Reverse Komi; New Players Welcome! Registration OPEN through the 7th

It’s a lot more precise; whether or not it’s accurate to match depends on a variety of factors, though it seems about right so far: I’ve won and lost games using this komi calculation.

I think it bears mentioning that they also accounted for the 0.5, 1.5, 2.5 &c. progression of traditional handicap (Black should win 43% of games, roughly, in OGS automatic handicap).

2 Likes

Ah, yes. I remember that now. OGS doesn’t expect black to win 50% in those cases. So my statement was inaccurate.

Edited: sorry, description of the statistics below was wrong.

In the EGF rating system, even games between players separarated by 1 rank difference shows the following trend for weaker player win% (sampled from 2006 until now):

Rank win%
6d 24
5d 29
3d 34
1d 37
4k 43
8k 43
13k 43

In the EGF rating system, 1 handicap stone for 1 rank difference shows the following trend for black’s win% (sampled from 2006 until now):

Rank win%
dan 43
kyu 42

So OGS’s 43% for all ranks corresponds well to EGF’s 43% for even games between players weaker than about 4k and 1 rank apart and also for 1 stone handicap games between players 1 rank apart.

4 Likes

It looks quite interesting so I’d like to join. Also, I like the scientific vibe in the chat: it brought me back to PhD times.

3 Likes

Pairings for Lagrange Points Round 2021-1 (1) Posted

Good luck to all the participants; I’m looking forward to both playing and watching some great matches.

As mentioned in the OP, pairing rules are in flux.

I am particularly interested in players thoughts on how the current pairing algorithm worked out and if/how it should be changed.

Should extreme reverse komi values be avoided, and if so, how extreme is too extreme? Should only very high (far from 0) reverse komi values be avoided, or should values close to 0 be avoided since they might feel too much like an even game, or is that not an issue?

For this round I rejected reverse komi values of more than 240, but I think limiting it to 120 (1/3 of the board) might be interesting if players want less extreme komi since that is a simple fraction which corresponds to 8 stones and 8 points, which is remarkable close to the highest official handicap: 9 stones. That said, I have played reverse komi games of almost 361 points and they do work, though as one can imagine it at that point does come down to whether or not the weaker player knows how to live in a corner against strong resistance, so there is a bit of degenerate behavior; I think most ddk players could beat anyone with a reverse komi that high: play 3-4 and capture a cutting stone.

Should players be required to have recent ranked games? For this round I only made sure players were non-provisional, but requiring participants to have completed at least one ranked game (live or correspondence, any board size, any rules) in the 3 weeks leading up to the round they’re signing up for might have benefits if players feel participants are not ranked accurately. I don’t know if this is really much of an issue: I updated the ratings for each player before generating the pairings and most players had some rating change since they signed up anyway.

5 Likes

Looking forward to it! I’m only wondering if you know @riiia’s forum name? I messaged on the main site but that make is not recognised in the forums. I’m worried I might miss messages via the main site but will keep an eye out.

I don’t believe I’ve had contact with them on the forums so far.

1 Like

Expecting wurfmau and I to make 200+ points on a ~10k is asking for a lot, so we’ll see I guess.

3 Likes

…But why? XD

Have you tried searching for yourself on OGS? :smiley:

Here are my scheduled games (so far):
sbk96 [9k] (B) vs. wurfmau3 [4d] (W; -165 komi): Friday, 19 March, 13:00 UTC
riiia [5k] (B) vs. wurfmau3 [4d] (W; -106 komi): Saturday, 20 March, 14:00 UTC
wurfmau3 [4d] (B) vs. vyzhael [9k] (W; 180 komi): Sunday, 21 March, 12:00 UTC
Samraku [2k] (B) vs. wurfmau3 [4d] (W; -75 komi): Tuesday, 23 March, 18:00 UTC

link to KoBa

3 Likes

I would like to present my 167 komi game against @wurfmau3 as evidence that larger differences work fine:

I opened with 3-3 points, tried to play extremely conservatively, keep everything connected, not make any corner shapes that had any chance of being dead, etc. Kept counting the score during the game and tried to estimate how much more territory I needed to grab to win, but I didn’t quite make it. Looking forward to the next round!

5 Likes

Registration for round 2021-2 (2) is now open! If you are interested in participating again this coming round, or want to join in, contact me with your OGS username (and preferably with a link to your profile, especially if you have non-ascii characters in your name) and the number of games you wish to play by Apr. 5 EST.

There will be a rating system based on the Japanese Points Rating System, assuming I can fix the bugs over the course of the next week. :wink:

Hope to see you all and more in round 2021-2 (2). :smiley:

For those who participated this round, here are some polls.

Maximum Komi Size
  • 120 (8 stones, 8 points)
  • 180 (12 stones, 12 points)
  • 216 (15 stones, 6 points)
  • 240 (17 stones, 2 points)

0 voters

Should Low Komi be Avoided?
  • Yes, I want to only play players of different levels than me.
  • No, I don’t mind occasionally meeting someone near my skill.

0 voters

1 Like

Can you remind me of the window for round two?

1 Like

Round 2021-2 (2) will run from when pairings are posted on Apr. 6, until Apr. 26, the day before Round 3 begins.

2 Likes

Lagrange Points 2021-2 (2) Begins

The teapoweredrobot vs. wurfmau3 match was a last-second proposal on my part which I’m not sure whether or not the players will be interested in, so that match is completely optional.

Results of Previous Round

Open Ratings and Volatilities

Samraku: 4-0-0
prior rating: 392 (2k) (vol = 84.00)
updated rating: 466 (4d) (vol = 58.50)

illusory_deceit: 0-0-0
prior rating: 294 (9k) (vol = 84.00)
updated rating: 294 (9k) (vol = 84.00)

KoBa: 1-1-0
prior rating: 406 (1k) (vol = 84.00)
updated rating: 406 (1k) (vol = 43.50)

teapoweredrobot: 0-2-0
prior rating: 336 (6k) (vol = 84.00)
updated rating: 286 (10k) (vol = 46.50)

He Who Walks in Shadows: 1-2-0
prior rating: 392 (2k) (vol = 84.00)
updated rating: 352 (5k) (vol = 45.36)

Kaworu Nagisa: 0-0-0
prior rating: 490 (6d) (vol = 84.00)
updated rating: 490 (6d) (vol = 84.00)

pdg137: 0-1-0
prior rating: 322 (7k) (vol = 84.00)
updated rating: 294 (9k) (vol = 44.50)

wurfmau3: 5-2-0
prior rating: 476 (5d) (vol = 84.00)
updated rating: 548 (10d) (vol = 55.11)

DashaTabasco: 0-1-0
prior rating: 252 (12k) (vol = 84.00)
updated rating: 224 (14k) (vol = 44.50)

riiia: 2-1-0
prior rating: 364 (4k) (vol = 84.00)
updated rating: 404 (2k) (vol = 45.36)

vyzhael: 0-2-0
prior rating: 308 (8k) (vol = 84.00)
updated rating: 258 (12k) (vol = 46.50)

LittlePebble: 0-0-0
prior rating: 126 (21k) (vol = 84.00)
updated rating: 126 (21k) (vol = 84.00)

sbk96: 0-1-0
prior rating: 308 (8k) (vol = 84.00)
updated rating: 280 (10k) (vol = 44.50)

I must say I am a bit scared at these results; 84.0 may be too high as a default volatility, and that’s if I close my eyes and hope really hard that that’s the biggest issue with my methodology. That said, the reverse komi recommendations from them don’t appear to be entirely absurd, and if the only issue ends up being general rank inflation (or deflation) relative to OGS, I can easily shift the ranking system to realign it without changing the relative ratings of the players.

Rules Update

I edited the OP, but forgot to mention the change here, so here it is.

Program Updates

What didn’t happen since Round 1 pairings were posted? I rewrote the pairing algorithm multiple times, eventually succeeding in creating a version of it that doesn’t work, thus taking a step backwards from Round 1 when it did work, I added a rating system based on the Japanese Points Rating System, scrapped it, and implemented another one which I suspect stands substantial risk of being incorrectly implemented, having degenerate behavior, or both, thus sentencing an unknown number of future persons to possible cruel and unusual reverse komi sizes, and did some code refactoring, which caused its own bugs.

While the participation this round is quite limited, I am mostly thankful for that as it gives me another three weeks to sort out the mess I’ve created here.

3 Likes

Registration Opens for Round 2021-3 (3)

Registration is now open for the third round of Lagrange Points. Contact me either here or on OGS with the number of games you desire and I’ll add you to the list.

OP and Wiki

As the edit period for the first 3 posts of this thread has expired, I requested a mod to wiki the rest of them. Feel free to add to the Index, but I would request that you please refrain from editing the OP itself. In the worst case scenario, I can just put the rules in a pastebin and link to it, but I’d prefer the convenience of just having it here.

Previous Poll Results

The vote was overwhelming for allowing matchups between closely-rated players, so there will be no restriction regarding that.

Maximum komi size had more diversity.

  • 1 vote: 72 (write-in)
  • 3 votes: 120
  • 1 vote: 180
  • 0 votes: 216
  • 2 votes: 240

I guess this could be seen as supporting the hypothesis that a very wide range of reverse komi works, but smaller values are best, but that could just be confirmation bias. Mean is 156, median and mode are both 120, and geometric mean is either ~21, or my calculations are in error.

Rating Fiasco

So the reverse komi was whack in round 2, that much is certain. I’d also like to keep some closer ties to the OGS rating system both so that people who regularly participate in Lagrange Points can be more confident in how to estimate appropriate reverse komi between them and another person on OGS who hasn’t played in Lagrange Points before, and so that we can take advantage of all the work which has been put into the OGS rating system instead of reinventing the wheel which, in retrospect, was a very silly idea indeed.

So it’s back to calculating directly from OGS ratings, but with an offset which only changes by 1 point for every won or lost game. So if you have a rating offset of 5 points, your reverse komi will be calculated as if you are 5 points stronger than your OGS rank indicates, regardless of how much stronger that OGS rank has become.

This is essentially a more straight adaption of the Japanese Points Rating system, but tied to the OGS ranks.

This will change very slowly, so from now on (or until further notice), players may request prior to the beginning of any round (when you’re registering is a good time) a rating offset adjustment. Suggest either a relative (I think I’m a stone stronger than my rating is suggesting) or absolute (I think I’m about a stone weaker at reverse komi games than I am at even games) rating offset, and give your reasons why you think so. All adjustments subject to veto or counter-suggestion, of course.

If this system ends up sticking, it may eventually fall out of sync, with “0” rating offset being either stronger or weaker than the typical new player joining Lagrange Points. In that case I will add or subtract the same number of points from all players’ rating offsets, with a cap on how much an inactive player’s rating offset can change with successive adjustments to prevent any unforeseen degenerate behavior in that case.

Hopefully this saner system proves more, well, sane. Thank you all for bearing with me during these wrong turns.

2 Likes

Round 3 Pairings

Results of Previous Round

Samraku: "397(5) / 2k" {2-1-0}
teapoweredrobot: "364(0) / 4k" {2-0-0}
pdg137: "321(-1) / 8k" {0-1-0}
wurfmau3: "463(1) / 4d" {0-2-0}

Ratings are back to normal, hopefully for good. If you feel you are over or under rated, feel free to contact me with reasoning to request a rank adjustment . Requested rank adjustments are performed between rounds, so these numbers will not reflect adjustments requested prior to the start of this round, but the pairings and reverse komi of this round do reflect these adjustments.

The first number is your rating: the sum of a direct conversion from OGS rank and your current offset, which is noted in parentheses. It is the offset which is adjusted by wins, losses, and manual adjustments. The numbers in braces are the {W-L-D} record from that round.

Program Updates

Pretty quiet here; just commented out a bunch of useless rating code and replaced it with the much simpler system here. I feel like git should be crying with all the commented out functions laying about haphazardly…

Wow - just had an amazing game with @vyzhael. Thanks again @Samraku for this tourney!

1 Like

Better Late than Never?

Sorry for not posting this earlier, but registration for Round 4 is open.

For future reference, while each round cannot start until I post the pairings, for obvious reasons, registration opens based on the calendar, so if I forget to post a notice, feel free to send me your registration anyway, and as long as everything else is in order with it, it will be accepted.

1 Like

I’m afraid in going to be busy so will have to miss this round.

1 Like