculate that. Bingo! The last sentence.... We still don't know how to calculate that. I thought I can do it somehow with Fritz so I posted this topic. >> Developing a tool to do move-based rating would be very, very >> difficult. But once you had it, it would be pretty straightforward >> to run. >> >> [...] >> >> In theory, competing tools could be tested and the world could adopt >> the best one. > > Not if it's very, very difficult to do. And do you have any criteria > in mind for choosing `the best one'? > > > Dave.
|
| | | | | | | | | | | | | |
Date: 07 Apr 2005 14:00:41
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:OWf*[email protected]... > David Kane <[email protected]> wrote: > > No. The ratings would have exactly the same objective: predicting > > future results. Because move based rating makes use of more information > > about the game than just the result, they are potentially more accurate. > > But if you calibrate your move-based system by using the results-based > system, then the move-based system is just an expensive way of calculating > the result-based rating! We already have a cheap way of calculating that. No. We use our knowledge of chess as played by humans to figure out the "conversion algorithm" between move history and winning expectancy. This would be a huge task. Once we have the conversion, though, we can keep using it with very little future work (feed in the moves and press go). Moreover it would be available for use by the millions of chessplayers for whom result ratings do *not* work, i.e. those who are not playing actively in a rating pool.
|
| | | | | | | | | | | | | | |
Date: 07 Apr 2005 23:35:53
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > David Richerby <[email protected]> wrote: >> David Kane <[email protected]> wrote: >>> No. The ratings would have exactly the same objective: predicting >>> future results. Because move based rating makes use of more >>> information about the game than just the result, they are potentially >>> more accurate. >> >> But if you calibrate your move-based system by using the results-based >> system, then the move-based system is just an expensive way of >> calculating the result-based rating! We already have a cheap way of >> calculating that. > > No. We use our knowledge of chess as played by humans to figure out the > "conversion algorithm" between move history and winning expectancy. Why do you need to? We already have a perfectly good way of estimating winning expectation by looking at results. What you're proposing is that there should be a way of looking at a player's moves and saying, ``He looks like an 1800 Elo player.'' We already know how to work out that somebody is an 1800 Elo player without going through all of this palaver. > This would be a huge task. Once we have the conversion, though, we can > keep using it with very little future work (feed in the moves and press > go). Ah, I see what you mean. You believe that the work in generating the rating system is a one-off job that only needs to be done once and will produce a very efficient algorithm that, given a game, will very quickly go though the moves and determine the rating of the player. I don't believe that it could possibly work like that. The system would have to be able to deal with any legal chess position and know what makes a good move in that position. Since there are far too many chess positions to store all of this in a pre-calculated table (which is to say, a 32-man tablebase), this would have to be re-calculated for each game. It would, surely, be at least as expensive as getting a computer to analyse games and suggest improvements to them. At the very least, that would be several seconds per move. At ten seconds a move, a tournament with 500 games (a 200-player, 5-round weekend Swiss) averaging 40 moves each would take nearly five days of computer time to rate, even before you consider the massive data-entry problem. > Moreover it would be available for use by the millions of chessplayers > for whom result ratings do *not* work, i.e. those who are not playing > actively in a rating pool. If they're not playing against other rated players, why do they need a rating? Dave. -- David Richerby Mentholated Peanut (TM): it's like a www.chiark.greenend.org.uk/~davidr/ roasted nut but it's invigorating!
|
| | | | | | | | | | | | | | | |
Date: 07 Apr 2005 18:48:37
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:fQF*[email protected]... > David Kane <[email protected]> wrote: > > David Richerby <[email protected]> wrote: > >> David Kane <[email protected]> wrote: > >>> No. The ratings would have exactly the same objective: predicting > >>> future results. Because move based rating makes use of more > >>> information about the game than just the result, they are potentially > >>> more accurate. > >> > >> But if you calibrate your move-based system by using the results-based > >> system, then the move-based system is just an expensive way of > >> calculating the result-based rating! We already have a cheap way of > >> calculating that. > > > > No. We use our knowledge of chess as played by humans to figure out the > > "conversion algorithm" between move history and winning expectancy. > > Why do you need to? We already have a perfectly good way of estimating > winning expectation by looking at results. We have "a" way that is neither good nor perfect for everyone. > > > This would be a huge task. Once we have the conversion, though, we can > > keep using it with very little future work (feed in the moves and press > > go). > > Ah, I see what you mean. You believe that the work in generating the > rating system is a one-off job that only needs to be done once and will > produce a very efficient algorithm that, given a game, will very quickly > go though the moves and determine the rating of the player. I don't > believe that it could possibly work like that. > > The system would have to be able to deal with any legal chess position and > know what makes a good move in that position. Since there are far too > many chess positions to store all of this in a pre-calculated table (which > is to say, a 32-man tablebase), this would have to be re-calculated for > each game. It would, surely, be at least as expensive as getting a > computer to analyse games and suggest improvements to them. At the very > least, that would be several seconds per move. At ten seconds a move, a > tournament with 500 games (a 200-player, 5-round weekend Swiss) averaging > 40 moves each would take nearly five days of computer time to rate, even > before you consider the massive data-entry problem. You have completely lost me. What I am proposing is that we determine how good the moves a player makes are and then apply the (yet to be determined) formula that converts those to winning prediction. Computationally it will be like the order of running a game through Fritz, probably more. When I said that this was little future work, I was referring to the fact that it would be computers doing the work. Your last comments indicate that you think that I have proposed that this would be something designed to be run by chess organizations to replace Elo ratings. I have never suggested that. However, I am heartened that you are attacking the idea on feasibility grounds, since maybe it means a minimal acceptance of the concept.
|
| | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 19:02:29
From: David Kane
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in message news:[email protected]... > > You have completely lost me. What I am proposing is that we determine > how good the moves a player makes are and then apply the (yet to be > determined) formula that converts those to winning prediction. > Computationally > it will be like the order of running a game through Fritz, probably > more. When Let me clarify. I am not suggesting that we only need a single game to get a good rating. Just as Elo ratings get better with more games, move-based ratings would get better with more moves. But if the information contained in the moves is put to its fullest use, we should be able to arrive at a better rating with fewer total games than Elo in some cases. The downside is that it will take more time to enter the games and the result, and then much more computational time. I also think that a hybrid system using both result and move history has merit since each rating method will have its range of validity.
|
| | | | | | | | | | | | |
Date: 07 Apr 2005 16:52:42
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > The calculation is trivial (although one could argue that collecting > the > information about results isn't quite trivial) How can you argue that the calculation is trivial when you haven't proposed what the calculation is? Even one one test you've proposed which we could actually measure (blunder rate as a function of time) requires orders of magnitude more effort to calculate than the Elo system. What you propose could end up being trivial to calculate, but how the hell would you (or anyone else) know? Until you actually have a system, it's absurd to talk about how easy that system is. This is especially the case when you seem to continually underestimate the practical difficulties involved in your system. If it's conceptually difficult (which it is, as nobody has been able to lay out the details yet) it's likely to be computationally difficult, as well. -Ron
|
| | | | | | | | | | | | | |
Date: 07 Apr 2005 10:04:15
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-932D8B.09524107042005@newssvr14-ext.news.prodigy.com ... > In article <[email protected]>, > "David Kane" <[email protected]> wrote: > > > The calculation is trivial (although one could argue that collecting > > the > > information about results isn't quite trivial) > > How can you argue that the calculation is trivial when you haven't > proposed what the calculation is? > I was conceding that the Elo calculation is trivial, before pointing out the more fundamental difficulty with Elo.
|
| | | | | | | | | | |
Date: 06 Apr 2005 18:38:45
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > If a huge percentage of the worlds' > chess players don't have meaningful move-based ratings, > that means the conditions to establish result-based ratings > are difficult to meet. Of course, if you had a record of their game scores (which would be required in order to rate them by a move-based rating system) you would also have a record of their results, which would make an Elo-like system easy to implement. If you want to claim a move-based rating system is easier, then, by all means, propose a system and we'll test it. That's the great thing about all this, you see: it's completely testable. For example, we know that the Glicko system is more accurate than the Elo system. You're dodging this point because you don't actually have a rating system. You just have an idea. If you think it's so damn easy, go for it. Propose a system. We'll test it. Until you have a system, it's absurd for you to claim it's easier or better than the system that actually exists. -Ron
|
| | | | | | | | |
Date: 06 Apr 2005 00:56:24
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > Two ideas seem promising. A simple way that would > almost certainly work well for club players, would be > to calculate something like blunder rate vs. ply depth > and fit that to empirical data. I think there's some merit to this approach, although, again, it runs up against the unsound-sacrifice-that-works problem. These moves SHOULD be the losing moves, but they end up being the winning moves. Are they only a blunder if the opponent properly refutes them? (This is not to say that the whole idea is invalid, but it is a major hole that needs to be addressed). > A more general algorithm would be to generate > something like the probability that players of a > certain strength would make a certain move in every > position. (The details are left as an exercize for > the reader) Then a best fit rating over all of the moves > of all of the games would be calculated. This doesn't work because all (say) 1800 players are not created equal. One 1800 player might be a very strong attacking player, but an indifferent defender and abysmal in the endgame. Another 1800 player but thrive in simplified, technical positions but struggle in raging attacks. The only thing "typical" about a typical 1800 player is that he wins enough to earn an 1800 rating. There are many, many roads to that rating.
|
| | | | | | | | | |
Date: 06 Apr 2005 09:41:33
From: mrlount
Subject: Re: Rating from moves??
|
Ron wrote: > In article <[email protected]>, > "David Kane" <[email protected]> wrote: > >> Two ideas seem promising. A simple way that would >> almost certainly work well for club players, would be >> to calculate something like blunder rate vs. ply depth >> and fit that to empirical data. > > I think there's some merit to this approach, although, again, it runs > up against the unsound-sacrifice-that-works problem. These moves SHOULD > be the losing moves, but they end up being the winning moves. > > Are they only a blunder if the opponent properly refutes them? > > (This is not to say that the whole idea is invalid, but it is a major > hole that needs to be addressed). > >> A more general algorithm would be to generate >> something like the probability that players of a >> certain strength would make a certain move in every >> position. (The details are left as an exercize for >> the reader) Then a best fit rating over all of the moves >> of all of the games would be calculated. > > This doesn't work because all (say) 1800 players are not created equal. > One 1800 player might be a very strong attacking player, but an > indifferent defender and abysmal in the endgame. Another 1800 player > but thrive in simplified, technical positions but struggle in raging > attacks. > > The only thing "typical" about a typical 1800 player is that he wins > enough to earn an 1800 rating. There are many, many roads to that > rating. Here is my main problem with the rating by moves idea, rather than the rating by result. Mrs Smith and Mr Brown pop to their local community centre to watch a chess tournament on the weekend. They stand behind board number 12 which has John rated 2200 playing white and Dave 1750 playing black. Mr Smith says to Mr Brown "Watch John play. I've seen him play before, he has an amazing tactical brain." Mr Brown "Ah, so I guess he's got this in the bag, especially since he's got the higher rating" Mr Smith "No, Dave wins this tournament every year. John has never won a single chess game, he plays 60 first class moves which Fritz can not better. He then loses a drawn endgame by a blatant blunder." Mr Brown "So what's the point of a rating if it tells me nothing?"
|
| | | | | | | | | | |
Date: 10 Apr 2005 10:55:14
From: tom_h
Subject: Re: Rating from moves??
|
Bark! wrote: > Na njuzima:[email protected], > tom_h <[email protected]> mukotrpnim radom iznade ovu umotvorinu: > > > cut > > David has given a few examples already that show that, but I wanted to > > add another one from a sort of different angle. Suppose you are at a > > casino and you win on the slots twice... these are the only games you > > ever played. You still should not expect to win again a third time, > > assuming the machine is working properly, etc. There are no moves to > > talk about, but it makes the point that past results are not > > necessarily the best predictors of future results... in this case, a > > probabilistic analysis of the odds for that machine would be the best > > predictor. > > cut > > Sorry, I have to comment this example. It's bad because you can't influence > the outcome of the slot game. There's no factor as your knowledge which you > can win slot games with. It's just luck. And chess is not about luck. I know that chess is different than slots... however, I was trying to demonstrate a point, which I think is applicable to both games, by using an example that might help people think about it in a different way. I can re-write a similar example using chess though. In my first tournament ever I get paired against a 2000. I am, let's say, actually 1200 strength. I play a lot of bad moves, but then my opponent plays a move that blunders his queen and quickly resigns because he is sandbagging for the upcoming World Open... In the next round I get paired against another 2000 who had taken a 1/2 point bye because he couldn't make the first round. It turns out his flight was delayed so he barely makes it to the second round with a couple of minutes to spare on his clock. He bashes out 30 moves or so, getting a winning position, and then his flag falls. Now I have 2-0 against 2000 rated players. However, I was lucky to get that score, and I am unlikely to win the next time I play against a 2000. ... Another thought occurred to me related to this thread. What is a rating supposed to mean anyway? I always assumed it was a predictor of future results, but the more I think about it, I realize that a lot of people probably don't think of it that way. Maybe people see it more as a recognition for past results. This would explain the rationale behind rating floors for example. So, for the sake of my argument above, I still assumed that a rating is supposed to be a predictor of future results. However, I will be the first to admit that if it is supposed to mean something else, then my slots example and the one given above don't make much sense. ... Finally, I probably won't respond again on this thread. The reason is just that it is taking me too long to compose a message, and I *really* need to be studying for classes instead. I apologize for jumping into the conversation and leaving so quickly, but I just won't have much time to spend on this over the next few days. Best wishes, Tom
|
| | | | | | | | | | | |
Date: 10 Apr 2005 21:39:38
From: David Richerby
Subject: Re: Rating from moves??
|
tom_h <[email protected] > wrote: > I can re-write a similar example using chess though. In my first > tournament ever I get paired against a 2000. I am, let's say, actually > 1200 strength. I play a lot of bad moves, but then my opponent plays a > move that blunders his queen and quickly resigns because he is > sandbagging for the upcoming World Open... > > In the next round I get paired against another 2000 who had taken a 1/2 > point bye because he couldn't make the first round. It turns out his > flight was delayed so he barely makes it to the second round with a > couple of minutes to spare on his clock. He bashes out 30 moves or so, > getting a winning position, and then his flag falls. Anyone can get lucky -- so what? Before too long, your rating will settle back to what it should be. Dave. -- David Richerby Revolting Hungry Cat (TM): it's like www.chiark.greenend.org.uk/~davidr/ a cat but it'll eat you and turn your stomach!
|
| | | | | | | | | | | | |
Date: 11 Apr 2005 00:54:05
From: Ron
Subject: Re: Rating from moves??
|
In article <63n*[email protected] >, David Richerby <[email protected] > wrote: > Anyone can get lucky -- so what? Before too long, your rating will settle > back to what it should be. The point is more that the value of any one move (as far as calculating a rating) is incredibly low. Therefore, the idea (that from a small number of games you could deduce someone's strength) is fairly limited. To get enough information you'd have to have a lot of games, and you'd have to have some sense of the conditions under which those games were played. The results might not therefore be meaningful to players who don't already have an Elo rating. Another thing to consider is that in a given game you may simply not have the opportunities to play moves that k you as a very strong player. (A trivial example will demonstrate the idea. 1.f3 e4 2.g4 Qh4 mate. How strong is black? Well, even assuming you can make probabilistic curves about who's likely to play what when, basically you're going to flatline black's rating above, say, 900 or so - a point above which you'd expect the players to see 2.Qh4 mate 99% of the time. Obviously, more complex games will have higher ceiling, but the point is that a 1200- and 1800- rated players may look like very different players when playing against a 1500-rated player, but they probably look pretty similar when playing against an 800-rated player. This may not seem that important, but remember that one of the alleged virtues of this idea is the ability to rate people who haven't played in tournaments, and therefore, basically haven't played against strong opposition.)
|
| | | | | | | | | | | | | |
Date: 10 Apr 2005 18:25:17
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-E9188F.17540410042005@newssvr13-ext.news.prodigy.com ... > > Another thing to consider is that in a given game you may simply not > have the opportunities to play moves that k you as a very strong > player. > This is exactly the same as beating rated someone much lower rated in Elo. Some moves don't have much information and some results don't have much information.
|
| | | | | | | | | | | | | | |
Date: 11 Apr 2005 06:58:24
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > > Another thing to consider is that in a given game you may simply not > > have the opportunities to play moves that k you as a very strong > > player. > > > > This is exactly the same as beating rated someone much lower > rated in Elo. Some moves don't have much information and some > results don't have much information. Sure. The Elo system accounts for this by not giving you many points for beating someone weaker than you. But -- and this is a key difference -- the Elo system gives a lot of value for a move like 2.Qh4++ (or rather, the result that comes from it) if both players are equally weak. The 1000-rated player who beats the 900-rated player gets on that trap gets a reasonable number of points. But since your system ignores strength, if has to give the players the same number of points regardless -- it has to treat the 1000-rated players win over the 800-rated player the same way it treats the 1200-players win over that same player, if they play similar moves. Also, you've been alleging that this system has value for people who haven't played in tournaments. I suspect, on the other hand, that the standard of play outside of tournaments is generally so low that you run into this problem. Tournaments deal with it by having you play progressively stronger players. But if you want to apply your system (which still has only been vaguely defined) to tournament players, you have to show a reason why it's better than the Elo system. So far, the priy virtue you've offered is that if doesn't required tournament participation. -Ron
|
| | | | | | | | | | | | | | | |
Date: 11 Apr 2005 00:49:44
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-4B226E.23582310042005@newssvr13-ext.news.prodigy.com ... > In article <[email protected]>, > "David Kane" <[email protected]> wrote: > > > > Another thing to consider is that in a given game you may simply not > > > have the opportunities to play moves that k you as a very strong > > > player. > > > > > > > This is exactly the same as beating rated someone much lower > > rated in Elo. Some moves don't have much information and some > > results don't have much information. > > Sure. The Elo system accounts for this by not giving you many points for > beating someone weaker than you. But -- and this is a key difference -- > the Elo system gives a lot of value for a move like 2.Qh4++ (or rather, > the result that comes from it) if both players are equally weak. The > 1000-rated player who beats the 900-rated player gets on that trap gets > a reasonable number of points. But since your system ignores strength, > if has to give the players the same number of points regardless -- it > has to treat the 1000-rated players win over the 800-rated player the > same way it treats the 1200-players win over that same player, if they > play similar moves. Not at all. A move that is highly predictable for a player of that rating contains no information and hence doesn't change the rating. This is exactly the same as result rating - results that are highly predictable don't change the rating. To repeat, the only way that move-rating would not work *in principle* is if there were something other than the moves you make that correlate with your results (over the long term), e.g. things like demeanor or appearance which influence the play of your opponents, independent of your moves. > But if you want to apply your system (which still has only been vaguely > defined) to tournament players, you have to show a reason why it's > better than the Elo system. So far, the priy virtue you've offered > is that if doesn't required tournament participation. Here are some questions that a move-rating tool might be able to answer, and Elo can't (or very poorly). Two chessplayers play every Sunday. "Elo rating" their pool of 2 would just give an uninteresting mathematical reduction of their recent results. It provides no answer to questions of interest to them such as: Are they improving? How good are they compared to other chessplayers? Players are playing, but not in a rating pool. Are they getting better? Is an influx of youngsters deflating the rating pool? By how much? Does the pressure of tournament games change the quality of chess compared to casual games? How is the quality of chess impacted by rule changes (time control changes etc)? Are there players whose results are better/worse than their moves would predict? How important is this factor? How good was Morphy? (Yes, I know about historical Elo, and am also aware of its deficiencies) How well did Fischer play in his Candidates streak? etc.
|
| | | | | | | | | | | | | | | | |
Date: 13 Apr 2005 00:44:31
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], David Kane <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: >cut > Two chessplayers play every Sunday. "Elo rating" > their pool of 2 would just give an uninteresting > mathematical reduction of their recent results. It provides > no answer to questions of interest to them such as: > Are they improving? How good are they compared to > other chessplayers? > cut Great example. Let's say they play daily for a month and the result is 0,5:0,5 every game. After a month they have the same ELO but they could both improve dramatically and their ELO doesn't show that.
|
| | | | | | | | | | |
Date: 06 Apr 2005 03:49:16
From: David Kane
Subject: Re: Rating from moves??
|
"mrlount" <[email protected] > wrote in message news:[email protected]... > > Here is my main problem with the rating by moves idea, rather than the > rating by result. > > > Mrs Smith and Mr Brown pop to their local community centre to watch a chess > tournament on the weekend. > > They stand behind board number 12 which has John rated 2200 playing white > and Dave 1750 playing black. > > Mr Smith says to Mr Brown "Watch John play. I've seen him play before, he > has an amazing tactical brain." > > Mr Brown "Ah, so I guess he's got this in the bag, especially since he's got > the higher rating" > > Mr Smith "No, Dave wins this tournament every year. John has never won a > single chess game, he plays 60 first class moves which Fritz can not > better. He then loses a drawn endgame by a blatant blunder." > > Mr Brown "So what's the point of a rating if it tells me nothing?" > The flaw in your little scenario (ignoring its unrealistic characteristics) is that "blunders" are moves, too. If John makes endgame moves like an 1100 player, then he won't get a 2200 rating. The point isn't solely that the player who wins makes better moves than the person who loses but also that win or lose the moves say something about playing strength. For example, if a player rated 1800 starts making moves characteristic of a 1600 player, he might still win if he is playing someone who is moving like a 1400 player. But his rating should go down.
|
| | | | | | | | | | | |
Date: 06 Apr 2005 12:23:13
From: David Richerby
Subject: Re: Rating from moves??
|
David Kane <[email protected] > wrote: > The point isn't solely that the player who wins makes better moves than > the person who loses Because, as has been demonstrated, this isn't necessarily the case. As Tartakower once said, ``The winner of the game is the one who makes the next to the last mistake.'' :-) Thanks, by the way, to everyone for an interesting and thought-provoking discussion on this matter. Dave. -- David Richerby Flammable Aluminium Priest (TM): it's www.chiark.greenend.org.uk/~davidr/ like a man of the cloth that's really light but it burns really easily!
|
| | | | | | | | | | | | |
Date: 06 Apr 2005 10:29:43
From: David Kane
Subject: Re: Rating from moves??
|
"David Richerby" <[email protected] > wrote in message news:rpt*[email protected]... > David Kane <[email protected]> wrote: > > The point isn't solely that the player who wins makes better moves than > > the person who loses > > Because, as has been demonstrated, this isn't necessarily the case. As > Tartakower once said, ``The winner of the game is the one who makes the > next to the last mistake.'' :-) > No one has demonstrated anything of the sort. Making blunders is tied to strength and has predictive value, but your side mysteriously denies that they are "moves" that I can use in my move-based rating scheme. Of course, blunders are the moves with the *most* predictive value. I see a lot of Kindergartners playing chess. I am positive that if I rank them according to "rate of leaving pieces left en prise" I will have a perfect ordering of their ability.
|
| | | | | | | | | | | | | |
Date: 06 Apr 2005 19:56:25
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > No one has demonstrated anything of the sort. Making > blunders is tied to strength and has predictive value, but your > side mysteriously denies that they are "moves" that I can use in > my move-based rating scheme. > > Of course, blunders are the moves with the *most* predictive > value. I see a lot of Kindergartners playing chess. I am > positive that if I rank them according to "rate of leaving > pieces left en prise" I will have a perfect ordering of their > ability. The last sentence is a good illustration what the flaw in your reasoning is: it is useless to use just one parameter (rate of leaving pieces en prise) if you can use the _actual_ results, which are a culmination of _all_ relevant parameters, including "move strength" and "leaving pieces en prise". I guess it's mainly because you oversee that the defined objective of a chess game is not a theory, but a predefined agreement. I also suspect you can't detach from the idea that "chess strenght" is some absolute, isolated value. Regardless of stategy, tactics, psychology, health, fitness, endurance, whatever, the strongest player is he or she who _wins_ the most games. As long as the target of chess is winning games, the best way to make a strength order is to use actual results, not just one parameter like "the quality of a move". Your suggestion that the Elo rating has flaws might be true; but your solution in which you use one isolated parameter to predict results instead of _actual_ results to predict results is even more flawed. Interesting, no doubt. But by definition certainly not more reliable. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | |
Date: 07 Apr 2005 00:13:32
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], CeeBee <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > >> No one has demonstrated anything of the sort. Making >> blunders is tied to strength and has predictive value, but your >> side mysteriously denies that they are "moves" that I can use in >> my move-based rating scheme. >> >> Of course, blunders are the moves with the *most* predictive >> value. I see a lot of Kindergartners playing chess. I am >> positive that if I rank them according to "rate of leaving >> pieces left en prise" I will have a perfect ordering of their >> ability. > > > The last sentence is a good illustration what the flaw in your > reasoning is: it is useless to use just one parameter (rate of > leaving pieces en prise) if you can use the _actual_ results, > which are a culmination of _all_ relevant parameters, including > "move strength" and "leaving pieces en prise". > > I guess it's mainly because you oversee that the defined objective > of a chess game is not a theory, but a predefined agreement. > I also suspect you can't detach from the idea that "chess > strenght" is some absolute, isolated value. > > Regardless of stategy, tactics, psychology, health, fitness, > endurance, whatever, the strongest player is he or she who _wins_ > the most games. > > As long as the target of chess is winning games, the best way to > make a strength order is to use actual results, not just one > parameter like "the quality of a move". > > Your suggestion that the Elo rating has flaws might be true; but > your solution in which you use one isolated parameter to predict > results instead of _actual_ results to predict results is even > more flawed. > > Interesting, no doubt. But by definition certainly not more > reliable. Ok. I agree with the tought that the result of a player is overall performance he made (moves, tactics, endgame...). And that brings back my question again - we know nothing about this guy, he plays a game, without ELO, how can we predict his rating just from one game - therefore only moves - not ELO? I guess we all can give our approximation (successful more or less) about his rating based on moves he plays. But even good player can have a bad day at work so he could play substandard moves. Someone answered my question - It can't be done. Full stop. So do you agree with this? Thanks again for a lively discussion.
|
| | | | | | | | | | | | | | | |
Date: 06 Apr 2005 23:16:59
From: CeeBee
Subject: Re: Rating from moves??
|
"Bark!" <[email protected] > wrote in rec.games.chess.analysis: > Ok. I agree with the tought that the result of a player is > overall performance he made (moves, tactics, endgame...). And > that brings back my question again - we know nothing about this > guy, he plays a game, without ELO, how can we predict his rating > just from one game - therefore only moves - not ELO? We can't but approximate, because rating is defined as a comparison of results in a pool of players. "Strong" is - based on the objective of the chess game - _defined_ by result (_winning_, not playing good moves), while that result is _influenced_ by factors like strong moves, etc. The discussion point has become: is "move rating" a better way to predict strength (=result) than "result rating"? The answer is clearly "no", as no proof has been given that move quality is more decisive in predicting future results than the actual former game results, and no valid model of the theory has been shown. A theory is supported by proof, and following that either confirmed by facts and predictions or disproven by falsification of that proof. A theory is not an idea that's "true until proven otherwise". > Someone answered my question - It can't be done. Full > stop. So do you agree with this? > > Thanks again for a lively discussion. The beauty of Usenet discussions is that no one is able to decide when there 's going to be a full stop to the discussion. Don't want further discussion - don't bring up new discussion points - don't respond to old ones. That's the only reliable way. If you ignore, you take the chance of getting a response. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | |
Date: 06 Apr 2005 13:31:56
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > > Your suggestion that the Elo rating has flaws might be true; but > your solution in which you use one isolated parameter to predict > results instead of _actual_ results to predict results is even > more flawed. > If you only think that it "might" be true, then you just don't understand Elo ratings well enough to continue. Your claim that I want to use "one isolated parameter" is also false. I've said that I want to use all the information available about all of the moves played. (I have also not precluded the possibility of some kind of hybrid system which weighs result history as well as move history) You are the one needlessly locked into looking only at one isolated parameter (the result). If my 6 year old, 9 year old and I all played a single game against Kasparov, we'd all lose and all have the same meaningless Elo rating. Anyone with any chess knowledge at all could easily rank the 3 of us by the quality of our losses. It's a clearcut illustration of rating results being objectively much inferior to rating the moves. Period. You seem to believe that accepting that inferiority is some kind of religious duty - I don't.
|
| | | | | | | | | | | | | | | |
Date: 06 Apr 2005 23:59:38
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > Your claim that I want to use "one isolated parameter" > is also false. I've said that I want to use all the information > available about all of the moves played. (I have also not > precluded the possibility of some kind of hybrid system > which weighs result history as well as move history) But so far you've only suggested two actual tests. One is "blunder rate" which I agree has some value, but your example of rating extremely weak players is not very useful because stronger players blunder so much less frequently that you may not be able to accomplish the original goal -- evaluate someone's strength through a single game (or a sample otherwise too small to acheive a meaningful Elo rating). Your other test (try to determine probabilistic curves for pariticular moves based on rating) fails because all players of the same rating don't play the same way. I think this is why some of this find aruging with you increasingly frustrating. You talk about your system like it's easy or obvious, but it breaks down. So when people accuse you of wanting to use one isolated parameter, that's because that's the only parameter you've suggested a reasonable methodology for analysis with. You want to use all information available. Great. Please tell us how you're going to do this.
|
| | | | | | | | | | | | | | | |
Date: 06 Apr 2005 21:40:40
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > You are the one needlessly locked into looking only > at one isolated parameter (the result). I'm sorry, but if you don't understand that the result of a chess game is not "one isolated parameter" but the _only_ objective of the whole game, there's not much to discuss left. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 00:15:01
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], CeeBee <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > >> You are the one needlessly locked into looking only >> at one isolated parameter (the result). > > > I'm sorry, but if you don't understand that the result of a chess > game is not "one isolated parameter" but the _only_ objective of the > whole game, there's not much to discuss left. Ok, peace, guys.
|
| | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 22:53:51
From: CeeBee
Subject: Re: Rating from moves??
|
"Bark!" <[email protected] > wrote in rec.games.chess.analysis: > Ok, peace, guys. With your rek you suggest there's a fight going on. This is not the case. Please understand first, judge later. The main problem in defending his alternative theory proposed by David is in questioning the validity of a set of rules that make up the game of chess. If you question the fact that in chess the agreement that winning the game is the objective, and your win/lose rate is thus the definition of strength relative to each other, there is not much sense in _discussing_, because you may claim that 2+2 does not equal 4, yet it's not a theory up for discussing, but a claim that an agreement is non-existent. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 16:00:11
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > "Bark!" <[email protected]> wrote in rec.games.chess.analysis: > > > > Ok, peace, guys. > > With your rek you suggest there's a fight going on. This is not > the case. Please understand first, judge later. > > The main problem in defending his alternative theory proposed by > David is in questioning the validity of a set of rules that make up > the game of chess. > > If you question the fact that in chess the agreement that winning > the game is the objective, and your win/lose rate is thus the > definition of strength relative to each other, there is not much > sense in _discussing_, because you may claim that 2+2 does not equal > 4, yet it's not a theory up for discussing, but a claim that an > agreement is non-existent. > You misunderstand the essence of the argument. The fact that something is an objective does not mean that it is the only or best way to predict future results. I've laid out numerous counterexamples in chess and other contests as clearly as I possibly can, but you continue to make this unfounded connection.
|
| | | | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 23:35:50
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > You misunderstand the essence of the argument. The fact that > something is an objective does not mean that it is the only or > best way to predict future results. I've laid out numerous > counterexamples in chess and other contests as clearly as > I possibly can, but you continue to make this unfounded > connection. In the rush you have forgotten the essence of your own argument. You claimed: > A move-based rating system does not have to be perfect to be > useful.The target is imperfect Elo ratings, not perfection. And in response to Ron's rek: >> If we had more games between these two opponents, we could >> calculate Elo >> based on performance. you claimed: > Or do even better rating the moves. What you have tried to claim is that "move quality" based prediction of a game result is more reliable than "result based" Elo prediction of a game result. However neither your theory nor your counterarguments have proven this. In fact all your counterexamples have been countered themselves extensively, yet you assume they have no validity and are beside the point because they don't support your theory. Your misunderstanding is that you think a theory is an idea true until disproven. It isn't. It's a coherent set of principles that explains a phenomonon or leads to a valid prediction. As soon as you prove that assessement of move quality gives a better prediction of the chances to win/lose/draw against another opponent than previous results in the pool of players in which those opponents play, you have your point. So far, you haven't been able to. You have only been able to point out that often good moves lead to good results (=wins), not that they give a better prediction of future wins or losses than the Elo rating. It's not a judgement about "personal flaws", but a conclusion based on the arguments you have presented. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | | | | | |
Date: 06 Apr 2005 18:52:27
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > > As soon as you prove that assessement of move quality gives a > better prediction of the chances to win/lose/draw against another > opponent than previous results in the pool of players in which > those opponents play, you have your point. > > So far, you haven't been able to. You have only been able to point > out that often good moves lead to good results (=wins), not that > they give a better prediction of future wins or losses than the > Elo rating. > No I've actually demonstrated clear cut examples where move-based evaluation is superior to result-based rating (which you conveniently avoid responding to) That demonstrates the potential of the concept and *should* shift the argument to practical matters - what are the limits, does the example extend to general cases, etc. Unfortunately the nonsensical arguments of principle (e.g."people-play-to-win") haven't gone away. I do regret stating my opinion that I think examining move history could do far better than examining result history, because it is really irrelevant to my interest in the concept. A move-based rating system, even if *inferior* to a result rating system, would still be extremely useful for the 99+% of the world's chess players who don't have result-based ratings.
|
| | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 14:50:34
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > No I've actually demonstrated clear cut examples where > move-based evaluation is superior to result-based rating You have _stated_ it, not proven. > I do regret stating my opinion that I think examining move > history could do far better than examining result history, > because it is really irrelevant to my interest in the concept. I have no problem whith that, save for the fact that it was the object of our discussion: is move based rating better than result based rating? No wonder we are discussing that, and no wonder my discarding that, and not something else. If you think it's irrelevant, the whole discussion has been void. > A move-based rating system, even if *inferior* to a result > rating system, would still be extremely useful for the 99+% > of the world's chess players who don't have result-based > ratings. Those systems are widely available, and on a sidenote, it remains to be seen if they actually do find it extremely usefull. Buy any copy of ChessMaster or Fritz and let them rate your play. It's a system which is also inferior to a real result based system like Elo, but gives them a result they can live with, but never will. No matter the dozens of messages popping up like "my rating in ChessMaster/Fritz is..... My rating at FICS is...... My rating at ICC is.... - now what do you think my _Elo/USCF_ rating is?" -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 09:31:33
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > > > No I've actually demonstrated clear cut examples where > > move-based evaluation is superior to result-based rating > > You have _stated_ it, not proven. > A plays B, C plays D. The goal is to handicap A vs C and B vs. D. Knowing the results has zero value. Knowing the moves has non-zero value. The rest is just details.
|
| | | | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 21:46:10
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > A plays B, C plays D. The goal is to handicap > A vs C and B vs. D. Knowing the results has zero value. > Knowing the moves has non-zero value. As I explained before, you don't prove that "knowing the moves" has non-zero value. You _assume_ it. If winning a game is a result of the better or the good moves - as you claim it to be - the result has by definition predictive value as well, just as the moves itself. Both players made better moves than the losers. As there are many moves leading to the win, some better, some less better than others, the win is by your own logic an indication of move strength, thus also no-zero value. Now what you have to do it prove that this non-zero value of move rating gives a better strength evaluation than the non-zero result based system. Apart from the more fundamental question if a one game calculation gives any sensible outcome at all. > The rest is > just details. Basically, those details are at the core of what you claim. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 18:31:45
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > > > A plays B, C plays D. The goal is to handicap > > A vs C and B vs. D. Knowing the results has zero value. > > Knowing the moves has non-zero value. > > As I explained before, you don't prove that "knowing the moves" has > non-zero value. You _assume_ it. So you're back to the extreme claim that knowing the moves has zero predictive value? > If winning a game is a result of the better or the good moves - as > you claim it to be - the result has by definition predictive value > as well, just as the moves itself. Wrong. The result of A vs. B and C vs. D don't tell me *anything* by themselves about A vs. C. > Both players made better moves than the losers. As there are many > moves leading to the win, some better, some less better than others, > the win is by your own logic an indication of move strength, thus > also no-zero value. > > Now what you have to do it prove that this non-zero value of move > rating gives a better strength evaluation than the non-zero result > based system. No. This is a clear cut case of moves being better than results.
|
| | | | | | | | | | | | | | | | | | | | | | | | | |
Date: 09 Apr 2005 00:35:46
From: CeeBee
Subject: Re: Rating from moves??
|
"David Kane" <[email protected] > wrote in rec.games.chess.analysis: > So you're back to the extreme claim that knowing the > moves has zero predictive value? I can't go back to a claim I never made, neither normal, nor extreme. We discussed the question "is a move based rating system better than a result based system". > No. This is a clear cut case of moves being better than results. So you basically don't have to prove it because it is a clear cut case. I see. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | | | | | | | | | | | |
Date: 08 Apr 2005 20:21:48
From: David Kane
Subject: Re: Rating from moves??
|
"CeeBee" <[email protected] > wrote in message news:[email protected]... > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > > > So you're back to the extreme claim that knowing the > > moves has zero predictive value? > > I can't go back to a claim I never made, neither normal, nor > extreme. > > We discussed the question "is a move based rating system better than > a result based system". > > > No. This is a clear cut case of moves being better than results. > > So you basically don't have to prove it because it is a clear cut > case. > > I see. > What is your answer? In the scenario I gave, who will do a better job of handicapping the next games, the person with the moves of the earlier games or the person with the results of the games? Your extreme defense of Elo rating seems to stem from a misunderstanding of what result-based rating is all about.
|
| | | | | | | | | | | | | | | | | | | | | | | | |
Date: 08 Apr 2005 00:13:54
From: Bark!
Subject: Re: Rating from moves??
|
Na njuzima:[email protected], CeeBee <[email protected] > mukotrpnim radom izna�e ovu umotvorinu: > "David Kane" <[email protected]> wrote in > rec.games.chess.analysis: > > >> A plays B, C plays D. The goal is to handicap >> A vs C and B vs. D. Knowing the results has zero value. >> Knowing the moves has non-zero value. > > As I explained before, you don't prove that "knowing the moves" has > non-zero value. You _assume_ it. Just a sec. You're saying that moves don't have any value? As you say non-zero. Just the result? Why the hell do we play chess then anyway??? If we play each other, how can someone kibitzing state sth like - CeeBee is making better moves - if those moves don't have any value? How can CM rate my play then? Based on what results? > cut
|
| | | | | | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 22:39:30
From: CeeBee
Subject: Re: Rating from moves??
|
"Bark!" <[email protected] > wrote in rec.games.chess.analysis: > Just a sec. You're saying that moves don't have any value? No. > As > you say non-zero. Non-zero= not equal to zero, i.e. bigger than zero. -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 16:53:32
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > A plays B, C plays D. The goal is to handicap > A vs C and B vs. D. Knowing the results has zero value. > Knowing the moves has non-zero value. The rest is > just details. I'm trying to make a map. The earth is covered with mountains, fields, cities, lands, and water. The rest is just details. -Ron
|
| | | | | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 10:16:11
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-8B880C.09533107042005@newssvr14-ext.news.prodigy.com ... > In article <[email protected]>, > "David Kane" <[email protected]> wrote: > > > A plays B, C plays D. The goal is to handicap > > A vs C and B vs. D. Knowing the results has zero value. > > Knowing the moves has non-zero value. The rest is > > just details. > > I'm trying to make a map. The earth is covered with mountains, fields, > cities, lands, and water. > > The rest is just details. > I alert you to the existence of satellite images and GPS devices that might be useful in completing the task. You respond that those sound awfully complicated and that a chain, magnetic compass, transit and level will work just fine.
|
| | | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 14:57:44
From: CeeBee
Subject: Re: Rating from moves??
|
CeeBee <[email protected] > wrote in rec.games.chess.analysis: > No ...wonder... > the dozens of messages popping up like "my rating in > ChessMaster/Fritz is..... My rating at FICS is...... My rating > at ICC is.... - now what do you think my _Elo/USCF_ rating is?" -- CeeBee "Eppur si muove!"
|
| | | | | | | | | | | | | | | | | | | | | |
Date: 07 Apr 2005 09:31:52
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > A move-based rating system, even if *inferior* to a result > rating system, would still be extremely useful for the 99+% > of the world's chess players who don't have result-based > ratings. I agree that such a system would be useful and interesting. I just don't think it's possible to any meaningful degree of accuracy except under conditions where results-based rating would be significantly more accurate. -Ron
|
| | | | | | | | | | |
Date: 06 Apr 2005 11:51:12
From: David Richerby
Subject: Re: Rating from moves??
|
mrlount <[email protected] > wrote: > Mrs Smith and Mr Brown pop to their local community centre to watch a > chess tournament on the weekend. > > They stand behind board number 12 which has John rated 2200 playing > white and Dave 1750 playing black. > > Mr Smith says to Mr Brown "Watch John play. I've seen him play before, > he has an amazing tactical brain." Dave puts his finger across his lips and says, ``Shh.'' > Mr Brown "Ah, so I guess he's got this in the bag, especially since he's > got the higher rating" Dave glares at Mr Smith and Mr Brown. ``Shhhhhhhhhh!!!'' > Mr Smith "No, Dave wins this tournament every year. John has never won a > single chess game, he plays 60 first class moves which Fritz can not > better. He then loses a drawn endgame by a blatant blunder." The tournament controller walks over and removes Mr Smith and Mr Brown from the playing area. :-) Dave. -- David Richerby Pickled Soap (TM): it's like a www.chiark.greenend.org.uk/~davidr/ personal hygiene product but it's preserved in vinegar!
|
| | | | |
Date: 05 Apr 2005 16:59:08
From: Ron
Subject: Re: Rating from moves??
|
In article <[email protected] >, "David Kane" <[email protected] > wrote: > If you calculated a result-based rating by selecting only Anand's > losses and only your wins, you would also reach an incorrect > conclusion > about your playing strengths. The flaw is in the selection of data, > not > some fundamental shortcoming in looking at the moves. I have no doubt > that looking at Anand's moves in his last few games and my moves in > my last few games would easily resolve who the better player is. But in the original posters question we're dealing with a very small selection of games. I don't disagree that I was cherry-picking data, but when given only a single game from your opponent, you can't assume that it's a typical game. If we had more games between these two opponents, we could calculate Elo based on performance. -Ron
|
| | | | | |
Date: 05 Apr 2005 10:08:08
From: David Kane
Subject: Re: Rating from moves??
|
"Ron" <[email protected] > wrote in message news:ronaldinho_m-37BC67.09590605042005@newssvr13-ext.news.prodigy.com ... > > If we had more games between these two opponents, we could calculate Elo > based on performance. > Or do even better rating the moves.
|
| |
Date: 04 Apr 2005 12:03:47
From: Peter Billam
Subject: Re: Rating from moves??
|
Bark! wrote: > Is it posible to somehow get unrated player's ELO from the moves he plays? > To be more concrete, I'd like to get my opponent's rating based on game(s) > that we played against each other so I can review my play against specific > sort of opponents (~1200, ~1600.. etc.) In <[email protected] >, Claus-J�rgen Heigl wrote: > You can also calculate your performance rating given > the results of your games. See here how this is done: > http://chess.about.com/od/organizations/ss/aa04l04_7.htm Calculating the rating from the moves is a good question, though. You'd need something like a strong chess-engine that could also play at levels adjustable by rating. Then using the strongest level you could go through the game measuring how sub-optimal the player's moves were. Then at different ratings, you'd repeatedly get the engine to replay the same positions, then evaluate the resulting moves and adjust the rating and repeat until you got a similar level of sub-optimality. (This would only work if the chess-engine's strongest level is far stronger than the player is.) It's possible for humans to guess an opponent's rating after a few games, so it should be possible for a computer to do it ... Regards, Peter -- Peter Billam, DPIWE/ILS/CIT/Servers, hbt/lnd/l8, 6233 3061
|
| | |
Date: 05 Apr 2005 19:40:43
From: David Richerby
Subject: Re: Rating from moves??
|
Peter Billam <[email protected] > wrote: > You'd need something like a strong chess-engine that could also play > at levels adjustable by rating. Then using the strongest level you > could go through the game measuring how sub-optimal the player's moves > were. Then at different ratings, you'd repeatedly get the engine to > replay the same positions, then evaluate the resulting moves and adjust > the rating and repeat until you got a similar level of sub-optimality. I can see a number of problems with this proposal. o It doesn't measure how good a player is but, rather, how much like the chosen engine a player is. o It's incredibly computationally expensive -- you need to be able to deal with thousands of games a week and you need to be able to somehow convert everone's current result-based rating into a move-based rating. o Engines don't understand closed positions very well so would claim that strong players were making bad moves in those cases. In addition, I'm not convinced that a move-based rating system would be sound. Should it really affect somebody's rating if it takes them fifteen moves to win a K+Q vs K endgame when the quickest win was in eight? Should somebody really be rewarded for finding the defence that leads to mate in ten rather than mate in five? Shouldn't they just have resigned anyway? > It's possible for humans to guess an opponent's rating after a > few games, so it should be possible for a computer to do it ... It's possible for humans to play go, write poetry, compose music, ... Dave. -- David Richerby Moistened Poetic Tree (TM): it's like www.chiark.greenend.org.uk/~davidr/ a tree but it's in verse and moist!
|
|