EGC 2024 13x13 Komi

[RobertJasiek]

For some side tournaments of the EGC 2024, the organisers claimed komi and handicap settings to follow those of earlier years but they misinformed us. In particular, during most previous 13x13 tournaments at EGCs, the even game komi was 8.5. This year under Japanese Rules, the komi was 6.5 in even games and in 1 handicap games. During the group phase, most games were handicapped. During the KO finals, I watched in particular these three even games each between two 7D players:

Quarter final: Black wins by 2.5.

Semi-final: Black wins by 1.5.

Final: Black wins by 5.5. (After the loser lost two groups.)

Therefore, again this year, evidence supports that the most suitable 13x13 komi in even games is 8.5, as was used, in particular, by Anton Steininger or me as 13x13 tournament directors in several previous years but also in further years when my recommendation was applied.

This year's tournament organisers thought that they would know it better but they did not. 13x13 even game komi must not be just copied from 19x19 komi.

[jlt]

https://lifein19x19.com/viewtopic.php?f=15&t=17750

Katago thinks that 6.5 is fairer than 8.5.

[Javaness2]

Pachi showed the same some time ago, that 6.5 is just fine as komi. Surprising for me, but there you go.

Given that the 13x13 is not a top event, https://www.eurogofed.org/egf/topevents.htm , I think that the EGF doesn't care. It doesn't have any specific rules for 13x13. Even for top events, the EGF doesn't particularly care about how they are organised.

[RobertJasiek]

With shallow searches (lightvector 100ks) and without manual iteration on searches but with fluctations, AI results cannot be trusted on boards with tactical complications. The small samples of human game scores have a low confidence, too. So it remains an open question. Currently, it is a matter of preference whether to approach humans or AIs. Just one thing is clear: maively copying the 19x19 komi lacks justification.

[kvasir]

I don't think it really matter what KataGo thinks the komi should be. Or Pachi and GnuGo.

If I understand correctly then part of the issue is that the tournament appears to have been held with different komi than was communicated earlier. That is problematic.

Now, I don't know what the circumstances were or what was communicated but it does seem important to communicate any changes as early as possible. Of course it is easier to understand some last minute changes than other. I can see how the time control could be changed when enough of the right type of clocks can't be found and how a minor change to the schedule can be due to some outside factors. That is far from ideal but it is harder to imagine how the komi can change if it was explicitly stated before.

I remember I played in the 13x13 once and it was very chaotic and I had no idea what was happening.

Btw is there a record of games played in 13x13 tournaments at EGCs in the past, including the handicap, komi and result of each game?

[jlt]

Anyway 13x13 tournaments are not even EGF-rated. The 2024 tournament results are here https://egc2024.org/fr/events.html#_13%C3%9713 (click on "standings"), however the table of "selection standings" is incomplete: there were 16 groups of 7 people, and a Round Robin was organized in each group. In addition many players forfeited their games in that phase.

[RobertJasiek]

is there a record of games played in 13x13 tournaments at EGCs in the past, including the handicap, komi and result of each game?

Only of a very few selected games, typically the final game in a few of the years. E.g., in 2000, I scribed the final game and I think it is in one of the congress bulletins of that congress.

[kvasir]

I took a look at the tournament standings.

The following was immediately clear:

• Black wasn't winning many games, at least not if the expectation was a 50% share.
• There was a final and a preliminary but only 5 rounds from the preliminary were posted on the webpage. So maybe there were two days of preliminaries and only one day was posted, or the extra rounds (if any) are missing for some other reason.
• The handicaps stated didn't appear consistent, sometimes they were the grade difference and other times they appeared to be the number of handicap stones.
• There weren't that many games with large rank difference in the final and white won all of those games.

So I just assumed that it would more useful to use the rank difference. Hopefully this is what was used to look up the handicap in the table that was posted on the webpage. If not then I don't know what to do.

I made the following calculations of black's win rate with different handicap. The confidence intervals (CI95) are binomial 95% confidence intervals using Wilson's score intervals with continuity correction (hopefully this is appropriate).

Code: Select all

Even games (rank difference < 2):
Black wins:   15  White wins:   10
Black win rate:  0.600  CI95:  0.389  -  0.782

Low handicap games (rank difference 2 - 4):
Black wins:   14  White wins:   32
Low handicap games:  0.304  CI95:  0.182  -  0.459

High handicap games (rank difference 5 - 9):
Black wins:   25  White wins:   73
Black win rate:  0.255  CI95:  0.175  -  0.355

Very high handicap games (rank difference > 9):
Black wins:   20  White wins:  127
Black win rate:  0.136  CI95:  0.087  -  0.205

All handicap games:
Black wins:   56  White wins:  219
Black win rate:  0.204  CI95:  0.159  -  0.257

All games:
Black wins:   71  White wins:  229
Black win rate:  0.237  CI95:  0.191  -  0.290

Now, I don't know if the intention was to give white an advantage or not. My opinion is that if white should have an advantage then that should be as close to a constant advantage as possible. In this case it appears that white advantage increased with larger rank difference.

[RobertJasiek]

In handicap games, these types of handicaps are fair in those games in which Black knows to use the handicap well. However, many players enter such tournaments with little or no idea how. If you want to reward lack of preparation etc., make the handicaps more favourable for Black. Those black players knowing how to use the handicap make winning very hard to impossible for White.

[kvasir]

If it is that most players don't know how to make good use of the handicap, then shouldn't the deviation from even win rate be smaller for more handicap than it is with less? It is after all easier to make good use of say 4 stones on 13x13 than it is to make good use of marginally smaller komi.

If it is a matter of handicap skill then it could also be suggested that the win rate should favor black with the more extreme handicap. "How hard can it be to make good enough use of all those stones?" - is something someone might ask.

Anyway, I'll only suggest the not so novel idea that keeping and collecting records of the results of such tournaments would be very useful for finding answers to questions about komi and handicap. That probably won't answer questions about if white should have an advantage, however, it could be a good idea to know how much white's advantage is.

[gennan]

So it seems that the 13x13 handicap was roughly 1 full stone per 4 ranks? I'm not surprised that this handicap was disadvantageous for black. I think 13x13 handicap should be more like 1 full stone per 2 or 3 ranks.

I have been using 1 full stone per 2 ranks in my youth club in the past, but perhaps that is a bit too advantageous for black, so I moved to 1 full stone per 3 ranks. OGS may move to 1 full stone per 3 ranks later this year: https://forums.online-go.com/t/proposal ... p/50314/10.

[gennan]

Anyway 13x13 tournaments are not even EGF-rated.

This may change in the future. A few years ago the AGM of the EGF accepted a motion to include 13x13 and 9x9 in the rating system, I think to allow particularly youth and beginners to play rated tournaments on smaller boards.

[jlt]

I made the following calculations of black's win rate with different handicap.

The preliminaries were played over two days, I think that only games of the first day appear on the site. Moreover many people forfeited their games. It's possible that people who didn't show up tend to be weaker players, so this decreases Black's winrate further.

So it seems that the 13x13 handicap was roughly 1 full stone per 4 ranks?

The EGC website says

Code: Select all

Rank difference | Handicap stones |  Komi for white
0	0	6.5
1	0	6.5
2	0	4.5
3	0	2.5
4	0	0.5
5	2	6.5
6	2	4.5
7	2	2.5
8	2	0.5
9	3	6.5
10	3	4.5
11	3	2.5
12	3	0.5
13	4	6.5
14	4	4.5
15	4	2.5
16	4	0.5
17	5	6.5
18	5	4.5
19	5	2.5
20	5	0.5

so assuming fair komi is 6.5, so that 1 full stone would be equivalent to 13 points, this is roughly

Code: Select all

rank difference | number of full handicap stones
4n+1	n	
4n+2	n+0.15	
4n+3	n+0.31	
4n+4	n+0.46	

[Javaness2]

Is there a reason for not using the handicap system suggested by Tim Hunt here :
https://web.archive.org/web/20080112004 ... handicaps/

The EGF system is the one favoured by RJ , you can see some brief discussion about its validity here https://senseis.xmp.net/?HandicapForSmallerBoardSizes

If you care about the system to be used for the 13x13 side event at the EGC, and I suppose that amongst EGF members you would be in a minority if you did care, you should surely care about it for the right reasons. Wanting to use a system because it has been used N-times before is not at all good if you're repeating a mistake. Decide on the right system. Ask the EGF to protect the tournament - secure its organisation, secure its rules, its schedule, ... Right now I don't think you can really claim it to have an official status. I mean here https://egc2024.org/en/events.html#_13%C3%9713 it is not described as the European Championship in 13x13 go, it is just described as '13x13 tournament'. If you were part of the EGF rules committee, you could already have done this

By the way what does the highlighted part mean in the announcement?

Round robin system for the group stage and regular tournament tree for the knockout stage.

I haven't heard of a regular tournament tree before, it's probably something very straightforward, but what exactly does it describe?

[RobertJasiek]

A side tournament handicap system can, e.g., make it fair for players with equal determination to use their skill (Anton Steiniger / Robert Jasiek), favour the strong players (blitz with handicap minus 3), or favour the lazy players (Tim Hunt).

"Regular tournament tree": an arbitrary description for "KO tree".