A more flexible, fluid system would work well for telling a player their level.

If you could reach 3d and maintain it, then you`d be 3d. If you could not maintain it, then you`d only be a 2d or 1d who happened to have been on hot form. However, knowing that you could get there would encourage you to keep trying, and to do the work necessary to make sure that eventually you could maintain it.

If you went down to 10k and stayed there, then you`d be 10k. If you went down to 10k and could not "maintain" that, then you`d only have been a stronger player who happened to have been drunk or emotionally discombobulated for a night or two.

How on earth can KGS`s system possibly be "accurate" when it is devoted to making sure nobody`s rank ever changes?

Change is part of reality! A system that embraced the fact that people change, both from day to day, and over longer time frames too, would surely be more accurate than one that denied that change occurs.

_________________
Learn the "tea-stealing" tesuji! Cho Chikun demonstrates here:

If you want to avoid playing against people who themselves go on tilt, and play badly, there is always the computer go room to consider.

_________________
http://tchan001.wordpress.com
A blog on Asian go books, go sightings, and interesting tidbits
Go is such a beautiful game.

KGS will not tell people "what they really are". It will only tell them what they were at the time they played their first few rated games. After that, the rank becomes set in concrete for frequent players, and the only way to avoid that is to NOT play rated games. Simply getting better at go won`t do the job.

I`ll tell you something interesting that I`ve noticed: it's a lot harder to beat a Japanese KGS 1k than a 1k from some other countries. I suspect it`s because players in the Japanese room tend to play mainly among themselves, and preserve a different level. I mention this as another example of how a system could be skewed. There are pockets of players, selected by nationality or some other shared factor, who rarely play outside those pockets, and they have a thwarting impact on the rating system too. It is quite possible that a J-room 3k might have been a 1k had they only played in the EGR from the start, but if they decided to switch to the EGR after that, then they`d find it unreasonably hard to rank up.

In other words, there can be no doubt that a system designed to accommodate change and instability would eventually yield more accurate results than one based on the premise that people don`t change and do not differ.

Let`s see how things pan out with Kaya. My guess is that people will be a good deal more satisfied with its rating system.

_________________
Learn the "tea-stealing" tesuji! Cho Chikun demonstrates here:

Some of its definite weak points:

- For a too high percentage of closely KGS-ranked players, the winning percentages for even or H1 games can be too far from ordinary 50% or 67% assumptions, e.g., can be 85 ~ 100%. (Comparion: Real world European ranks (with the known national exceptions of French kyus, 1 rank tougher CZ, FI, P players) almost always meet 50% or 67% assumptions reasonably.)
- It is absurdly tough to improve a KGS-rank.
- Regular players are punished for playing regularly: their rating increment is absurdly tough.
- Thus it is punished to maintain the same account but rewarded to create new accounts when the old account's KGS-rank dropped. The opposite would be fair.
- Sudden global, manual rating shifts.
- Sudden very hectic rating shifts of some individual players. Possibly alternated with fake periods of rating stability, then another hectic shift by, say 2 ranks. Etc. Absurd!
- The H1 default for players one KGS-rank apart encourages too many players to decline even games. Thereby the percentage of more meaningful even games is kept low unnecessarily. The default ought to be even games.
- I have never had the impression that winning an H1 game would reward the winner sufficiently, but I have always had the impression that losing an H1 game hurts the loser too badly.
- Bad system parameters: see my earlier discussions in various forums.

This reminds me of the way the dual-n-back program works. There are two thresholds, an upper (80%) and a lower(40%). If you manage to cross the upper one once, you jump immediately (to the sound of cheers) to the next difficulty level. You then stay there until either you promote again or you cross the lower threshold three times. This gives you a chance to get a good taste of what the harder level is like before being demoted, and the idea is that by getting accustomed to the greater difficulty, you will have better chances of improving.

_________________
Patience, grasshopper.

This is the way our national ranking system works - it's quite hard to get promoted, but its also quite hard to get demoted. The ranks are discrete levels, and to get promoted requires you to gain 1000 rating points, but to get demoted requires you to lose 1000 points (effectively a 2000 point band in which you can travel). The number of points you win or lose for a game varies by rank, e.g. you might get 1500 points for a win as a 29k, but only 50 points as a 3d. Anyway, it allows you to get used to the new level and grow into it.

This in contrast to KGS's system which actively discourages you from playing after you've been promoted because of the risk of a loss which will immediately demote you again. Of course, KGS is not in the business of massaging ego's or letting people grow into their rank, its about accurate and up to date handicapping.

Our ranking system is not without it's flaws, for example, it's very resistant to external improvement. But for active players the rank preservation is very attractive.

This is not only very important, it is also pretty much unsolvable.

Many, if not most, players seem to believe that they win games because they played well, and that they lose games because they were tired, frustrated, tipsy, or out of shape. You hardly ever see players claim that they only won because their opponent was tired, frustrated, tipsy or out of shape, or that they lost because their opponent simply played better.

This is actually a well known psychological phenomenon, called illusory superiority.

The same phenomenon explains why most people think they are above average drivers, believe they have an above average intelligence, think themselves more popular than their friends, and are generally more likely to consider themselves to be superior at all sorts of things.

So unless you somehow manage to alter the very psychological make-up of people, a fair rating system will always result in complaints about the rating system.

You may indeed be right that many people think they are better than they really are.

But that does not change the argument all that much! It`s not my rank that I am complaining about, but the way ranks get STUCK if you play a lot of rated games. I`d prefer a rating system that was more responsive to change, both upwards and downwards. There are many others who feel the same.

I`ve played with a lot of people who appeared to be on tilt, too. I felt sorry for them, because I can sympathise with their situation. If only the KGS system was more responsive to change and did not make so much of history, the occasional tilty outburst would not be so hard to recover from.

Look at it from another point of view - suppose somebody weaker managed to get a 5d rank by playing their first ten or so games at a blitz time limit and managing to fudge a few wins on time. After that they might decide to play slower games for the real practice - but since they are worse than their rank, and KGS denies the possibility of change, then it will take a long time for such a person to fall to their real level, and in the meantime they will pollute the system.

Yes, once more, I don`t deny that many people falsely believe they are better than they really are. But some people DO get stronger, or at least make a sincere attempt to do so, and it would be a real improvement in the KGS experience if their needs were better met.

_________________
Learn the "tea-stealing" tesuji! Cho Chikun demonstrates here:

Yes, but this does not invalidate the other arguments. It just means that not every post in every discussion mentions also these basics, e.g. because nobody wants to go through his recent hundreds of games to generate statistics exactly how many of the games were won / lost by tiredness versus superior / inferior play.

I suspect that a more responsive rating system would result in just as many complaints about how volatile the ratings are, people complaining that they dropped two ranks just because they lost a few games, and now they can't get games at their own level. People complaining that their opponent is too overrated or too underrated because he won/lost a few games, etc.

Which is impossible, as you cannot know whether your opponent was tired or not.

And it doesn't matter anyway. Your playing strength includes the way you play when tired and when your opponent is tired. It includes stupid blunders and brilliant tesuji. It includes your whole game, not just the parts you like. Many players seem to have this notion of their "real strength", which is usually how strong they would be if they removed all the parts of their play they don't like. And it is nonsense. Being able to play well even when you are tired, even when you in byoyomi, even when the game is decisive to win a large prize, all of that is part of your playing strength.

I know several players that are stronger than myself not because they read deeper, or know more about the game, but because they are able to play more consistently, because they have more stamina, because they never give up, and will grab any chance I give them. That does not mean I am "actually just as strong". No, those players are stronger than me. They win more games. If you win more of your games against the same opponents, you are stronger, it is as simple as that. No excuses, no illusions.

That still doesn`t change the argument. KGS is TOO stable. Nobody is asking for it to be made TOO responsive, either - only that it is made more responsive than it currently is.

_________________
Learn the "tea-stealing" tesuji! Cho Chikun demonstrates here:

Which just means that a rating system needs a proper balance. Neither too static nor too volatile.

There`s a flaw with this argument too.

What if you DO get stronger? Is the past still relevant?

If some 70-year-old retiree, let`s call him Trevor, who`s been 5k all his life became stronger in his new-found free time, would you insist on basing his rank on all the games he ever played, from up to 60+ years before his improvement? He might be better at go now than his friends Bob and Sosuke...oh, but by your reckoning Bob and Sosuke are stronger because they have won several thousand more games than Trevor. Does Trevor have to win several thousand more games from this point to prove that he has improved?

People CHANGE. When people change, this should be recognised.

An ideal rating system would enable people to rank up on the basis of good results or rank down on bad ones. Exactly how easy or difficult changing rank should be is admittedly hard to determine, but my opinion is KGS errs too far on the conservative side. You real level would be shown by the level you can maintain, not by your peaks or troughs.

_________________
Learn the "tea-stealing" tesuji! Cho Chikun demonstrates here:

First of all, this sounds inconsistent. If KGS doesn't punish you for losing games, then that minimizes the extent to which tilt sets you back.

As for ranks getting stuck, this is a bad picture. What happens is that if you play a lot of games, the system gains evidence about how strong you are. It then uses this evidence later. But you ask "What if you get stronger? Then you're stuck, because it's using the old games." The thing is, there is no way any system can recognize that improvement (faster than a KGS-like system does) without making errors.

Say you play 100 games a month, which is a pretty fast clip, and might be the kind of thing that makes people complain about getting stuck (when I played 30-40 games a month, I never felt like my rank was stuck, fwiw). Well, if you play 100 games with a 50% winning chance, there is a better than even chance you will win or lose 6 straight at some point. This, in turn, indicates that there are decent odds that at some point during a given month you'll gain or lose a rank for no reason at all, if the system is flexible (my mental model is the Tygem system, but the point applies to any system that ignores the amount of data collected). The only way to avoid that problem is to make ranks get "stuck". But as the example shows, it's best not to think of it as being stuck. It's best to think of it as avoiding random, meaningless, stupid fluctuations.

Of course, these kinds of arguments can only get you so far: they're heuristics. But it just so happens that we have a wonderful french mathematician who's tested these things with statistics.

P.S. It is emphatically false that KGS assumes no one ever changes. Old games lose weight, which is a built in assumption that players strengths can change. It's true that KGS does poorly with players who suddenly make small gains in strength, but that's not the same thing.

_________________
Occupy Babel!

Where did I say that all his old games should be relevant?

Of course your playing strength can change. And of course your rating should be based on more recent data. I would consider that entirely obvious.

All I am arguing against is people who want their "tired" games to somehow not count.

While we're coming at this from similar points of view, I differ with this point. It is not entirely meaningless. It's just impossible to systematically test (this may be my favorite philosophical distinction, fwiw).

Sometimes we see a player who is almost perfect in most regards, but also has a glaring flaw. We recognize that he's better than his performance in the sense that if he fixes that flaw he will be exceptional. And it does matter in one sense. Suppose A is exceptionally talented, but is also an alcoholic, while B is less talented, but is completely disciplined. A wins when he's been sober, but loses more often because of his personal problems. But if A ever gets his act together, B will never match him, no matter how hard he works.

Less extreme examples are common, and while they're prone to bias and wishful thinking, that doesn't mean they're not sometimes real. They're just not the business of a ratings system.

_________________
Occupy Babel!

I can often see this from his moves, as I can see it from my moves years after I played a game.



This is a bad excuse for a bad rating system. First it forces players to become tired, then it blames them for losing when tired.

A rating system is not responsible for a player being tired by solely his own responsibility, but a rating system is responsible for making a player tired in more games.



There are three factors for blunders in server games:
1) blunder due to playing strength
2) blunder due to short thinking time
3) blunder due to the rating system having caused tiredness or frustration

Roughly guessed for my games and per game on average, (1) is below 5%, (2) is about 50%, (3) is about 45%. (During the night, (3) can easily dominate (2).)



It is not nonsense that (3) exists.



Nobody doubts this. It does not remove (3) though.



Nobody doubts this.



Wrong. It is NOT as simple as that. (3) exists. Make the KGS rating system better and I will have a better rating.



(3) is not an excuse or illusion but a fact.

So what. Has he identified and suggested solutions for players affected by the too static rating system?

I want a rating system good enough for not creating players to play more tired games than they would if the system were good.