Fast != slow

[uPWarrior]

It is strange that no Go server has ever supported blitz ratings, but there you go, they never have. There doesn't seem to be much demand for them to appear. I rather doubt that any amateur organisation has the capacity to construct a second rating list.

Doesn't OGS now do it? I can't check quickly since I'm on iPad and OGS "hover to find out what this is" doesn't work that well here.

They do, there are blitz, live and correspondence ratings now (there is also an "overall" rating, but I don't know how it's calculated. Maybe ignoring the timesettings?)

[snorri]

An amateur organization could decide to implement blitz ratings, but the discussion over where to draw the line for how fast blitz is would likely end in fisticuffs.

[Uberdude]

EGF ratings already have a tournament class weighting factor so that faster games change your rating less.

[palapiku]

I believe the reason it's common to have a single rank, instead of different ranks for different classes, lies in the origin of the dan/kyu system. Your rank is not just a number, it's almost like a title. Nowadays it is usually calculated in the same way as elo-style ratings, but it can also be given after an examination, or simply announced by your teacher, or given as an honor. Sometimes for life. It's not really meant to be an exact estimate of your current ability the way chess ratings are. This is still generally the case with pro dan ranks.

[EdLee]

EGF ratings already have a tournament class weighting factor so that faster games change your rating less.

Nice.

[EdLee]

the discussion over where to draw the line for how fast blitz is would likely end in fisticuffs.

A general time setting:

Start_time + N_period * ( Byo_yomi / Moves_per_period ) + eXtras

They'd have to decide on the (range of) values for S, N, B, M, X.

Samples:

Popular on KGS: S 30 mins, N 3, B 30 secs, M 1. ⁽¹⁾
Popular on IGS: S 1 min, N infinite, B 10 mins, M 25. ⁽²⁾
"Weekend" AGA tourneys: S 45 mins, N 3, B 30 secs, M 1. ⁽³⁾
US Open: S 90 mins, N 3, B 30 secs, M 1. ⁽⁴⁾
Popular KGS blitz: S 1 min, N 3, B 10 secs, M 1.

Personally...
^(1)(2)(3) All blitz.
⁽⁴⁾ The initial time of 90 minutes is nice, but the 30 sec byōyomi becomes blitz.

[tentano]

It really makes a lot more sense for blitz to have its own rating. As it is, I wonder if it doesn't scare people off trying competitive blitz because they're going to damage their normal rating with it if they fail. There's a reason so many people keep a separate account for blitz games.

Of course, that also means there should be more blitz tournaments. As far as I know they're fairly rare. That might be the main reason a separate rating never happened.

[DrStraw]

the discussion over where to draw the line for how fast blitz is would likely end in fisticuffs.

A general time setting:

Start_time + N_period * ( Byo_yomi / Moves_per_period ) + eXtras

They'd have to decide on the (range of) values for S, N, B, M, X.

Samples:

Popular on KGS: S 30 mins, N 3, B 30 secs, M 1. ⁽¹⁾
Popular on IGS: S 1 min, N infinite, B 10 mins, M 25. ⁽²⁾
"Weekend" AGA tourneys: S 45 mins, N 3, B 30 secs, M 1. ⁽³⁾
US Open: S 90 mins, N 3, B 30 secs, M 1. ⁽⁴⁾
Popular KGS blitz: S 1 min, N 3, B 10 secs, M 1.

Personally...
^(1)(2)(3) All blitz.
⁽⁴⁾ The initial time of 90 minutes is nice, but the 30 sec byōyomi becomes blitz.

You number show up immediately how hard it would be to reach consensus. To me, blitz is anything where the game can expect to end in less than 15 minutes. 15 minutes to one hour would be considered normal on a server. But for a game which contributed to any rating to which I gave much credence I would want the game to last at least 90 minutes and preferably longer.

This is one reason why I don't have much faith in online rankings except as a general guide.

[wineandgolover]

I think it would be great if the AGA supported dual rankings and encouraged blitz tourneys or side events. I think the kids would love it, and it would help grow our game.

And I dont even like blitz.

[Bantari]

probably be impossible (or at least - very hard) to actually qualify to any degree.

On the contrary, it's very easy (for me).

So, how do you quantify?
I know that you can easily see that you overlooked stuff and blundered here and there, but this is now what I am talking about.

Lets look at an example.

You play a 100 moves fast, and make 7 mistakes which you would not make in slower games. But how do you say that 3 of these mistakes were 3k level, one was 7k level, and the rest were 5k level? And how do you then calculate that this brings you down from 1k to 4k? How does the formula go? When you look at a mistake, how do you tell exactly what level this mistake is?

Also - as part of the equation - how do you account for your opponent making mistakes, and these in turn inducing your mistakes? Or maybe allowing (and even inducing) your brillant plays, which you would have had no chance to play in slower games?

So - how do you account for some possibly (much?) better moves which your intuition (or your opponent) allowed you to make when the brain was not involved in the decision making? I know for a fact that some of the moves I make in fast games can actually be better than the moves I would make in similar situations if i thought more.

All we can see from fast games is that "I blundered, overlooked atari, lost a group, and lost the game, would never have happened in a slower game" - but what does this really mean for your overall move level average? It seems to concentrate on the decisive blunder the most. But is that correct?

Is that also influenced by the difference in the way you approach the game? For example: intuitive vs calculated players? Or those relying on memorizing shapes vs memorizing sequences? Stuff like that...

I find it really interesting.

PS>
The monkey is really cool!

[EdLee]

Bantari, as I said, it's just my feeling.
If you want some actual numbers, be my guest:
grab a bot -- the best current ones are around 4~6 stones from pro on 19x19 ? --
and do some statistical analyses to find out the relative levels
when you play it at different time settings.
The procedures are very simple; it's the logistics --
bot availability, human operators, time, etc. -- that take effort.
Enjoy and let us know your findings.

[Bill Spight]

You play a 100 moves fast, and make 7 mistakes which you would not make in slower games. But how do you say that 3 of these mistakes were 3k level, one was 7k level, and the rest were 5k level? And how do you then calculate that this brings you down from 1k to 4k? How does the formula go? When you look at a mistake, how do you tell exactly what level this mistake is?

One problem is that the average loss per move is quite small per stone difference in strength, so that a nine stone difference means a loss of around one point per move on average. So what about a move that loses 30 points?

Suppose that you or I make 7 4-5 kyu plays that we otherwise would not make. That would bring our play down by about 1/2 stone. OTOH, we could make a single play that loses that much and not even notice.

[Boidhre]

probably be impossible (or at least - very hard) to actually qualify to any degree.

On the contrary, it's very easy (for me).

So, how do you quantify?
I know that you can easily see that you overlooked stuff and blundered here and there, but this is now what I am talking about.

Lets look at an example.

You play a 100 moves fast, and make 7 mistakes which you would not make in slower games. But how do you say that 3 of these mistakes were 3k level, one was 7k level, and the rest were 5k level? And how do you then calculate that this brings you down from 1k to 4k? How does the formula go? When you look at a mistake, how do you tell exactly what level this mistake is?

Also - as part of the equation - how do you account for your opponent making mistakes, and these in turn inducing your mistakes? Or maybe allowing (and even inducing) your brillant plays, which you would have had no chance to play in slower games?

So - how do you account for some possibly (much?) better moves which your intuition (or your opponent) allowed you to make when the brain was not involved in the decision making? I know for a fact that some of the moves I make in fast games can actually be better than the moves I would make in similar situations if i thought more.

All we can see from fast games is that "I blundered, overlooked atari, lost a group, and lost the game, would never have happened in a slower game" - but what does this really mean for your overall move level average? It seems to concentrate on the decisive blunder the most. But is that correct?

Is that also influenced by the difference in the way you approach the game? For example: intuitive vs calculated players? Or those relying on memorizing shapes vs memorizing sequences? Stuff like that...

I find it really interesting.

PS>
The monkey is really cool!

Someone could play a bunch of games under blitz and slow time controls and then post them without the time stamps (I don't know if you can remove them from the file or would have to manually create a new sgf) and ask stronger players to identify which games are which. Could be interesting.

[Matti]

The European ratings deal with this by requiring sufficient thinking time for a tournament to be included with full coefficient of 1 (75 minutes sudden death or 60 minutes plus 15 minutes for 45/60 moves in Japanese/Canadian byoyomi). There is also a minimal requirement (30 minutes or 25 minutes plus 5 minutes for 45/60 moves in Japanese/Canadian byoyomi) for a tournament to be included at all, and then, it is only included with a coefficient of 0.5 (or 0.75 for slightly slower games).

When this system was prepared I considered also that it is possible to give the EGC main tournament a higher weight like 1,25, but the I thought not to make an exception for one tournament.

[xed_over]

Based on Ed's feeling that he plays 2-4 stones weaker in blitz games, wouldn't that be more or less true for everyone else too? That's the primary reason for wanting separate ratings, correct?

But if everyone's rating drops a couple of stones, then the relative rating difference between each other would be relatively unchanged, and in any rating system, the only thing that matters would be this relative difference and the win/loss probabilities of that difference.

If the difference is unchanged, I see no need to have separate rating calculations.