AGA Rating System
Table of contents |
Overview
''The AGA Rating System'' provides a complete description of the AGA's current rating system. Other documents describing the system are: ''Inside the AGA Rating System'', an explanation of the AGA rating sytem by Paul Matthews, dated 28 July 1990, and a FAQ dated May 2002. The rating algorithm was originally developed by Paul Matthews, and was reimplemented by Philip Waldron in 2009-2010 after Matthews' departure from the AGA. Paul's updated tournament rating system is freely available at http://www.accelrat.com. Philip's algorithm is now open-source and licensed under the GPL; it can be downloaded from the AGA's ratings page
AGA ratings consist of a numerical value (RP or rating prior) and a sigma or uncertainty. With each game played the sigma decreases. Smaller values of sigma cause a player's rating to change more slowly.
Numerical format
The AGA system has no official ranks, only numerical ratings. Generally, positive AGA ratings represent dan ranks, and negative AGA ratings represent kyu ranks. Unofficially, in order to convert a rating to a rank, round it towards zero (not, as might be expected, to the nearest whole number). E.g., ratings in the range 5.0 to 5.9 represent 5 dan, and those in the range -5.9 to -5.0 represent 5 kyu. (There are no ratings between -1.0 and 1.0.) As an exception, any rating of 6.0 or higher represents 6 dan, the highest AGA rank. (Jie Li is 7 dan, but I assume this is a Chinese rank. His AGA rating is 9.1.)
Examples:
- -3.9 is a weak 3k
- -3.1 is a strong 3k
- 3.1 is a weak 3d
- 3.9 is a strong 3d.
- There are no ratings between -1.0 and 1.0.
Self Promotion
From http://www.usgo.org/resources/downloads/aga-rating.pdf
- The default value of rp_sigma seems sufficient for routine rating adjustments; however, a rapidly improving player may play at a rank several hundred points above his or her old rating, and a boost is needed. Players who declare a rank more than 50 points higher than their rating, have the mean and standard deviation parameters of their RP function increased. By adding points to the RP mean, points are added to the whole system, helping to counteract the tendency for the ratings of stable players to deflate as other players improve. The larger standard deviation allows an improving player's rating to float more freely, upward or downward, and have less effect on the ratings of opponents. Note that a player who performs poorly when playing above his or her rating risks a larger loss of rating points.
The key part about how the math works is "the mean and standard deviation parameters of their RP function increased." RP is your rating prior, or in other words your rating before the tournament begins. Increasing the mean of the RP is a fancy way of saying your rating prior is increased. Unfortunately it doesn't say how much it's increased by.
Self-promotions are inferred by AGA ratings algorithm when it reads the tournament report and notices the player's entry rating is higher than the player's current rating.
From http://www.usgo.org/ratings/RatingQuals.asp an example tournament report, with a comment added to show entry rating.
TOURNEY Princeton Go Club Rated Games, Princeton, NJ, June 15-30, 2004 start=6/15/2004 finish=6/30/2004 rules=AGA PLAYERS 489 Matthews, Paul 4k # <--- this is where the player's entry rating is set 3199 Mott, Rick 4k GAMES 489 3199 b 0 7 3199 489 w 0 7
Comments
aokun: I think the AGA system, consisting of numerical ratings only and with no ranks at all, is missing an opportunity. The rating math is fine, as far as I know, but I think the Association should award ranks as well.
The ratings go up and down and are a measure of current strength. That should stay the same. But when your rating gets to a certain level, your rank should too, and not go down. You should play at your rating or above, as now, with the rank as an indicator of achievement rather than current strength. The rank should come with a certificate, or ribbon, or medal or some other bauble.
The trouble with the ratings, aside from their inevitable descent with age, is that they are so flavorless and, until you are a shodan, negative. Most other martial arts and competitive games use evocative titles and tangible rewards to encourage members. Karate and Judo, like go Asian-origin martial arts, award colored belts and stamps in faux-leather booklets to mark promotion through ranks. Consistent with our psychology, this encourages effort. Archery in the US operates in a similar way; as you reach certain levels you are ranked, the ranks having names like "junior bowman."
The 12-year-old karate student, after two years of effort, can tell mom he is a brown belt and show her the belt and papers. The similarly positioned go student can tell his mom that he is now an AGA -2.71, but that is not a rank, only a short-term indication of playing strength. Likewise, the 35-year-old judo practioner who, as a 24-year-old, earned a black belt, remains deservedly a black belt as his midriff expands, not subject to having the belt confiscated just because he can't clobber the next generation.
Awards, ribbons and ranks are tangible forms of fun and recognition for go players. We should have them. At the very least, we should have something to reward folks who make shodan.
Jared: The closest thing we have is tournament prizes.
If I am remembering this correctly, there is a distinction between AGA ranks (kyu, dan) and AGA ratings (-1.0, 3.5). I think an analogy would be between Elo ratings in chess and the various classes (Master, International Master, etc). So in the AGA system, the highest rank is 7-dan, but the ratings have no cap. This probably explains why Jie Li is listed as a 7-dan, since his actual rating could very well be 9.x. KGS uses a similar system -- players like toxxicu are 7-dan but several levels higher than other 7-dan players.
-- Matt Noonan
SAS: Some more (and more certain) information would be useful. The following is my understanding of the AGA system, but I don't live in America and would appreciate some confirmation for all of this.
- But AGA doesn't set players ranks: in the United States, even dan players set their own ranks. (I would particularly appreciate confirmation of this point. Being used to the British system, I find it rather surprising.) However, to prevent sandbagging, players who have an AGA rating are not allowed to set their rank below the rank that their rating represents.
The AGA rating page is here: http://www.usgo.org/ratings/default.asp
The above is essentially correct. The AGA does not (at present) set or define "ranks", only "ratings". However, it is common practice to convert a rating to a rank as outlined above. Players may not enter a tournament at a rating less than their current official rating. Unrated players, and players that believe they are stronger than their official ratings enter using their best estimate of their current strength, with approval/guidance from the TD.
Joe Berry