Winning statistics since Jan 2018
Interesting, so as one might suspect, the impact of difference in grades becomes much greater in the low kyu and dan ranks, which might explain some of the discussion above as well.
I recall 3 ranks of difference not making remotely as much of a difference in the ddk and even high sdk (under 7k?) ranks as well, when compared to lower sdk, and especially dan ranks.
(e.g. the win % likelihood for the weaker player in a game between a 5k vs 2k feels much greater to me than a 3d vs. a 6d, and even greater if it is a 14k vs. 11k; even greater again if it is a 28k vs. 25k game.)
To me it feels a victory by an opponent 3 stones weaker happens much more frequently in the double-digit kyu ranks, and especially below 20-25k or so.
Overturning an even game against a stronger opponent felt much easier to me and more likely at that rank as well (for example, 14k vs 11k) than in the SDK and dan ranks â although admittedly I didnât spend any more than a month as a DDK myself, and it was long ago, and I donât watch many DDK games.
My impression is that there are many gaps and potential errors which can change the game in one playerâs favour or cause it to swing wildly at DDK levels.
(For example, there are often many +/- 70+ point moves being played by both players throughout many DDK games, even throughout the midgame and endgame, which is very rare in mid or high dan level games, and much rarer in high SDK games â it can often be enough to end a game decisively if a mid-dan player blunders to that degree)
And these types of gaps begin to somewhat close up in the SDK and dan levels and produce more consistent games the more one increases in rank.
So it is a bit easier to determine whether a 5k is âactuallyâ closer to 2k, than whether a 27k is âactuallyâ closer to a 24k by looking at the game record.
But it would potentially explain the comments by @Gia and @Cchristina in this thread regarding 3 ranks being a small difference, and perceptions that a 3-rank difference doesnât mean much at their ranks.
At 1k-5k or progressively higher dan levels, my impression is that the win likelihood against a significantly stronger player (3+ stones) continues to decrease noticeably with each progressive rank (or at least the lack of as many opportunities to overturn a game, and as much variance created from opponent mistakes, is more noticeable after progressing several ranks).
So one would expect to see noticeable differences between a 3d and a 6d which stay consistent over a large number of games and leave less opportunity for the 3d to win against the 6d, as compared to the much greater possibility for variation and game-overturning tactics often seen in games between players with the same 3 stones of rank difference, but ranked 10k-30k.
I donât know if statistics for players around 20k should be taken seriously. The initial rank of beginners is often set at 20k, although they may be weaker. And people in the 15k-19k range may actually be stronger than that because they are quickly improving and their EGD rank canât keep up with that if they donât play frequently enough.
Note that the predicted/observed winrates by rating difference are only applicable to the start of a game.
As the game progresses, Iâd expect that the winrate continually shifts, depending on the level of the players, the lead that a player has, and the phase of the game.
A 10 point lead at move 30 is not as decisive as a 10 point lead at move 240.
A 10 point lead at move 30 is not as decisive for a 20k as it is for a 9d.
This research may be relevant (mapping AI winrate to human winrate, depending on the level of the human player and the phase of the game):
Iâd like to read that research paper, but I havenât found it yet.
Yeah I downloaded it from there previously also. Thanks for linking.
I think as well the rating gaps between ranks are not constant
In elo the difference in rating translates to a probability of winning, but the difference in ranks will be a bigger rating gap at higher ranks than at lower rank.
Like at 15 kyu maybe a three rank gap is 150 points, but at 4d it could be 300 points. So it could be a much more decisive advantage at higher ranks translating rating gaps to probabilities.
Iâm a bit confused. I thought that we do have different gaps between the ranks at different skill levels, and that the reason for this is so that the probabilities of a win for a given gap remain constant across the skill range? (obviously, this could only be expected to hold for moderately small gaps)\
Yes. OGS has that. Itâs reflected by the non-linear formulas used to convert between rating and rank:
The EGF rating system (implicitly) uses similar non-linear conversion formulas (but the AGA rating system doesnât IIRC). See this post for a comparison between the EGF and OGS conversions:
Is that the reason? Are you sure itâs that the win rate for even games remains constant?
Or is it that the handicap to rank gap win rate stays roughly constant.
That is if a 2kyu can win 50 something percent of the time with giving three stones to 5 kyu, and a 6dan can win 50 something per cent of the time against a 3dan giving three stones, are we expecting that the win rate in even games between the 5kyu and 2kyu and the 6d and 3d is going to be the same also?
Itâs not clear to me that this should hold necessarily.
Oh you are right! The changing size amount of rating per rank is for handicap constancy.
Huh - fascinating: whatâs the relationship between handicap stone effect on winrate and rank ⌠does it amount to the same thing?
In the EGF rating system, handicap is accounted for by elevating the rank of the black player by (n - 0.5) where n is the number of handicap stones (on 19x19), and use the corresponding elevated rating to calculate the expected winrate. IIRC OGS does something similar (I think the details can be found in the forums).
Yeah I think OGS treats one of the players, say the weaker as if they three stones stronger when they have a three stone handicap for the rating calculation.
That ties the rating system to the handicap/ranks quite well.
But that doesnât seem to suggest that without handicap, a three stone/rank gap should give the same expected winrate at all rank differences right for even games right?
When you do the rank to rating calculation youâll get bigger gaps at higher ranks than at lower ranks and thereâs no adjustments because thereâs no handicap stones in the even game.
Indeed. Those rating-rank conversion formulas give bigger rating gaps between players 3 ranks apart at higher ratings than at lower ratings. So a 4d against a 7d in an even game is expected to have a (much) smaller chance to win than a 20k against a 17k. The former might be something like 10% and the latter something like 40% (also see the statistics posted by @Atorrante and @jlt earlier today). At the same time, the chance should be close to 50% when the weaker player gets 3 stones handicap.
Hi everyone, I am a beginner in Go.
So far, I havenât resign because I donât know when to do it.
In fact I didnât even realize I lost 300 points until the score was calculated.
I want to know if players like me make my opponents feel impatient?
If my behavior violates the rules of OGS?
Additionally, am I responsible for deciding that my game is no longer winnable and resigning it?
In order to determine whether a game should be resigned, you need to
- know the life and death status of groups
- be confident that if you try to kill one of your opponentâs group which is alive, heâll know how to defend it.
- count.
Beginners are not expected to know all that.
Thats why playing offline is more enjoyable than online. Feel the stone, feel the board, watch opponentâs expression
If you donât disable analysis in your custom game preferences you can use the score estimator during the game.
If you see that you are more than [your choice: âŚ, 20, 30, 40, âŚ] points behind, you can resign.
The in-game score estimator (accessible to the players during the game) can easily be off by dozens of points, because its life and death skills are only beginner level, if even that (and this is by design).
In my own games itâs not uncommon for the SE to be off by 20-30 points at some point during the game, and just be wrong about whoâs leading.
So I wouldnât recommend resigning whenever the SE says youâre behind by 30 points. With such a relatively small margin, there is a significant chance that the SE is wrong and you are in fact leading. And even if itâs right, a beginner game may still swing the other way in a later stage of the game.
If youâre a beginner and you donât know yourself what is dead/alive during the game and you want to rely on the SE when considering resignation, Iâd recommend to wait until the SE says youâre behind by a much larger margin (say 70 points) before resigning.
I wouldnât worry too much about your opponent getting bored when you resign âtoo lateâ. When your rating has settled on a certain level, itâs reasonable to expect similarly rated opponents to have the similar issues with positional judgement as you. So when you have no clue whoâs leading and by how much, Iâd say itâs safe to assume your opponent has no clue either.
As for potentially violating the rules of OGS, I think OGS moderators can tell the difference between a beginner not knowing about life and death, and a more advanced player trying to annoy their opponent.
I fully understand your wish to learn to resign. Thatâs great, meaning you are not so ill to lose.
Now you mostly have to rely on your own estimation of your chances to win. No real shortcut on this.
If you canât see even if you care, itâs simply because you have too few experience and still miss some basic (like cutting/connecting, alive/dead, closing boundariesâŚ)
But the good side is that you are concerned by the goal of the game, and yes, thatâs the way to go, to not forget to take the biggest, or the most urgent at each of your moves. So far, so good.
Keep playing, keep the goal and soon it will be more clear when resigning.
Ps: donât worry about attitude (OGS, opponents), it takes a bit of time to grasp the game basics, and, guess, itâs like everyone went more and less through these early steps.
Two recommendations: (1) Prioritize playing live 9x9 games until you understand life and death and have a basic knowledge of counting, and (2) if you want to continue playing 19x19, at least play fewer games simultaneously, so you can concentrate on making better moves.