jriechel Nov 9 1992, 4:33 pm show options Newsgroups: rec.games.chess From: jriec...@Bonnie.ICS.UCI.EDU - Find messages by this author Date: 10 Nov 92 06:19:31 GMT Local: Mon, Nov 9 1992 10:19 pm Subject: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse Why not start newcomers ratings to zero, and reset all currently rated players, also, to a rating of zero? If your rating was ever exactly zero, then the system could be programmed to display =PAR= instead of 0. With this system, there would be no pretension to approximating the USCF and FIDE systems. Of course, it would still use Elo's equations, but it would LOOK DIFFERENT. Which is usually enough to convince people it's unique. James Riechel E. Mark Ping Nov 10 1992, 9:31 am show options Newsgroups: rec.games.chess From: ema...@soda.berkeley.edu (E. Mark Ping) - Find messages by this author Date: 10 Nov 1992 21:55:00 GMT Subject: Re: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article <9211092219.aa02...@Bonnie.ics.uci.edu> jriec...@Bonnie.ICS.UCI.EDU writes: } }Why not start newcomers ratings to zero, and reset all currently rated }players, also, to a rating of zero? If your rating was ever exactly }zero, then the system could be programmed to display =PAR= instead of 0. } }With this system, there would be no pretension to approximating the USCF }and FIDE systems. Of course, it would still use Elo's equations, but it }would LOOK DIFFERENT. Which is usually enough to convince people it's }unique. } }James Riechel This has the same problem as the current system. All of the ratings will gravitate towards 0 instead of 1000 or 1500, but the problem remains: in the current system, the differences between rated players is decreasing as everyone moves toward the averages. This problem would probably be solved by having the provisional player's score not affect the opponents rating until around 5-10 games are played. That's not perfect either, but there are no tournaments on the ICS (except for blitz), so there's no easy way to use the provisional player's post-tournament rating to calculate into the other players rating. ------------ E. Mark Ping ema...@soda.Berkeley.EDU "Stand aside, I take large steps." --Michael Dorn "Say, that's a nice bike." --Cyberdyne Systems T-1000 "Pituita es." --Unknown Latin Scholar Shirish Chinchalkar Nov 12 1992, 10:14 am show options Newsgroups: rec.games.chess From: chi...@cs.cornell.edu (Shirish Chinchalkar) - Find messages by this author Date: Thu, 12 Nov 1992 22:31:57 GMT Local: Thurs, Nov 12 1992 2:31 pm Subject: Re: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse The reason for deflation of ICS ratings goes beyond introduction of new players into the system. IN THE ABSENSE OF NEW PLAYERS, the ICS ratings will STILL drop to 0. I have mailed the following to the ICS admins, but have not received any response yet. I would like to know if my reasoning is correct or not. The bug is in not using a temporary variable when doing integer operations in the file ratings.c in the IRS code. The reason this drop was not so prominent earlier is that there was a ratings floor and because very few games were played before the FS-2 boom. Shirish. ---------------------------------------------------------------------- It is often the case on ICS that when A beats B, A's rating increases by x points and B's rating drops by x+1 points. This reduces the total number of points in the pool by 1. At the rate of 1000 games per day and 1000 registered players, the average rating is expected to drop at the rate of about 1 per day, which is indeed the case. Example: matt (1540) paar (1381) White valkyries.andrew.cmu.edu Wed Nov 11 09:55 :26 1992 Statistics for matt: Standard : 1549 Blitz : 1391 Statistics for paar: Standard : 1371 Blitz : 1161 So matt's rating increased by 9 points and paar's rating dropped by 10 points. Consider the following fragment of irs code: deltaR = bOrigRating - wOrigRating; expectedWins = 1.0 / (pow(10.0, (deltaR / 400.0)) + 1.0); *wr = wOrigRating + (K * (wWin - expectedWins)); deltaR = wOrigRating - bOrigRating; expectedWins = 1.0 / (pow(10.0, (deltaR / 400.0)) + 1.0); *br = bOrigRating + (K * (bWin - expectedWins)); "wWin - expectedWins" is equal in magnitude to "bWin - expectedWin". Since ratings are integral quantities, for the above example, you get: *wr = 1540 + 32*(1 - 0.714) = 1540 + 9.149 *br = 1381 + 32*(0 - 0.286) = 1381 - 9.149 Now, in integral arithmetic, 1540 + 9.149 = 1540 + 9, but 1381 - 9.149 = 1341 - 10. This problem can be eliminated by truncating *first* before updating the ratings: int t; deltaR = bOrigRating - wOrigRating; expectedWins = 1.0 / (pow(10.0, (deltaR / 400.0)) + 1.0); t = K * (wWin - expectedWins); *wr = wOrigRating + t; deltaR = wOrigRating - bOrigRating; expectedWins = 1.0 / (pow(10.0, (deltaR / 400.0)) + 1.0); t = K * (bWin - expectedWins); *br = bOrigRating + t; Michael Schmahl [Black-Robe Mage] Nov 12 1992, 1:31 pm show options Newsgroups: rec.games.chess From: f...@camelot.acf-lab.alaska.edu (Michael Schmahl [Black-Robe Mage]) - Find messages by this author Date: 12 Nov 92 20:39:35 GMT Local: Thurs, Nov 12 1992 12:39 pm Subject: Re: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse jriec...@Bonnie.ICS.UCI.EDU writes > Why not start newcomers ratings to zero, and reset all currently rated > players, also, to a rating of zero? If your rating was ever exactly > zero, then the system could be programmed to display =PAR= instead of 0. > With this system, there would be no pretension to approximating the USCF > and FIDE systems. Of course, it would still use Elo's equations, but it > would LOOK DIFFERENT. Which is usually enough to convince people it's > unique. > James Riechel but perhaps it's not a good idea for the ICS to be different from the USCF/FIDE ratings. I think ICS has a good potential for estimating the strength of chess programs, for example. Another example is the chessplayer who does not play OTB regularly. Besides using ICS to keep his chess sharp, he can also use it to gauge how he is doing in general. But this is not useful if ICS and USCF are not even similar. It's even worse if ICS looks like USCF but is 300 or so points off.... -- The secretary says he's on another line. Can I hold for a long, long time? -TMBG Michael Schmahl a.k.a. Black-Robe Mage Michael Schmahl [Black-Robe Mage] Nov 12 1992, 1:50 pm show options Newsgroups: rec.games.chess From: f...@camelot.acf-lab.alaska.edu (Michael Schmahl [Black-Robe Mage]) - Find messages by this author Date: 12 Nov 92 21:17:36 GMT Local: Thurs, Nov 12 1992 1:17 pm Subject: Re: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse E. Mark Ping writes - Hide quoted text - - Show quoted text - > In article <9211092219.aa02...@Bonnie.ics.uci.edu> jriec...@Bonnie.ICS.UCI.EDU writes: > } > }Why not start newcomers ratings to zero, and reset all currently rated > }players, also, to a rating of zero? If your rating was ever exactly > }zero, then the system could be programmed to display =PAR= instead of 0. > } > }With this system, there would be no pretension to approximating the USCF > }and FIDE systems. Of course, it would still use Elo's equations, but it > }would LOOK DIFFERENT. Which is usually enough to convince people it's > }unique. > } > }James Riechel > This has the same problem as the current system. All of the ratings will > gravitate towards 0 instead of 1000 or 1500, but the problem remains: in > the current system, the differences between rated players is decreasing as > everyone moves toward the averages. This problem would probably be solved > by having the provisional player's score not affect the opponents rating > until around 5-10 games are played. That's not perfect either, but there > are no tournaments on the ICS (except for blitz), so there's no easy way to > use the provisional player's post-tournament rating to calculate into the > other players rating. > ------------ > E. Mark Ping > ema...@soda.Berkeley.EDU > "Stand aside, I take large steps." --Michael Dorn > "Say, that's a nice bike." --Cyberdyne Systems T-1000 > "Pituita es." --Unknown Latin Scholar Perhaps some other form of normalization is desirable. Say under the assumption that the distibution of ratings is apporximately normal, take the average of everyone's ratings, subtract that from the ratings, divide by the standard deviation, multiply by the desired standard deviation (probably about 300) and add the desired average. The effect will be that people will still have the same distribution and be in the same place relative to the population, but you can get better approximations and don't have to worry about rating degradation. -- - Hide quoted text - - Show quoted text - The secretary says he's on another line. Can I hold for a long, long time? -TMBG Michael Schmahl a.k.a. Black-Robe Mage Kenneth Sloan Nov 12 1992, 6:54 pm show options Newsgroups: rec.games.chess From: s...@cis.uab.edu (Kenneth Sloan) - Find messages by this author Date: Thu, 12 Nov 1992 22:20:33 GMT Local: Thurs, Nov 12 1992 2:20 pm Subject: Re: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article <1992Nov12.203935.26...@raven.alaska.edu> f...@camelot.acf-lab.alaska.edu (Michael Schmahl [Black-Robe Mage]) writes: >... >but perhaps it's not a good idea for the ICS to be different from the >USCF/FIDE >ratings. I think ICS has a good potential for estimating the strength >of chess >programs, for example. Another example is the chessplayer who does not >play OTB >regularly. Besides using ICS to keep his chess sharp, he can also use >it to gauge >how he is doing in general. But this is not useful if ICS and USCF are >not even >similar. It's even worse if ICS looks like USCF but is 300 or so points off.... You can meet all of the goals above even if the ICS scale looks completely different from the USCF scale. If both ICS chess and USCF OTB chess are the same animal, and both scales are reasonably good, and the rating pools don't differ too much...then you should be able to translate from the ICS scale to the USCF scale at will. But...the advantage of making the two scales *look* different is that we won't see people waving about their ICS rating without qualification, pretending that it is their USCF (or FIDE) rating. This is one reason that I think it was a mistake to re-align the USCF Postal (oops - Correspondence) scale to rough alignment with the USCF OTB scale. I claim that its more confusing to do things that way. But, I've been wrong before... -- Kenneth Sloan Computer and Information Sciences s...@cis.uab.edu University of Alabama at Birmingham (205) 934-2213 115A Campbell Hall, UAB Station (205) 934-5473 FAX Birmingham, AL 35294-1170 Amarnath Mukherjee Nov 13 1992, 7:52 am show options Newsgroups: rec.games.chess From: amarn...@aura.cis.upenn.edu (Amarnath Mukherjee) - Find messages by this author Date: 12 Nov 92 23:04:29 GMT Local: Thurs, Nov 12 1992 3:04 pm Subject: Re: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse In article <1992Nov12.222033.2...@cis.uab.edu> s...@cis.uab.edu (Kenneth Sloan) writes: >But...the advantage of making the two scales *look* different is that we >won't see people waving about their ICS rating without qualification, >pretending that it is their USCF (or FIDE) rating. True..there's also another issue that I think is relevant: Many of us play there to have fun and the ratings are important only in so far as they can entice stronger players (than ourselves) to play with us. Also, most people play much more seriously (alsmost for blood) in tournaments...ICS is a friendly place to check out other players and have fun. For instance, I've never played an IM OTB, but one gets to play aries, ekw, scratchy etc. on ICS and learn from them. Also, we get to play jarmo, rkl and others from Europe sitting in the US and vice versa! I for one have no complaints! Thanks much gang for a wonderful job!! Amarnath Mukherjee (ICS- am) John Anthony Chanak Nov 13 1992, 11:56 am show options Newsgroups: rec.games.chess From: John Anthony Chanak - Find messages by this author Date: Fri, 13 Nov 1992 06:52:31 -0500 Local: Fri, Nov 13 1992 3:52 am Subject: Re: ICS Ratings -- Average Ratings -- Provisional Ratings Reply to Author | Forward | Print | Individual Message | Show original | Report Abuse Excerpts from netnews.rec.games.chess: 12-Nov-92 Re: ICS Ratings -- Average .. Shirish Chinchal...@cs.c (2575) > The reason for deflation of ICS ratings goes beyond introduction of new > players into the system. IN THE ABSENSE OF NEW PLAYERS, the ICS ratings > will STILL drop to 0. I have mailed the following to the ICS admins, but > have not received any response yet. I would like to know if my reasoning > is correct or not. This is a bug and has been fixed as of yesterday (Thursday). It is the first truly believable explanation I have yet heard; the bug is there; the numbers work out. > The bug is in not using a temporary variable when doing integer operations in the file ratings.c in the IRS code . There are lots of ways to squash bugs... > The reason this drop was not so prominent earlier is that there was a ratings > floor and because very few games were played before the FS-2 boom. > Shirish. > ---------------------------------------------------------------------- [explanation deleted] -John Chanak