It is currently Thu Mar 28, 2024 3:31 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 17 posts ] 
Author Message
Offline
 Post subject: Komi of Internet 13x13 Amateur World Go Championship
Post #1 Posted: Tue Apr 14, 2015 5:00 am 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
http://www.pandanet.co.jp/event/13ro_to ... ndex_e.htm

The handicap system (-3 komi per 1/2 rank difference) is designed to let the weakest player win but the even game 6.5 komi is really dubious. It should be 8.5. 6.5 looks like "We use it 19x19, so let's try it for 13x13, too.". Komi should have justification. I see none for 6.5 for 13x13. 8.5 has the good justification of many close games in European 13x13 Championships.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #2 Posted: Tue Apr 14, 2015 5:19 am 
Lives in gote

Posts: 309
Liked others: 3
Was liked: 41
Rank: 5 dan
Do you have statistics to support your claim?

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #3 Posted: Tue Apr 14, 2015 5:30 am 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
RobertJasiek wrote:
8.5 has the good justification of many close games in European 13x13 Championships.


Do you have statistics for the score difference on 13x13 at the EGC?

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #4 Posted: Tue Apr 14, 2015 6:41 am 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
I do not have formal statistics but from 1993 to 2014 I attended each European (Congress) 13x13 Championship and played / watched preliminaries and finals to the final games. Almost all years used 8.5 komi. In evenly ranked games (kibitzed or my own), 0.5 scores have been frequent and, to the best of my memory, the mean and peak of close scores (neither Black nor White was preferred).

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #5 Posted: Tue Apr 14, 2015 7:19 am 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
How about the claim that 6 points per rank (3 points per half rank) is to the advantage of the weaker player? Do you have numbers for that?

The EGC uses roughly 3.5 points per rank and sees white win much more often, so that would be too little. Also given that a player gets roughly 14 points per rank on 19x19, 6 doesn't sound so bad on 13x13.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #6 Posted: Tue Apr 14, 2015 8:24 am 
Oza
User avatar

Posts: 2777
Location: Seattle, WA
Liked others: 251
Was liked: 549
KGS: oren
Tygem: oren740, orenl
IGS: oren
Wbaduk: oren
6.5 has been fairly commonly used lately in Japan for the 13x13 crowdfunding event and the computer-human matches. No one has seemed unhappy with it.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #7 Posted: Tue Apr 14, 2015 9:15 am 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
HermanHiddema wrote:
How about the claim that 6 points per rank (3 points per half rank) is to the advantage of the weaker player? Do you have numbers for that?


1) Typical per rank komi increments in 13x13 EGCs have been 2 or sometimes 3, and this has turned out to be reasonable. (Needless to say, weak players not thinking at all need more komi per rank. The 2 or 3 komi per rank is fair in matches in which both the stronger and the weaker players think well about their moves.)

2) An increment of 6 komi per rank on 13x13 very quickly leads to so many komi that the simplest opening strategies followed by dull, but blunder-free, play by Black are good enough to win. Any significantly weaker player should be able to win with "take territory early, then just connect your important stones and block your already existing territories" as his strategy.

Quote:
The EGC uses roughly 3.5 points per rank


We are not speaking about 19x19. 19x19 has 361 intersections, 13x13 has 169 intersections. To compare komi increments approximately, multiply by 169/361.

Quote:
and sees white win much more often,


Is this a 19x19 even game statistics for all ranks? If so, it shows that weak black players do not always know how to use their advantage of starting well. This says very little for 13x13.

Quote:
so that would be too little. Also given that a player gets roughly 14 points per rank on 19x19,


Rank-handicap-komi relations are not sufficiently reliable for high handicaps to make such statements, IMO.

Quote:
6 doesn't sound so bad on 13x13.


You miss the special properties of the 13x13 board: its size favours fighting / exchange strategies, which Black can initiate well, and he must if indeed the even game komi is 8.5. 19x19 strategy does not work well on 13x13.

oren wrote:
6.5 has been fairly commonly used lately in Japan for the 13x13 crowdfunding event and the computer-human matches. No one has seemed unhappy with it.


Interesting. Where can we read more about those events?

Now I would like to know whether, on average in such 6.5 komi 13x13 tournaments, Black did use his first move advantage fully by using fighting / exchange strategies or whether Black would play too cautiously as if it were a 19x19 game. The wrong komi can motivate the wrong strategies and thus close games with many 0.5 results for the wrong reasons. I am not conconvinced by such results until I see Black fighting / exchanging properly in many of the 13x13 games.

Have they also tried many 8.5 komi games? They should to find out the best komi. Otherwise, it is like decades of the wrong 5.5 komi on the 19x19 board: the wrong komi encourages the wrong strategies and we get misleading statistics.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #8 Posted: Tue Apr 14, 2015 9:56 am 
Oza

Posts: 2493
Location: DC
Liked others: 157
Was liked: 442
Universal go server handle: skydyr
Online playing schedule: When my wife is out.
RobertJasiek wrote:
We are not speaking about 19x19. 19x19 has 361 intersections, 13x13 has 169 intersections. To compare komi increments approximately, multiply by 169/361.


It seems problematic to assume that komi is proportional to board size, unless you think that correct komi for a 9x9 is about 1.5? If nothing else, I suspect that komi is also proportional to the number of moves to be expected in a game, since the first move out of 40 or 50 should be much more important proportionally than the first out of 250.

If you're referring to the idea that the change is komi for one (19x19) stone's difference in ranks should be set proportionally, this seems more reasonable, but it still breaks down for smaller solvable boards, I believe.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #9 Posted: Tue Apr 14, 2015 10:16 am 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
RobertJasiek wrote:
HermanHiddema wrote:
How about the claim that 6 points per rank (3 points per half rank) is to the advantage of the weaker player? Do you have numbers for that?


1) Typical per rank komi increments in 13x13 EGCs have been 2 or sometimes 3, and this has turned out to be reasonable. (Needless to say, weak players not thinking at all need more komi per rank. The 2 or 3 komi per rank is fair in matches in which both the stronger and the weaker players think well about their moves.)


According to sensei's library, it is more like 3.5

Quote:
2) An increment of 6 komi per rank on 13x13 very quickly leads to so many komi that the simplest opening strategies followed by dull, but blunder-free, play by Black are good enough to win. Any significantly weaker player should be able to win with "take territory early, then just connect your important stones and block your already existing territories" as his strategy.

Quote:
The EGC uses roughly 3.5 points per rank


We are not speaking about 19x19. 19x19 has 361 intersections, 13x13 has 169 intersections. To compare komi increments approximately, multiply by 169/361.


No, I am talking about 13x13. 19x19 uses 14 points per rank. Multiplying that by 169/361 suggests 6.5 for 13x13

Quote:
Quote:
and sees white win much more often,


Is this a 19x19 even game statistics for all ranks? If so, it shows that weak black players do not always know how to use their advantage of starting well. This says very little for 13x13.


No, it is 13x13. And black not knowing how to use their handicap/komi advantage is exactly why they get handicap/komi. They are weaker players.

There's a list of statistics from the 2010 EGC at http://senseis.xmp.net/?HandicapForSmal ... Sizes#toc5

Quote:
Quote:
so that would be too little. Also given that a player gets roughly 14 points per rank on 19x19,


Rank-handicap-komi relations are not sufficiently reliable for high handicaps to make such statements, IMO.



Not for high handicaps, no, but the first few stones will not be far off. And the maximum rank difference at this event will be around 7 stones in A class.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #10 Posted: Tue Apr 14, 2015 11:13 am 
Oza
User avatar

Posts: 2777
Location: Seattle, WA
Liked others: 251
Was liked: 549
KGS: oren
Tygem: oren740, orenl
IGS: oren
Wbaduk: oren
RobertJasiek wrote:
Have they also tried many 8.5 komi games? They should to find out the best komi. Otherwise, it is like decades of the wrong 5.5 komi on the 19x19 board: the wrong komi encourages the wrong strategies and we get misleading statistics.


No. They're playing with 6.5 now. Maybe in the future if black can win significantly more, it would change.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #11 Posted: Tue Apr 14, 2015 10:49 pm 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
HermanHiddema wrote:
According to sensei's library, it is more like 3.5


Which Sensei's pages, how do you conclude from them that they are saying exactly this?

Quote:
19x19 uses 14 points per rank. Multiplying that by 169/361 suggests 6.5 for 13x13


Exactly that's what I and skydyr and trying to say: one must not simply try to derive the komi of one board size to another board size because proportionality is not ensured.

Quote:
And black not knowing how to use their handicap/komi advantage is exactly why they get handicap/komi. They are weaker players.


You are not understanding what I have been writing: WRT to correct even game komi, I do not speak of the handicap system - I speak of the even game 13x13 komi! In even games, there is no weaker player in terms of ranks!

Quote:
There's a list of statistics from the 2010 EGC at http://senseis.xmp.net/?HandicapForSmal ... Sizes#toc5


These statistics say essentially nothing about even games 13x13 because these statistics are for all played games, including the handicap games.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #12 Posted: Tue Apr 14, 2015 11:29 pm 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
RobertJasiek wrote:
HermanHiddema wrote:
According to sensei's library, it is more like 3.5

Which Sensei's pages, how do you conclude from them that they are saying exactly this?

The same one I linked earlier, http://senseis.xmp.net/?HandicapForSmal ... Sizes#toc5

The system there gives 1 handicap every five ranks. If 8.5 is an accurate komi, then the value of the first move is around 17 and the first few handicap stones will have a similar value. 17/5 = 3.4 which is where I get the "roughly 3.5" from.

Quote:
Quote:
19x19 uses 14 points per rank. Multiplying that by 169/361 suggests 6.5 for 13x13

Exactly that's what I and skydyr and trying to say: one must not simply try to derive the komi of one board size to another board size because proportionality is not ensured.

It seems you are confused between "what is appropriate even game komi for 13x13" and "what is an appropriate komi increment per rank for 13x13 handicaps"

You said: "We are not speaking about 19x19. 19x19 has 361 intersections, 13x13 has 169 intersections. To compare komi increments approximately, multiply by 169/361.". So it was you who suggested this method for calculating komi increments. I simply applied it and got a value of 6.5. Note that this is not "even game komi 6.5" but "komi increment per rank 6.5"

Quote:
Quote:
And black not knowing how to use their handicap/komi advantage is exactly why they get handicap/komi. They are weaker players.

You are not understanding what I have been writing: WRT to correct even game komi, I do not speak of the handicap system - I speak of the even game 13x13 komi! In even games, there is no weaker player in terms of ranks!

Quote:
There's a list of statistics from the 2010 EGC at http://senseis.xmp.net/?HandicapForSmal ... Sizes#toc5

These statistics say essentially nothing about even games 13x13 because these statistics are for all played games, including the handicap games.

Again the above two points seem to be confused on the topic, which is "komi increments in handicap games", not "correct komi in even games".

Those statistics show that the EGC komi increment of roughly 3.5 per rank is insufficient to give the weaker players an even chance of winning games.

It was your assertion that 6 points per rank (3 per half rank) would give the weaker player a better than even chance to win. I was asking if you have any numbers to back up that assertion. The only numbers I know of (which I posted) suggest that at least 4 points per rank is required.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #13 Posted: Wed Apr 15, 2015 10:34 pm 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
HermanHiddema wrote:
If 8.5 is an accurate komi, then the value of the first move is around 17


Ok, where you mean the miai value.

Quote:
and the first few handicap stones will have a similar value. 17/5 = 3.4 which is where I get the "roughly 3.5" from.


Now I understand what you are referring to: you divide by 5 because the referred handicap table adds another handicap stone every 5 ranks. Ok, this makes sense numerically.

Quote:
Those statistics show that the EGC komi increment of roughly 3.5 per rank is insufficient to give the weaker players an even chance of winning games.

It was your assertion that 6 points per rank (3 per half rank) would give the weaker player a better than even chance to win. I was asking if you have any numbers to back up that assertion. The only numbers I know of (which I posted) suggest that at least 4 points per rank is required.


1) I do not have statistics - I have only much kibitzing and playing.

2) The aforementioned numerical relation does not work well in practice.

3) I have observed when kibitzing and playing 13x13 the following (and similarly 9x9): there are two types of weaker players: a) those applying basic go theory well and having ca. 50% winning chance under the handicap system 1 stone per 5 ranks and else 2 komi decrement per rank versus b) those not applying basic go theory well and needing some greater or much greater handicap to have ca. 50% winning chance. This has consequences for handicap - winning statistics, which indifferently measure both (a) and (b) players: they measure how weak on average the weaker players play instead of measuring how strong the weaker players should play. This is so also because 13x13 tournaments are rare and (b) players get too little motivation to develop a reasonable 13x13 skill. (a) players get their motivation nevertheless from their pride to play well and seriously attempt to win. I recommend the aforementioned handicap system because I want to give roughly equal chances to stronger players and (a) players. A few others suggest greater handicaps because they want to tolerate the average of (a) and (b) players and do not care for equal winning chances for white top players, whose disadvantage is overlooked in the overall averaging statistics.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #14 Posted: Wed Apr 15, 2015 11:00 pm 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
@Robert: Ok, so your assertion is that the 6 pts komi increase per rank is perhaps not unfair on average for all players, but that there exist weaker players whose playing style is particularly suitable for defending an existing advantage and that such players would be advantaged by this system?

So you initial statement of "The handicap system (-3 komi per 1/2 rank difference) is designed to let the weakest player win..." is perhaps not entirely accurate in that respect. The weakest (or weakest few) players may not possess the appropriate playing style (or may lose to players who are only slightly stronger, because their initial advantage becomes insufficient).

Now an additional complication is, of course, that there is no handicap, only komi. Some players who are particularly skilled at defending a handicap advantage may not be equally skilled at defending a komi advantage.

I guess we'll see what happens. It will be interesting data.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #15 Posted: Wed Apr 15, 2015 11:46 pm 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
Something like this happened in the 1994 final game of the EGC, IIRC, 13x13 tournament, in which Black was 20k and won the game due to playing out a ladder worse than this example:

Click Here To Show Diagram Code
[go]$$B
$$ +---------------------------+
$$ | . . . . . . . . . . . . . |
$$ | . . . . O O . . . O O . . |
$$ | . . O O X X . . . X O . . |
$$ | . . X X . . X . . X X . . |
$$ | . . . . . . . . . . . . . |
$$ | . . . . O . . . . . . . . |
$$ | . . . X . 6 X . . X . . . |
$$ | . . . . 5 O 4 1 . . . . . |
$$ | . . . . . 3 2 O X X . X . |
$$ | . . . X . . X O O X X O . |
$$ | . . . . . . X . O X O . . |
$$ | . . . . . . . . . O . O . |
$$ | . . . . . . . . . . . . . |
$$ +---------------------------+[/go]


The only reason for the 20k winning the tournament was, OC, a too great handicap.

EDITED

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #16 Posted: Thu Apr 16, 2015 8:15 am 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
I don't see why it is a problem that a 20k won the 1994 EGC 13x13 tournament?

If the handicaps are truly equalizing, then we would expect a 20k to have as much chance of winning it as a 7d, and we would see a 20k win the event roughly once every 30 years or so.

We would also see a representative sample of the playing population reach later rounds.

So, for example, if we look who qualified for the 2010 final 16 KO (http://www.egc2010.fi/results/13x13-finals.pdf), we can see that the handicap system was apparently very much to the advantage of the stronger players. Only one kyu player reached the last 16, and only one player weaker than 5 dan reached the last 8.

On the other hand, if we look at the 2014 results (http://www.egc2014.com/rezultate/1313.htm) we see that 6 kyu players, ranging from 2 kyu to 12 kyu, reached the last 16. So did 10 dan players, ranging from 2 dan to 6 dan. Since 3 qualified dan players failed to show up, sadly, only 7 dan players actually participated, which gave one of the kyu players two byes in a row, after which he won the tournament. In the last 8 it was 4 kyu players vs 3 dan players, in the last 4 it was 3 kyu and 1 dan player, and the final was 2 kyu players against each other.

I don't know what handicap system 2014 used, but the result was much closer to equalizing all players than 2010.

Now of course one can argue about whether one finds it desirable to truly equalize all the players or not, but if that is your set goal, then in 2010 the system clearly failed, and in 2014 the system got quite a reasonable result.

Top
 Profile  
 
Offline
 Post subject: Re: Komi of Internet 13x13 Amateur World Go Championship
Post #17 Posted: Thu Apr 16, 2015 10:16 pm 
Judan

Posts: 6087
Liked others: 0
Was liked: 786
The problem is not that a 20k won the tournament but that the handicap was so favourable for 20k players that no-thinking strategies and tactics sufficed to win the tournament. Even a 20k should be able to consider more than one next move, but, in the type of sequence shown, she did not.

For EGCs with a handicap system 1 stone per 4 or 5 ranks, ca. 2 points komi decrement per rank, there have been years in which mostly dans made it to the KOs and other years in which a good mix of dans and kyus qualified including cases in which also kyus proceeded to the later rounds. With 2010, you just happen to mention a year in which the dans were particularly successful.

ATM, I do not recall the exact handicap system of 2014. IIRC, it was a tough one for the strong dans, for whom it was a lottery whether they were in groups with kyus mostly of type (a) or (b). I saw very different play quality by (a) versus (b) players, and accidentally the groups were very differently mixed - some being (a)-heavy, others being (b)-heavy. The KO table shows the result of that group distribution.

7D in such tournaments tend to be opponent-killers regardless of the handicap system; I think that no system can reflect well the possible range of playing strengths summerised under the "7d" tag. We saw this in 2010; and Silt 6d (the only non-7d in the semi-finals) is one of the stronger 6d players. In 2010, I lost to Ilya Shikshin in the quarter finals because he played much better in that games; in the 2009 final game I lost against him with 4 points because I made one significant endgame mistake, which I should have avoided, so, again, I think that the mentioned handicap system was fair. It is just that the weaker player needs to be even more careful than usual to avoid (also smaller) blunders: in 2005, Cho Seok-bin 7d threw me out when I overlooked a non-standard large scale killing move which I should have been able to discover and predict. In the 2012 final game against a 7d, I did not blunder but played suboptimal endgame during the opening moves. I do not want a greater handicap for me - I want myself to play better in order to deserve beating the 7d. The mentioned handicap system is good enough for that.

With reasonable handicaps, there is an occasional 10 to 13k in the finals but 14 - 30k are under-represented. IMO, this is so not because the handicaps would be unfair but because there are too many (b) players among the weaker DDK players.

My aim would be to give type (a) players reasonably equal chances. One sees that almost all strong dans are type (a) players while, say, only rougly 50% of the SDK players are. Therefore, with a type (a) handicap, there can be more dans than kyus in the finals if the preliminaries had equally many dans and kyus. This varies from year to year. The congress mean is, I think around 1k. In 13x13 (or other side) tournaments, it can happen that already the preliminaries have more dan than kyu participants. Besides, it is more likely that qualified kyus fail to participate in the KO and and lose their first KO game.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 17 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group