Personally, most of my games fall in the category of game won by a big group captured in the middle of the game or game won by 10+ points difference. Hardly does 6.5 Komi make any meaningful difference. On 13x13, though, 6.5 makes a difference, but often as a result of a complicated situation that gives a side extra few points. The result of these complicated situations for players with similar level feels more like a coin toss, than a real difference in skill. I derail a bit from the main post, but the game sometimes feel more probabilistic than deterministic.
I notice though that in higher level games, the score is often very close.
Most is not all. For some of your games you still need komi.
It matters more and more as you get stronger.
[8d white without komi] vs [8d black] supposed to have same winrate as [8d white with komi] vs [9d black]
so, if you are not against playing even games with 1 or more ranks stronger opponents, it makes perfect sense to also be not against playing without komi as white
Komi didnât exist a century ago. To manage equity the players had to switch colors in a serie of games.
In that case the way players search a win differs. No more searching the perfect balanced move, instead white wants trouble and black security. The opening theory indeed changed a lot with the introduction of komi in the rules.
If we have enough consistency in our games at our low level or not, thatâs a question for sure. But the existence of komi change the perspective anyway for all levels of players.
I guess that if many games finish by resignation, thatâs for some reason like being testing things, no compromise, no alternative ways considered, too much self-esteem, bad reading⊠I mean still missing some more balanced way to drive a game.
No, a game without komi is equivalent to one half rank. So (n dan without komi vs n dan black) is equivalent to (n dan white with komi vs (n+1/2) dan black).
The EGF rating system predicts winrates like this:
n=1: WR = 43%
n=5: WR = 37%
n=8: WR = 28%.
As @Groin said, the modern komi system as we know it, only started to formalize in the mid-20th century, although its early format involved a transition from the teai system (playing many games without komi with different ratios as black or white and handicaps). From my research previously (still not completely), they didnât just appear out of nowhere, but measures of half a stone or different advantages for the extra âfirst moveâ (sente) were used for players of different strengths. And game records with this early âkomiâ format can be traced back at least to the 19th century (perhaps even older).
And players had known about the âfirst moveâ advantage (ć æć©) for a long time, even going as far back as the first surviving scrolls around the 5th century. Iâve written about it years ago, and there appeared to be some form of âmeta-countingâ system back then that not only counted âstonesâ within a game, but across several games (and using different counting units, that is, the first player was given 3 âchipsâ, and the second player has to win games or use the captured stones to take those chips back in order to âwinâ across multiple matches, which most likely linked to ancient gambling customs). Although this custom seemed to be lost around the 10th century and mostly forgotten, or was originally only a regional practice. But it certainly showed players knew about the âfirst move advantageâ and the need to compensate one way or another for thousands of years.
descriptive versus normative statements.
So, reverse komi is needed for âhandicap 1â games to have 50.0% winrate against 1 rank stronger opponent?
Yes, exactly. Thatâs what IGS does.
Do OGS rank system considers it?
No it doesnât.
On OGS, if White is 1.0d, the effect of winning an even game against a 1.0d is the same as winning an H1 game against a 1.0k, and the effect of losing an even game against a 1.0d is the same as losing an H1 game against a 1.0k.
So, OGS is systematically wrong and need to be changed?
Yes OGS is wrong. And not only that, the whole calculation of rating points for handicap games looks wrong
(Assuming the rating calculator tool Play Go at online-go.com! | OGS reflects accurately what is done on the server.)
@anoek , is it true?
Itâs probable the bug is in both, but the ratings calculator uses different code (JS) than the real rating system (Python).
If you are interested, here is the code:
- Ratings calculator: online-go.com/src/views/RatingCalculator/glicko2.ts at e8ec958d23b58821bc95c6b6bae74b3274490686 · online-go/online-go.com · GitHub
- Prod(ish): goratings/goratings/math/glicko2.py at master · online-go/goratings · GitHub
The implementations should be identical +/- language, so if you are able to identify the bug/fix in one, it should be pretty easy to port it to the other.
demo board without komi
I passed as black and clicked score estimate: w+6.3
I played 4-4 as black and clicked score estimate: b+6.6
placed two 4-4 like in handicap 2: b+18.9
âŠ
- | -6.3 |
---|---|
1 | 6.6 |
2 | 18.9 |
3 | 32.3 |
4 | 46.5 |
5 | 60.5 |
6 | 71.2 |
7 | 83.6 |
8 | 98 |
9 | 108.7 |
we are supposed to give 2 stones handicap to 2 ranks weaker opponent, and 3 stones handicap to 3 ranks weaker opponentâŠ
so, Kata thinks there is 12.7 points difference on average between each current and next rank
so, komi difference is indeed around half rank difference
I donât understand the code. In particular on line 134
E = 1 / (1 + exp(-g_phi_j * (player.mu - p.mu)))
This corresponds to the formula
I suppose this is a winning probability, and I canât find where the handicap has been taken into account.
Ah yeah, that file might be pure glicko (only considering wins and losses). I think handicap adjustment happens somewhere nearby:
- goratings/analysis/util/RatingMath.py at 6cab3090a54c8b876ac6a8d0886dc9124dadd575 · online-go/goratings · GitHub
- online-go.com/src/lib/rank_utils.ts at e8ec958d23b58821bc95c6b6bae74b3274490686 · online-go/online-go.com · GitHub
Disclaimer: I havenât spent much time in the ratings code myself