![]() StartingPoints Paths Referenced by
|
rank - worldwide - discussion
Niklaus: Isn't the "unit" of the ranking system supposed to be handicap stones? So the only difference between the different scales should be where you set the bar for 30k (or 29k, since everybody below is 30k), and that difference should be the same on every level onwards? Most of the chart above is more or less consistent, but there are some areas (such as IGS double digit kyus and Japanese high dans), where the "unit" seems to be more or less than one handicap stone. That doesn't make sense to me, because it makes the rating system a lot less useful. Charles Matthews: All the evidence is that server grades are not one stone apart, but closer. Where they are not used for handicap go much or at all, that's an artefact of the rating algorithm. dnerra: If you look at the details of the rank computation on IGS, then, ignoring handicap games (which are truly negligible on IGS), the IGS definition of the "unit", i.e. the one stone difference is: The stronger player has a winning percentage of 72%. (See "help math" on IGS if you want to understand why.) Of course this assumption is pretty much nonsense. The winning percentage of a 15k against a 16k is a lot closer to 50%, whereas a 7D has probably a higher percentage playing against a 6D. It would be pretty easy to fix the system IGS by using empirical values, e.g. from the European rating database, instead. exswoo: The ranking system is purely relative since each system only has certain people from limited areas (both physical and virtual). So the ranking system works within that group because the stone differences do work out. What's happening is that the average strength of a player within each system is different, which gives rise to the major variances at each end of the spectrum. From the table above, it's clear that the average player in each system is in the single digit kyu's, with the exact level varying about 3 or so stones. Also, I'd like to question the Korean ranking table because while I can't vouch for the low end of the spectrum, I've played some 12 and 14 gups and they seem to be only about 2 or 3 stones stronger than the AGA rankings. Jan: In the Netherlands (at least at my Go club) there is another system in use, the 'klassen'-system. ('Klasse' means class). Roughly, a difference of two classes is equivalent to one stone. I believe the formula is: kyu level = (klasse - 20) / 2; For example, I'm currently about klasse 50, which corresponds to 15 kyu. A player of klasse 20 would be between 1kyu and 1dan. I'm not exactly sure where this information fits in, so I thought I'd mention it here. Maybe someone can fill in some more detail... Andre Engels: A player of klasse 20 would be 1 kyu - klasse 20 and 21 correspond to 1 kyu, 22 and 23 to 2 kyu, 18 and 19 to 1 dan, etcetera. Migeru: Why don't the deshis that have more than one rating (usually a region and an internet server) post their ranks here? That way we can get an idea of the equivalence if there is one. For what it's worth: I am NNGS 12k*, KGS 14k and IGS 19k... This doesn't really match the table on RankWorldwideComparison. Charles: You need a large sample (100?) to make good sense of this. Often players have several accounts, they play casually on servers (or seriously), their tournament ratings can be out-of-date, their club grades are unreliable and so on. There have been surveys before that didn't prove much. Tim Brent: I feel the on-line rankings should change less often, say on a monthly basis, as opposed to daily. It would be better, in my opinion, in that you would get a more accurate reading of your rank or strength on the server (or in a real-life club). One of my bad habits on IGS when I played there was going crazy every day to check my rating (the early stages of Rating Paranoia). HolIgor: I've heard that KGS declares the objective of keeping its rating close to AGA rating. Answer: Yes. Ian: As someone who is a low kyu and plays a lot on KGS I find I can win games more easily by taking the suggested number of handicap stones. Infact practice has shown this to be the best method (barring taking more than the recommended amount) Does this make me a sandbagger? Answer: No. amc: I have a question about ranking. I know that "theoretically" a difference in one rank is equal to one stone, right? But what does this mean exactly? Does it mean that two players of different ranks playing lots of games with the appropriate amount of stones should split the results 50/50? But if this is so, isn't this more or less easy to determine? What exactly causes the differences between the different ranks? (servers, regional association, etc.) Answer (Charles Matthews): Yes, it is assumed that for purposes of playing handicap games the correct rank difference is the number of stones to give 50-50 results, while one stone too many or too few will give more like 67-33. There is some bias for small numbers of stones in that a two-stone handicap is more like 1.5 stones in practice. The observed difficulties in cross-calibration of ranking systems will stem from a number of factors, such as: different pools of players worldwide not strongly interacting; lack on some servers of handicap games played; styles and depth of thought being different for quick server games and tournament play. W: What about differences in style? I do not think a linear rating system will ever be completely accurate. I am sure there are cases of players A, B, and C, where A can give B two stones and play even, B can give C two stones, and C can give A two stones, if the styles of play and focus of learning are different enough. How would you rate such players easily? Or do these situations not occur at higher ranks? This is a copy of the living page "rank - worldwide - discussion" at Sensei's Library. ![]() |