Life In 19x19
http://lifein19x19.com/

Is Komi standard play?
http://lifein19x19.com/viewtopic.php?f=11&t=10387
Page 1 of 3

Author:  Joelnelsonb [ Mon Jun 02, 2014 5:03 am ]
Post subject:  Is Komi standard play?

I've seen some things that make Komi sound like it's not always accepted as equal. I've even heard people speak of an "even game" meaning no komi. So is komi not a standard for regular gameplay? I've also seen different komis used (other than 6.5) and I wonder how you go about deciding on which Komi to use? But really my question is: if two experienced players who considered each other equal rivals sat down to play a friendly, fair game, just in private, would they play with a Komi? If so, how much?

Author:  RBerenguel [ Mon Jun 02, 2014 5:10 am ]
Post subject:  Re: Is Komi standard play?

Joelnelsonb wrote:
I've seen some things that make Komi sound like it's not always accepted as equal. I've even heard people speak of an "even game" meaning no komi. So is komi not a standard for regular gameplay? I've also seen different komis used (other than 6.5) and I wonder how you go about deciding on which Komi to use? But really my question is: if two experienced players who considered each other equal rivals sat down to play a friendly, fair game, just in private, would they play with a Komi? If so, how much?


They would play with komi, but how much is subject to debate. 6.5 is more or less the standard now, and will probably be 7.5 sooner or later. It is adjusted depending on how often black wins, since black moves first. A good way to solve "how big a komi" is either komi bidding or "split the pie."

  • Bidding: Offer an amount of points for playing first (Auction komi)
  • Split the pie: a random player starts, white can swap if he prefers playing as black with that move as first (Pie Rule)

Author:  DrStraw [ Mon Jun 02, 2014 5:27 am ]
Post subject:  Re: Is Komi standard play?

Black has a decided advantage when there is no komi. That is the whole point of komi - to compensate white for this advantage. So technically there is no such thing as an even game without komi - it simply is not an even chance between two equal players. What most people refer to as an even game without komi is really the equivalent of a one stone handicap.

As to the question of how big the komi should be to compensation white, there is a lot of debate. When I first started playing it was 4.5 and the move to 5.5 was called the "big komi". Now it has moved to 6.5, and as RBerenguel says, it may soon change to 7.5. The truth is that until you get to the high dan level it is not going to make a lot of difference in most games because the overwhelming majority of amateur games are not that close.

The concept of 0.5 komi to white in handicap games is not the same idea, even though it has the same name. In this case it is used merely to prevent a jigo. Historically white was considered to have won a jigo for the simple reason that to advance you had to be able to beat stronger players - not tie with them. As an extension of this I would say don't worry about komi if playing over the board - just take the correct handicap and try to improve. Online just take the default komi for the server you are on. When you get to the dan level you can start thinking about it more but until then it is not going to affect your play.

Author:  Joelnelsonb [ Mon Jun 02, 2014 5:36 am ]
Post subject:  Re: Is Komi standard play?

Also, shouldn't it be adjusted for smaller games? Seems like 6.5 is a lot for a 9x9.

Author:  oca [ Mon Jun 02, 2014 5:51 am ]
Post subject:  Re: Is Komi standard play?

Joelnelsonb wrote:
Also, shouldn't it be adjusted for smaller games? Seems like 6.5 is a lot for a 9x9.

Here is a link on that subject :
HandicapForSmallerBoardSizes

Author:  Hushfield [ Mon Jun 02, 2014 5:53 am ]
Post subject:  Re: Is Komi standard play?

That's what I thought at first as well, but the advantage of having the first move on such a small board is much bigger than having the first move on a larger 19x19 board. Therefore black can get a lot more because of his first move, and white needs more compensation. 6.5 points appears to be just the right amount.

I'm not sure if this is "universally accepted" (there's bound to be some debate), but I think a large portion of the go playing community believes that 6.5 is a fair komi, no matter the size of the board (9x9, 13x13 or 19x19).

Author:  RBerenguel [ Mon Jun 02, 2014 6:29 am ]
Post subject:  Re: Is Komi standard play?

Hushfield wrote:
That's what I thought at first as well, but the advantage of having the first move on such a small board is much bigger than having the first move on a larger 19x19 board. Therefore black can get a lot more because of his first move, and white needs more compensation. 6.5 points appears to be just the right amount.

I'm not sure if this is "universally accepted" (there's bound to be some debate), but I think a large portion of the go playing community believes that 6.5 is a fair komi, no matter the size of the board (9x9, 13x13 or 19x19).


Personally I'd see 4.5 as much better komi for 9x9.

Author:  DrStraw [ Mon Jun 02, 2014 6:31 am ]
Post subject:  Re: Is Komi standard play?

I think that the smaller the board the bigger the advantage of first move and so the higher the komi should be. DGS currently has 7x7, 9x9, 13x13 and 19x19 tournaments going. The komi for the 7x7 is the highest at 8. Some people on the forums there are arguing that 9 would be better.

Author:  Bonobo [ Mon Jun 02, 2014 7:08 am ]
Post subject:  Re: Is Komi standard play?

See also http://senseis.xmp.net/?komi for more exhaustive information about komi.


Greetz, Tom

Author:  yoyoma [ Mon Jun 02, 2014 7:52 am ]
Post subject:  Re: Is Komi standard play?

People gave a lot of expert-level answers. But I want to emphasize the basic answer, because maybe it sounds like there is some sort of controversy.

When people say "even game", they mean using komi (6.5 for Japanese rules).

Author:  RBerenguel [ Mon Jun 02, 2014 7:57 am ]
Post subject:  Re: Is Komi standard play?

DrStraw wrote:
I think that the smaller the board the bigger the advantage of first move and so the higher the komi should be. DGS currently has 7x7, 9x9, 13x13 and 19x19 tournaments going. The komi for the 7x7 is the highest at 8. Some people on the forums there are arguing that 9 would be better.


As I said, it's my personal opinion. On the other hand, Go Seigen and Miyamoto Naoki played a couple of well-known 9x9 games to determine 9x9 komi (or so I've read here) and both were won by black by 4 points. So, 4.5 would seem to be the perfect komi in this case.

Going by the numbers, I have checked my 9x9 databases, which currently hold ~400 pro games and ~800 amateur (from OGS, where at least one of the players is better than 1 dan and the other at least 5k, IIRC) games. And probably 4.5 is overkill (of course, not taking into account how much 6.5 skews gameplay,) seeing the results, but 5.5 seems like somewhat better, at least for amateur play. For pro play... I'd bet for just too few games... And probably pie rule or bidding may be better, given the huge jump from 6.5 to 5.5

Checking only games where there is a definite result (so resignations or timeouts are not accounted) and komi is 6.5, and here's what changing the komi gives:

Amateur games (remember: filtered for komi=6.5 AND numeric result, no resignations or timeouts)

Code:
Totals (248 games)
Komi 6.5:     W: 55.6452    B: 44.3548
Komi 5.5:     W: 48.3871    B: 51.6129
Komi 4.5:     W: 34.6774    B: 65.3226

Pro games: (remember: filtered for komi=6.5 AND numeric result, no resignations or timeouts)

Code:
Totals (118 games)
Komi 6.5:     W: 56.7797    B: 43.2203
Komi 5.5:     W: 44.9153    B: 55.0847
Komi 4.5:     W: 37.2881    B: 62.7119

Here's the awk script I used, together with the invocation for it to work with sgfinfo:

Code:
# sgfinfo -propKM:6.5 -propRE *.sgf | LC_NUMERIC="C" gawk -f komi.awk
# sgfinfo -propKM:6.5 -propRE *.sgf | LC_NUMERIC="C" gawk -f komi.awk
# Remember: LC_NUMERIC="C" !!

BEGIN{
    black_n=0
    white_n=0
    black_4=0
    white_4=0
    black_5=0
    white_5=0
}

/+[0-9]\.[0-9]/{
    split($1, res, "+")
#    printf "DEBUG: Won:" res[1] " by " res[2] "\n"
    if (res[1]=="B"){black_n++}
    if (res[1]=="W"){white_n++}
    # Komi 5.5:
    if (res[1]=="W"){
   if (res[2]<1){
       black_5++
   }else{
       white_5++
   }
    }else{
   black_5++
    }
    if (res[1]=="W"){
   if (res[2]<2){
       black_4++
   }else{
       white_4++
   }
    }else{
   black_4++
    }
}

END{
    total=black_n+white_n
    printf "Totals (" total " games)\n"
    printf "Komi 6.5: \t W: " white_n/total*100 "\t B: " black_n/total*100 "\n"
#    printf "DEBUG: total_5: " black_5+white_5 "\n"
    printf "Komi 5.5: \t W: " white_5/total*100 "\t B: " black_5/total*100 "\n"
#    printf "DEBUG: total_4: " black_4+white_4 "\n"
    printf "Komi 4.5: \t W: " white_4/total*100 "\t B: " black_4/total*100 "\n"

}

Author:  HermanHiddema [ Mon Jun 02, 2014 8:06 am ]
Post subject:  Re: Is Komi standard play?

Why are you filtering out resignations? That's a perfectly valid result, which takes into account komi. Also, currently OGS always gives the stronger player white, and since stronger players generally resign on smaller margins, filtering out resignations may bias the result.

Author:  Dusk Eagle [ Mon Jun 02, 2014 8:07 am ]
Post subject:  Re: Is Komi standard play?

I don't think this analysis for 4.5 and 5.5 komi is valid unfortunately. Even at my pitiful level, I can accurately count a 9x9 game to within a point or two and adjust my play drastically if I'm going to lose. I also will have no need to play for a 1.5 win if I know that the simpler way of play gives me a 0.5 win.

If we want to determine what the win rate would be at 4.5 or 5.5 komi, I think the only valid way would be to generate a large sample of games played at 4.5 or 5.5 komi. Since computers are actually strong at 9x9 games now, doing a large number of computer playouts at these komis could be one way to do this. There may still be pitfalls to this approach but it would be interesting to see the results.

Author:  RBerenguel [ Mon Jun 02, 2014 8:16 am ]
Post subject:  Re: Is Komi standard play?

HermanHiddema wrote:
Why are you filtering out resignations? That's a perfectly valid result, which takes into account komi. Also, currently OGS always gives the stronger player white, and since stronger players generally resign on smaller margins, filtering out resignations may bias the result.


Re: OGS and white, I have asked for this to be fixed on the 9x9 (and hopefully all other) ladder games, at least.

I'm filtering out resignations to rule out "very early resign," where a big blunder happens before the game is "relatively" close. Also, there's no easy way I can count games where one party resigns. Was the game within 2 points when the player resigned? Was the game within 1 point? The only way to check would be to run pachi (or fuego) on these cases with a large number of playouts to rule out even more mistakes, and from personal experience pachi is somewhat dumb at the later stages of even 9x9 games.

As DuskEagle says, the best way would be to have a large pool of real games played with these settings, but 9x9 computer go is not strong and fast at the same time, sadly, specially if we expect very good level of play.

My experience in 9x9 go is that winning as black feels like a quite high mountain to climb, whereas as white it's somewhat more easy paced.

Author:  illluck [ Mon Jun 02, 2014 8:47 am ]
Post subject:  Re: Is Komi standard play?

I think you are overthinking it, RBerenguel. Including resignations is the simplest way to ensure there are no biases. Early resigns should cancel out, and there is no need to count the score - it's not meaningful for games that are scored either.

Author:  bayu [ Mon Jun 02, 2014 9:00 am ]
Post subject:  Re: Is Komi standard play?

RBerenguel wrote:
OGS and white, I have asked for this to be fixed on the 9x9 (and hopefully all other) ladder games, at least.


If there are a lot of 9x9-tourney and ladder games of shin-OGS in your database, than this bias makes a lot invalid: The stronger player is always white, so you'd expect white to win more often (with fair komi). I've seen people complaining about the high 9x9-komi in the chat. If they were always taking white against stronger players, they would complain that 6.5 points is never enough, I'm sure.

I don't know why the devs of shin-OGS had the idea to do that in tournament games and still stick to it. In ladder games it is kind of debatable whether the defender (not the stronger player) always has white. In the tournaments it doesn't make sense to me at all.


illluck wrote:
I think you are overthinking it, RBerenguel. Including resignations is the simplest way to ensure there are no biases. Early resigns should cancel out, and there is no need to count the score - it's not meaningful for games that are scored either.


The way I understand the setting, rberenguel took for the amateur games only 6.5-komi games and checked how many swings in results you get if you assumed that the komi were lower. So you have to get rid of resignations.

Author:  RBerenguel [ Mon Jun 02, 2014 9:11 am ]
Post subject:  Re: Is Komi standard play?

Probably guilty of overthinking (wouldn't be the first time). okay, ran again without filtering resignations (in the AWK file just remove the regexp at the beginning of the main clause)

Amateur games:
Code:
Totals (875 games)
Komi 6.5:     W: 51.6571    B: 48.3429
Komi 5.5:     W: 49.6    B: 50.4
Komi 4.5:     W: 45.7143    B: 54.2857


Pro games with 6.5 komi:
Code:
Totals (263 games)
Komi 6.5:     W: 52.4715    B: 47.5285
Komi 5.5:     W: 47.1483    B: 53.6122
Komi 4.5:     W: 43.7262    B: 57.0342


Was tempted to check 5.5 komi games (which are roughly 50% of the games I have) but it would mean rewriting how I test for changes in score and I'm a little too busy now to do it, maybe later.

Re: Bayu and swings: Indeed, this was the way I thought it. And yes, OGS games are rigged like this with the stronger player always getting white, so it is to be expected that he wins more often. I have asked to randomise nigiri in ladder games: if games are no handicap, why should they be always stronger player has white? I'd prefer full nigiri for the ladder. I think it makes more sense specially in 9x9, because even if the komi is large enough or small enough, the way you approach the game as B or as W are very different, and being on top of the ladder should account for both.

Author:  Uberdude [ Mon Jun 02, 2014 9:14 am ]
Post subject:  Re: Is Komi standard play?

Being the highest ranked (ex)player on OGS I would dislike never being able to play a black fuseki!

Author:  illluck [ Mon Jun 02, 2014 9:34 am ]
Post subject:  Re: Is Komi standard play?

You can't try to adjust komi after the game - players will play differently with different komi, especially on 9x9 when it's so easy to count.

Author:  leichtloeslich [ Mon Jun 02, 2014 10:17 am ]
Post subject:  Re: Is Komi standard play?

HermanHiddema wrote:
currently OGS always gives the stronger player white

Wow, that's seriously braindead and entirely defeats the purpose of giving white komi.
I hope it's a bug and not a feature.

illluck wrote:
You can't try to adjust komi after the game - players will play differently with different komi

I'm amazed this needed to be pointed out.

I regularly sacrifice a few points in 9x9 endgames to simplify the game.

Page 1 of 3 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/