Page 1 of 9 123 ... LastLast
Results 1 to 15 of 129
  1. #1
    CC Candidate Master
    Join Date
    Jan 2010
    Location
    Meadow Springs wa
    Posts
    234
    Quote Originally Posted by Bill Gletsos
    Your April 1996 rating was 1397. You received a 100 increase as part of the April 2000 uplift.
    How does that help historical consistency?

    I was talking to to Bill Jordan a couple of years ago, discussing players ratings way back then, and now, and how players ratings' were higher now than formerly. It would have been well explained by these adjustments, but Bill didn't mention it, and I was completely unaware of it.
    John Summerfield

  2. #2
    Illuminati Bill Gletsos's Avatar
    Join Date
    Jan 2004
    Location
    Sydney
    Posts
    17,058
    Quote Originally Posted by Santa
    How does that help historical consistency?
    There was no historical consistency to start with.

    BTW what happened to the Compaq computer?
    The Force can have a strong influence on the weak-minded.
    Mos Eisley spaceport The toolbox. You will never find a more wretched hive of scum and villainy.

  3. #3
    CC Candidate Master
    Join Date
    Jan 2010
    Location
    Meadow Springs wa
    Posts
    234
    Quote Originally Posted by Bill Gletsos
    There was no historical consistency to start with.
    The question is how your actions help historical consistency.
    John Summerfield

  4. #4
    Illuminati Bill Gletsos's Avatar
    Join Date
    Jan 2004
    Location
    Sydney
    Posts
    17,058
    Quote Originally Posted by Santa
    The question is how your actions help historical consistency.
    My answer remains the same.
    There was no historical consistency to start with.

    BTW what happened to the Compaq computer?
    The Force can have a strong influence on the weak-minded.
    Mos Eisley spaceport The toolbox. You will never find a more wretched hive of scum and villainy.

  5. #5
    Monster of the deep Kevin Bonham's Avatar
    Join Date
    Jan 2004
    Posts
    40,360
    What is the point of historical consistency if the previous ratings had (i) become seriously flawed as the result of use of insufficiently dynamic systems and other problems and (ii) significantly lost parity with FIDE ratings at the pointy end?

    Why strive to be consistent with suboptimal figures?
    Moderation Requests: All requests for, comments about, or questions about moderation of any kind including thread changes must be posted in the Help and Feedback section and not on the thread in question. (Or by private message for routine changes or sensitive matters.)

    ACF Newsletter Information - All Australian players and administrators should subscribe and check each issue for relevant notices

    My psephology/politics site (token chess references only) : http://kevinbonham.blogspot.com.au/ Politics twitter feed https://twitter.com/kevinbonham

  6. #6
    CC Candidate Master
    Join Date
    Jan 2010
    Location
    Meadow Springs wa
    Posts
    234
    Quote Originally Posted by Kevin Bonham
    What is the point of historical consistency if the previous ratings had (i) become seriously flawed as the result of use of insufficiently dynamic systems and other problems and (ii) significantly lost parity with FIDE ratings at the pointy end?

    Why strive to be consistent with suboptimal figures?

    I made some points about the earlier system some years ago, and proposed a new system which was ready for testing when the ACF decided it was happy with the present system.

    Whatever was wrong with the old system (and I'd like your reasoned views on that), it does not justify making random changes to current ratings, changes that are not related to players' performance.

    If the current maths is working well, why are regular changes needed?

    Why is parity with FIDE important?
    John Summerfield

  7. #7
    Monster of the deep Kevin Bonham's Avatar
    Join Date
    Jan 2004
    Posts
    40,360
    Quote Originally Posted by Santa
    Whatever was wrong with the old system (and I'd like your reasoned views on that), it does not justify making random changes to current ratings, changes that are not related to players' performance.
    What "random changes" supposedly unrelated to player performance did you have in mind?

    If the current maths is working well, why are regular changes needed?
    They aren't. Since the Glicko system was started there have been two major changes, the 150 point uplift which was to correct problems caused by the prior systems and one 70 point uplift essentially for FIDE parity.

    There have been a number of minor improvements but many of these are along the lines of taking a good thing and making it work even better (for instance re-runs of past ratings with more complex formulae to implement intermediate ratings). Yet for some reason there is this perverse line of argument that says that if you take a good system and fine tune it to make it even better, then that means it must be suspect compared to a hypothetical inferior system that just sits there being inferior. I don't get this.

    A crucial difference between ratings now and 20+ years ago is that a ratings system can now make far more complicated use of the capacities of computers. It is no surprise that better and more complex ways of modelling and predicting player performance have come about, just as they have in many other similar fields of endeavour. I sometimes work in population modelling for threatened species, for example. You can do stuff now you just couldn't do 30+ years ago because of the advances in computing.

    Why is parity with FIDE important?
    It's no big deal, but "historical consistency" with previous more flawed systems is even less of a bigger deal. As for the case in favour, it allows for the use of FIDE ratings of players arriving from overseas to seed their ACF rating and also allows for easy seeding of such players in tournaments. True, the ACF could instead decree official adjustements but anyone who is concerned about historical consistency can make those themselves to past ACF ratings by adding whatever uplifts they consider represent a genuine adjustment to the base rather than just correction for deflation.

    As for the previous system my major issues with it were insufficiently dynamic response to two classes of players: improving players (esp juniors) and inactive players who play rarely. At times fixes such as bonus points were tried for the first but because the model was simplistic it was prone to overshoot.
    Moderation Requests: All requests for, comments about, or questions about moderation of any kind including thread changes must be posted in the Help and Feedback section and not on the thread in question. (Or by private message for routine changes or sensitive matters.)

    ACF Newsletter Information - All Australian players and administrators should subscribe and check each issue for relevant notices

    My psephology/politics site (token chess references only) : http://kevinbonham.blogspot.com.au/ Politics twitter feed https://twitter.com/kevinbonham

  8. #8
    CC Candidate Master
    Join Date
    Jan 2010
    Location
    Meadow Springs wa
    Posts
    234
    Quote Originally Posted by Bill Gletsos
    My answer remains the same.
    There was no historical consistency to start with.
    A key objective of Elo's recommendations was for ratings to be consistent over time, and he tested his system with data spanning 125 years and found, with the FIDE implementation, they were. See 2.8 "The test of time" in The rating of Chessplayers Past and Present.

    Whatever the faults of the previous system, it wasn't given to enormous jumps overnight, and whatever its faults, the new system should have addressed them.

    The question hasn't been addressed, but needs to be.
    John Summerfield

  9. #9
    Banned Hydra
    Join Date
    Jan 2010
    Posts
    7
    Quote Originally Posted by Santa
    A key objective of Elo's recommendations was for ratings to be consistent over time, and he tested his system with data spanning 125 years and found, with the FIDE implementation, they were. See 2.8 "The test of time" in The rating of Chessplayers Past and Present.

    Whatever the faults of the previous system, it wasn't given to enormous jumps overnight, and whatever its faults, the new system should have addressed them.

    The question hasn't been addressed, but needs to be.
    But what happened to the Compaq computer?

  10. #10
    CC Candidate Master
    Join Date
    Jan 2010
    Location
    Meadow Springs wa
    Posts
    234
    Quote Originally Posted by Kevin Bonham
    What "random changes" supposedly unrelated to player performance did you have in mind?



    They aren't. Since the Glicko system was started there have been two major changes, the 150 point uplift which was to correct problems caused by the prior systems and one 70 point uplift essentially for FIDE parity.
    The new ratings system should be capable of handling those itself given some time, and doing so smoothly.

    Elo discusses rating a pool comprising only unrated players (3.3), a suggestion is to enter them all into the system with the same ratings and large K (up to 50). After 30 games, he says, the players' relative ratings will be consistent.

    I might do that and calculate their performance ratings, then start again using their performance ratings as their initial ratings.

    There have been a number of minor improvements but many of these are along the lines of taking a good thing and making it work even better (for instance re-runs of past ratings with more complex formulae to implement intermediate ratings). Yet for some reason there is this perverse line of argument that says that if you take a good system and fine tune it to make it even better, then that means it must be suspect compared to a hypothetical inferior system that just sits there being inferior. I don't get this.
    I'm not the only person asking about how the new system works. In Shaun Press's time, the calculations and formulae were published in the ratings book. When I made changes, those were documented.

    Now, all I've been able to find is that the Ratings Officers were asked to develop a new system like, but not exactly the same, the Glicko2 system.

    I've been reading old tournament results in old magazines. They publish estimated ratings changes attributable to that tournament. I can see that Geoff Saw's performance rating in his first tournament was 1525 (obviously I did something right! and another Waverley junior performed at 1076 - better than I did!. That tournament (unofficially, of course) cost me 8 points.

    A crucial difference between ratings now and 20+ years ago is that a ratings system can now make far more complicated use of the capacities of computers. It is no surprise that better and more complex ways of modelling and predicting player performance have come about, just as they have in many other similar fields of endeavour. I sometimes work in population modelling for threatened species, for example. You can do stuff now you just couldn't do 30+ years ago because of the advances in computing.
    I've been using computers for over 40 years, I well understand the changes that have occurred in the time frames you talk about. The ACF has enjoyed the use of computers for 20 years or so.

    If the ACF published the formulae and algorithms used to calculate ratings, then others could implement them themselves, cast critical eyes over the results and, maybe, question discrepancies and suggest improvements.
    It's no big deal, but "historical consistency" with previous more flawed systems is even less of a bigger deal. As for the case in favour, it allows for the use of FIDE ratings of players arriving from overseas to seed their ACF rating and also allows for easy seeding of such players in tournaments. True, the ACF could instead decree official adjustements but anyone who is concerned about historical consistency can make those themselves to past ACF ratings by adding whatever uplifts they consider represent a genuine adjustment to the base rather than just correction for deflation.
    For a very long time, the ACF has entered foreign FIDE-rated players into Australian tournaments with ACF=FIDE.

    There has never been any need to do this, it would be permissible to make a more accurate conversion. If the evidence is there that FIDE ratings are 200 points above Australian ratings, the transformation is easy: enter a 2300 player at 2100. Or, rank players with FIDE ratings and assume Australians without FIDE ratings have FIDE=ACF+200.

    For ratings calculations, those players should be handled as unrated players, and their ACF-equivalent ratings not stored.

    As for the previous system my major issues with it were insufficiently dynamic response to two classes of players: improving players (esp juniors) and inactive players who play rarely. At times fixes such as bonus points were tried for the first but because the model was simplistic it was prone to overshoot.
    The reason that it was prone to overshoot was easily fixed. Ratings changes for a period were calculated with players' ratings unchanging for the period. if my rating was 1300 on a ratings list, all my ratings changes (and those of all my opponents) were calculated with that rating.

    Instead, tournaments should have been ordered by date (I suggest the date of the middle round to cater better for tournaments running for some weeks), and ratings updated at the end of each tournament. Then, use the end-of-tournament rating for the players' next event.

    If I earned a rating increase for my first tournament of a period of 100 points, my next tournament would be rated with my rating set to 1400, not 1300.

    That would pretty much kill the overshoot. However, the ACF would not agree to that change.

    Elo has K pretty much stable for a player, but Elo was principally targetting established, strong players. Really, there is no reason it should not be adjusted from time to time for any number of reasons such as
    A prolonged absence from the game
    Recent results inconsistent with a long-established rating.
    Illness affecting a players' skill (a stroke maybe) or ability to play (impending blindness?).

    There's no real need to keep players ratings for ever, I'd be happy to drop players ratings three (but you choose a number!) years after their last game. Entering them again, if they return, should not be a big deal, and there's no good reason to suppose their old rating has any meaning.
    John Summerfield

  11. #11
    Illuminati Bill Gletsos's Avatar
    Join Date
    Jan 2004
    Location
    Sydney
    Posts
    17,058
    Quote Originally Posted by Santa
    I made some points about the earlier system some years ago, and proposed a new system which was ready for testing when the ACF decided it was happy with the present system.
    Yes, you were the one who deleted over 8,000 names from the ACF master file in 1993 without ACF Council approval.
    You continually refused to re-add them.
    As such you were replaced as ACF ratings Officer in mid 1996.

    BTW what happened to the Compaq computer?
    The Force can have a strong influence on the weak-minded.
    Mos Eisley spaceport The toolbox. You will never find a more wretched hive of scum and villainy.

  12. #12
    Illuminati Bill Gletsos's Avatar
    Join Date
    Jan 2004
    Location
    Sydney
    Posts
    17,058
    Quote Originally Posted by Santa
    A key objective of Elo's recommendations was for ratings to be consistent over time, and he tested his system with data spanning 125 years and found, with the FIDE implementation, they were. See 2.8 "The test of time" in The rating of Chessplayers Past and Present.
    The point is he may have said that but it has not been the case.
    most people who have studied the FIDE ratings believes that the FIDE Elo system has been inflating virtually every year.

    Quote Originally Posted by Santa
    Whatever the faults of the previous system, it wasn't given to enormous jumps overnight, and whatever its faults, the new system should have addressed them.
    As usual you dont understand. This has nothing to do with Glicko Vs Elo.

    The problem was with the ACF Elo system of the 1980's & 1990's.
    Because of the 336 rule players at the top of the rating list continued to gain points they did not deserve and players at the bottom of the list continued to lose points that they should not have lost. Also the bonus point scheme in place generally caused massive overshoots in one list that lead to loss of points by those players in the next list.
    All the while the ratings of players were deflating compared to their FIDE ratings.

    The ACF Council decided that it would apply an uplift to the December 1999 ACF ratings to try and combat this long standing deflation of the ACF ratings and inflation of the FIDE ratings.

    BTW what happened to the Compaq computer?
    The Force can have a strong influence on the weak-minded.
    Mos Eisley spaceport The toolbox. You will never find a more wretched hive of scum and villainy.

  13. #13
    Illuminati Bill Gletsos's Avatar
    Join Date
    Jan 2004
    Location
    Sydney
    Posts
    17,058
    Quote Originally Posted by NorthsCC
    But what happened to the Compaq computer?
    Yes he seems to be deliberately ignoring this question.
    The Force can have a strong influence on the weak-minded.
    Mos Eisley spaceport The toolbox. You will never find a more wretched hive of scum and villainy.

  14. #14
    Illuminati Bill Gletsos's Avatar
    Join Date
    Jan 2004
    Location
    Sydney
    Posts
    17,058
    Quote Originally Posted by Kevin Bonham
    Since the Glicko system was started there have been two major changes, the 150 point uplift which was to correct problems caused by the prior systems and one 70 point uplift essentially for FIDE parity.
    Actually the 150 point uplift occurred in April 2000 under the ACF Elo system. It had nothing to do with the Glicko system.
    The Force can have a strong influence on the weak-minded.
    Mos Eisley spaceport The toolbox. You will never find a more wretched hive of scum and villainy.

  15. #15
    Monster of the deep Kevin Bonham's Avatar
    Join Date
    Jan 2004
    Posts
    40,360
    Quote Originally Posted by Santa
    The new ratings system should be capable of handling those itself given some time, and doing so smoothly.
    How is any system supposed to fix a system-wide deflation caused by the failures of a previous system? (Bill notes correctly that the uplift was pre-Glicko; I had thought it came just after Glicko started, but in any case the point was that the old system created the problems that led to that uplift, not Glicko.)

    Elo discusses rating a pool comprising only unrated players (3.3), a suggestion is to enter them all into the system with the same ratings and large K (up to 50). After 30 games, he says, the players' relative ratings will be consistent.
    I don't see how this is relevant to anything whatsoever since it has no bearing on a system that consists largely of players with established ratings being continued from one system to the next.

    Furthermore even if the players' relative ratings are consistent with each other a closed system proves nothing. It is how a system responds to situations caused by players entering and leaving the pool at various strengths and trajectories that is a bigger test.

    I might do that and calculate their performance ratings, then start again using their performance ratings as their initial ratings.

    I'm not the only person asking about how the new system works. In Shaun Press's time, the calculations and formulae were published in the ratings book. When I made changes, those were documented.

    Now, all I've been able to find is that the Ratings Officers were asked to develop a new system like, but not exactly the same, the Glicko2 system.

    I've been reading old tournament results in old magazines. They publish estimated ratings changes attributable to that tournament. I can see that Geoff Saw's performance rating in his first tournament was 1525 (obviously I did something right! and another Waverley junior performed at 1076 - better than I did!. That tournament (unofficially, of course) cost me 8 points.
    I don't see how any of this is relevant to the quote from me it was written in reply to. You asked why regular changes are needed. I disposed of the argument that (supposedly) regular fine tuning demonstrates inferiority in a system (in fact it can often demonstrate that a system is well maintained). You then go talking about the difficulty in finding details of the system but that is not the same thing as evidence that the system itself is not working.

    If the ACF published the formulae and algorithms used to calculate ratings, then others could implement them themselves, cast critical eyes over the results and, maybe, question discrepancies and suggest improvements.
    People can do that anyway. You don't need to know the formulae used to calculate ratings to have a reasonable idea of whether those ratings are proving predictively accurate (or not) in practice.

    For ratings calculations, those players should be handled as unrated players, and their ACF-equivalent ratings not stored.
    That is simply throwing away information that whatever the flaws in FIDE ratings is nonetheless still predictively useful.

    There has never been any need to do this, it would be permissible to make a more accurate conversion.
    An adjustment for parity makes no difference to the accuracy of a conversion except that it removes the need for one.

    That would pretty much kill the overshoot. However, the ACF would not agree to that change.
    Ordering tournaments by middle date is not practical since a weekender late in a ratings period could get rated before a weekly tournament with an earlier middle date that extends into the next period, even though most games in the latter event would be earlier.

    Elo has K pretty much stable for a player, but Elo was principally targetting established, strong players. Really, there is no reason it should not be adjusted from time to time for any number of reasons such as
    A prolonged absence from the game
    Recent results inconsistent with a long-established rating.
    Illness affecting a players' skill (a stroke maybe) or ability to play (impending blindness?).
    What Glicko does is effectively adjust k based on the evidence and the time a player has had off the game so Glicko is several steps ahead of you there. Of course any system that required adjustment for illness would be a complete nightmare to police; ratings officers are ratings officers, not health monitors of their player base.

    There's no real need to keep players ratings for ever, I'd be happy to drop players ratings three (but you choose a number!) years after their last game. Entering them again, if they return, should not be a big deal, and there's no good reason to suppose their old rating has any meaning.
    You seem to like throwing away information that may still be quite predictively useful. Keeping the old rating but allowing its influence compared to new results to wane over time is what Glicko does.

    The only problem we have at the moment is some very inactive players who know they are overrated as hell and don't want to lose ratings points, trying to pretend their ancient rating still means anything.
    Moderation Requests: All requests for, comments about, or questions about moderation of any kind including thread changes must be posted in the Help and Feedback section and not on the thread in question. (Or by private message for routine changes or sensitive matters.)

    ACF Newsletter Information - All Australian players and administrators should subscribe and check each issue for relevant notices

    My psephology/politics site (token chess references only) : http://kevinbonham.blogspot.com.au/ Politics twitter feed https://twitter.com/kevinbonham

Thread Information

Users Browsing this Thread

There are currently 1 users browsing this thread. (0 members and 1 guests)

Similar Threads

  1. Good books for social players
    By Kevin Bonham in forum Coaching Clinic
    Replies: 24
    Last Post: 08-11-2011, 05:57 PM
  2. Siberian Chess Tiger Days Revisited [fiction]
    By Axiom in forum General Chess Chat
    Replies: 24
    Last Post: 15-07-2009, 01:30 AM
  3. Sponsorship withdrawal and coverage sf. 2009 Aus Open
    By peter_parr in forum Australian Chess
    Replies: 83
    Last Post: 10-01-2009, 04:02 PM
  4. Box Hill bulletins
    By Davidflude in forum Australian Chess
    Replies: 0
    Last Post: 08-07-2008, 07:06 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •