Life In 19x19

I have played go off and on for a couple years (more off than on). I am probably realistically like 25k on a good day. I played a couple rated games on KGS earlier in the year where my rank fluctuated between 27k and 29k. Today I tried to play a 26k bot and it told me I could not play that rated because it would be more than six stones handicap. I look at my user information and it shows "Rank: 2d?". If I look at the ratings graph, it shows somewhere between 29 and 30k (which would seem correct). What the heck is going on?

Alex

Known bug when you don't play a rated game for 6 months. Supposed to change you back to plain [?], but sometimes bug happens and you get [2d?] instead. Always [2d?], never some other rank.

Or maybe you're better than you think you are

Think of it as a good excuse to play a couple games with stronger players...

linuxdaemon wrote:I have played go off and on for a couple years (more off than on). I am probably realistically like 25k on a good day. I played a couple rated games on KGS earlier in the year where my rank fluctuated between 27k and 29k. Today I tried to play a 26k bot and it told me I could not play that rated because it would be more than six stones handicap. I look at my user information and it shows "Rank: 2d?". If I look at the ratings graph, it shows somewhere between 29 and 30k (which would seem correct). What the heck is going on?

Alex

KGS simply uses an algorithm that takes into account every game for a long time, weighted by time since you played, whether your opponent improves shortly afterwards etc. When the number of games is small, and it has been a long time since the last one, this algorithm doesn't work very well and gives strange results.

What you can't see on the displayed rank is the algorithm's recorded uncertainty. This will be very high, so playing just a single game will move it a great deal to near where it should be.

Edit: Maybe there is an actual bug, as yoyoma suggests, but I've definitely observed the phenomenon of weak players reaching non-2d? ranks after not playing for a long time.

Thanks for the replies. That make some sense, and I had a feeling it had something to do with the length of time I had not played anything. 2d didn't make any sense though .

@daniel_the_smith: Maybe playing in the 2d area will bring it out of me...but somehow I don't see that happening

When you play a lot, your rating does not move. When you don't play, hectic big rating jumps are, uh, called a feature instead of a bug, sigh. Vice versa it would make more sense.

RobertJasiek wrote:When you play a lot, your rating does not move. When you don't play, hectic big rating jumps are, uh, called a feature instead of a bug, sigh. Vice versa it would make more sense.

If you want your rank to be more volatile, you could open a few more accounts and alternate play amongst them.

I do not want to use several accounts because 1) I like the name, 2) I am well known under that name, 3) I do not support an insane rating system by fitting its design weaknesses but quite contrarily a rating system ought to handle all players equally well regardless of their (possibly changing) playing frequency.

RobertJasiek wrote:I do not want to use several accounts because 1) I like the name, 2) I am well known under that name, 3) I do not support an insane rating system by fitting its design weaknesses but quite contrarily a rating system ought to handle all players equally well regardless of their (possibly changing) playing frequency.

Why would you want a server to assign you a rank at which you are statistically expected to be a poor match for your opponents in even games?

linuxdaemon wrote: I am probably realistically like 25k on a good day. I played a couple rated games on KGS earlier in the year where my rank fluctuated between 27k and 29k. Today I tried to play a 26k bot and it told me I could not play that rated because it would be more than six stones handicap. I look at my user information and it shows "Rank: 2d?".

RobertJasiek wrote:When you play a lot, your rating does not move. When you don't play, hectic big rating jumps are, uh, called a feature instead of a bug,

Luckily, the answer to both posters problems is identical! You can change the handicap setting in the game offer window and still play rated. If you simply play games handicapped at the rating you believe you should be, KGS will compensate accordingly and your rank will quickly move to where it should be. There is no "penalty" to any of your opponents because once your rank takes on the new value, KGS considers what you should have been ranked the whole time. (Note: This only applies if you are able to actually maintain the correct win/loss record to justify your new rank....but if you don't, then your rank estimate was off so all is still well).

jts wrote:Why would you want a server to assign you a rank at which you are statistically expected to be a poor match for your opponents in even games?

I would not want. (Nor have I said so.)

Mef wrote:Luckily, the answer to both posters problems is identical! You can change the handicap setting in the game offer window and still play rated.

Gulp. 1) I dislike handicap games. 2) KGS rating of handicap games is even worse than that of even games, so I prefer to avoid handicap games on KGS (and in EGF tournaments). 3) Circumventing the even game rating problems does not solve their treatment.

If you simply play games handicapped at the rating you believe you should be, KGS will compensate accordingly and your rank will quickly move to where it should be.

No. I tried. The opposite has been the case for me.

RobertJasiek wrote:

Mef wrote:Luckily, the answer to both posters problems is identical! You can change the handicap setting in the game offer window and still play rated.

Gulp. 1) I dislike handicap games. 2) KGS rating of handicap games is even worse than that of even games, so I prefer to avoid handicap games on KGS (and in EGF tournaments). 3) Circumventing the even game rating problems does not solve their treatment.

If you simply play games handicapped at the rating you believe you should be, KGS will compensate accordingly and your rank will quickly move to where it should be.

No. I tried. The opposite has been the case for me.

Not to entirely derail the thread....but a quick look at http://kgs.gosquares.net/ for a certain well-known account of yours informs me that over the last two months you have won 51.6% of your games (N=64, all rated). Now, games that would still be included in your rating would be over the last six months... Jul-11 through Dec-11 gives a 51.0% winning percentage (N=516). Over the last 7.5 years it shows you winning 53% of your rated games, 50% of your even games, 50% of your games taking handicap and 47% of your games giving handicap. If I were attempting to do an assessment of the server's rating system, I would be tempted to use your account as a stunning endorsement.

Perhaps this is merely a case of perception not fitting data?

There are reasons why, with that account, I win only some 50% of my games. E.g., I also use it to play during the night around 3 am, when I am about a rank weaker. If KGS did measure all my games over the years, the rating could be considered fair for that usage behaviour. However, I use the account differently during different periods. In particular, when I did play badly during a night and dropped to 4d, then typically I do not play during the middle of the night at least until I am 5d again. It takes 1 night and about 25 games to drop from 5d to 4d but it takes 3 or 4 months and hundreds of games to raise from 4d to 5d. There is no symmetry! Dropping is much faster than increasing.

When I dropped to 4d without having played a bad night before, then more often than not I noticed this behaviour: A sequence of exactly 1 lost game suffices to drop from 5d to 4d but right afterwards a sequence of 2 or 3 wins does not suffice to raise from 4d to 5d. Again there is no symmetry! Dropping one rank can occur with ordinary winning percentages like 49% (or even with much higher percentages for the last few days). For players playing a lot (something like 20+ games per day) raising one rank from 4d to 5d seems to require winning of at least 68%, better over 70%, and yet better over a period of many weeks.

When I play almost no games during a couple of weeks, then 1 win can cause an increment of, say 7% of a rank. Thus the by far easiest method for improvement is first not to play at all for several weeks or some months, then choose known easy (or at least not the toughest) opponents (and don't play blitz), play seriously and win 10 or 20 games in a row (in a real world tournament, such is essentially impossible except for the very top players but on KGS it is rather easily possible when playing seriously during the day rather than the night; ask, e.g., Cornel: he also has such winning series).

None of these rating system behaviours makes sense, except for punishing harshly the most active, single account players.

RobertJasiek wrote:There are reasons why, with that account, I win only some 50% of my games. E.g., I also use it to play during the night around 3 am, when I am about a rank weaker. If KGS did measure all my games over the years, the rating could be considered fair for that usage behaviour. However, I use the account differently during different periods. In particular, when I did play badly during a night and dropped to 4d, then typically I do not play during the middle of the night at least until I am 5d again. It takes 1 night and about 25 games to drop from 5d to 4d but it takes 3 or 4 months and hundreds of games to raise from 4d to 5d. There is no symmetry! Dropping is much faster than increasing.

When I dropped to 4d without having played a bad night before, then more often than not I noticed this behaviour: A sequence of exactly 1 lost game suffices to drop from 5d to 4d but right afterwards a sequence of 2 or 3 wins does not suffice to raise from 4d to 5d. Again there is no symmetry! Dropping one rank can occur with ordinary winning percentages like 49% (or even with much higher percentages for the last few days). For players playing a lot (something like 20+ games per day) raising one rank from 4d to 5d seems to require winning of at least 68%, better over 70%, and yet better over a period of many weeks.

When I play almost no games during a couple of weeks, then 1 win can cause an increment of, say 7% of a rank. Thus the by far easiest method for improvement is first not to play at all for several weeks or some months, then choose known easy (or at least not the toughest) opponents (and don't play blitz), play seriously and win 10 or 20 games in a row (in a real world tournament, such is essentially impossible except for the very top players but on KGS it is rather easily possible when playing seriously during the day rather than the night; ask, e.g., Cornel: he also has such winning series).

None of these rating system behaviors makes sense, except for punishing harshly the most active, single account players.

Since it's a slow day and we've already started down the rabbit hole...

Once again, your perception simply does not match the data. First of all your assertion that it takes 25 losses to demote, and hundreds of wins to promote is patently absurd, because that would mean your rank would drop a stone for every 100 games you played at a 50% win/loss (and over your 12,000 game KGS career your rank would have fallen 120 stones!).

Luckily we can supplement the analytics data with you actual KGS rank graph. I'm assuming that your reference to a "bad night" is the spot in April where your graph falls sharply, however if you actually look at the game data, that wasn't a bad night, it was a 4 day stretch were you went 15-32, and as the data shows it was part of a greater 4+ month period of having a sub-50% win-rate.



In early to mid July you have a sudden upswing - the result of going 78-35 in a 5 day period. Note - you have a good run at 70+ winning %, but it's also in the overall context of two and a half months of winning. If you normalize this for your game playing rate it would have been the same as going 23-10 back in April -- very similar to the streak that knocked you down (which makes sense because they both only took 4-5 days).



After you promote again your performance for the next few months is similar to January-March levels, though actually a little worse. (your overall win-rate August to November is 43.6%)



Ultimately we get to a 53% win-rate on the year, with about 60% of those games handicapped as a 5d and 40% as a 4d, and your rank is more or less exactly where it started. Your rank was stable except for two "swings", both swings occurred over a similar amount of time and required a similar number of games relative to your game playing rate.

Any sense of taking hundreds of games to build your rank back up, or losing it after only a handful of games is only present due to a failure to recognize the context in which the streaks occurred. The system's moves make perfect sense when you remove yourself emotionally from the problem and look only at the data presented.