Komi Statistics
DanGo
message correctly.
These figures relate to the games in the Jan. 2002 release of the
GoGoD database. Whether the figures indeed suggest that a 6.5 (or even
7.5) komi would be more appropriate, I leave to the readers.
(A) Komi [5.5] 7000+ games, excluding some 300+ games with 5 point
komi. These are mostly games from top tournaments and matches (Japan,
Korea).
period | B+ % | B+0.5% | What if komi 6.5
---------------------------------------------
1941-1979 | 53.1% | 2.7% | 50.4%
1980-1984 | 53.4% | 2.4% | 51.0%
1985-1989 | 51.9% | 2.7% | 49.2%
1990-1994 | 53.8% | 2.9% | 50.9%
1995-1999 | 54.0% | 3.1% | 50.9%
2000-2001 | 57.1% | 4.9% | 52.9% (*)
total | 53.6% | 2.9% | 50.7%
(*) This sample might be too small (450) for a reliable estimate.
(B) Komi [6.5] 590 games almost all after 1998, Korean.
period | B+ % | W+0.5% | What if komi 5.5
---------------------------------------------
1998-2001 | 51.7% | 3.0% | 54.7%
(C) Komi [2.75] 688 games. Chinese rules.
period | B+ % | B+0.75% | What if komi 3.25
---------------------------------------------
1980-2001 | 54.3% | 6.4% | 47.9%
Jan van Rongen
s/NOSPAM/xs4all/
PS searches performed using Kombilo0.2.
Robert Jasiek
DanGo wrote:
> period | B+ % | B+0.5% | What if komi 6.5
> ---------------------------------------------
> 1941-1979 | 53.1% | 2.7% | 50.4%
> 1980-1984 | 53.4% | 2.4% | 51.0%
> 1985-1989 | 51.9% | 2.7% | 49.2%
> 1990-1994 | 53.8% | 2.9% | 50.9%
> 1995-1999 | 54.0% | 3.1% | 50.9%
> 2000-2001 | 57.1% | 4.9% | 52.9% (*)
> total | 53.6% | 2.9% | 50.7%
The right column presumes that no player changes his strategy (even not
subconsciously) independendly of knowing at the game start whether the
komi is 5.5 or 6.5. This is very unrealistic.
The variance in the second column might indicate that the sample is too
small. Furthermore, it is unclear why periods of greatly different length
can be compared equally.
> (B) Komi [6.5] 590 games almost all after 1998, Korean.
>
> period | B+ % | W+0.5% | What if komi 5.5
> ---------------------------------------------
> 1998-2001 | 51.7% | 3.0% | 54.7%
Same problem.
> (C) Komi [2.75] 688 games. Chinese rules.
>
> period | B+ % | B+0.75% | What if komi 3.25
> ---------------------------------------------
> 1980-2001 | 54.3% | 6.4% | 47.9%
Same problem.
--
robert jasiek
DanGo
wrote:
>
[]
>The right column presumes that no player changes his strategy (even not
>subconsciously) independendly of knowing at the game start whether the
>komi is 5.5 or 6.5. This is very unrealistic.
I also wrote that it is up to the reader to draw his own conclusions.
But the samples are not too small for the periods 1980-1999. On
average 300 games per year is not a small sample, no matter what the
size of the population is. It is not a random sample though, they are
the top games.
I do not agree with your observation "very unrealistic". FI, in the
2001-06-18 game between Cho Hun-hyeong and Seo Pong-su (Wangwi
tournament) "the result was originally given as W+0.5. Seo had asked
the scorer during the game what the komi was and was told 6.5. But
this was incorrect and the result was officially reversed." (That
Korean tournament still had 5.5 komi).
Cheers
Jan van Rongen
s/NOSPAM/xs4all.nl
Robert Jasiek
DanGo wrote:
> I do not agree with your observation "very unrealistic".
So you think that many white players aim for a 0.5 loss when the komi is
6.5 instead of 5.5 and that they do not try to invest any extra efforts
for winning in view of such circumstances?! No, really, nobody could
believe that!
--
robert jasiek
John Fairbairn
news:3C52DFAA...@snafu.de...
>
> The right column presumes that no player changes his strategy (even not
> subconsciously) independendly of knowing at the game start whether the
> komi is 5.5 or 6.5. This is very unrealistic.
>
But you've already told us professionals are very weak and have low
strategic ability, so their attempts to change strategy are not likely to
have much effect.
I do find the approach of pulling the wings off flies and saying they can't
fly rather a sterile way of conducting a discussion. Or is it just me
getting old?
DanGo
wrote:
>
>
"Many White players" .... Oh, does white play first in Germany? Does
he/she give the komi? Yeah, nobody could believe that !
"So you think" ... read what I write, please don't try to guess what
I think. I gave an example of a pro who in the midst of a game didn't
know what the komi was. Seo was playing black and "officially
reversed" of course meant that it was changed from W+0.5 to B+0.5.
Cheers, Jan
Robert Jasiek
John Fairbairn wrote:
> But you've already told us professionals are very weak and have low
> strategic ability, so their attempts to change strategy are not likely to
> have much effect.
Professionals are weak at precisely incorporating a one point strategy in a
high temperature strategy during the middle game. They are strong at
calculating the late yose well. Therefore independently of their middle game
weakness they can not only have the desire to gain another point during the
endgame but change their strategy as a consequence, especially if this means
winning instead of losing. This is a very likely effect.
jan.van...@NOSPAM.nl (DanGo) wrote:
> "Many White players" .... Oh, does white play first in Germany? Does
> he/she give the komi? Yeah, nobody could believe that !
>
> "So you think" ... read what I write, please don't try to guess what
> I think. I gave an example of a pro who in the midst of a game didn't
> know what the komi was. Seo was playing black and "officially
> reversed" of course meant that it was changed from W+0.5 to B+0.5.
It is fruitless to discuss while missing each other's words:)
--
robert jasiek
BobbySixer
This is almost troll fodder
>
>
>
>
>
Robert Jasiek
BobbySixer wrote:
> >So you think that many white players aim for a 0.5 loss when the komi is
> >6.5 instead of 5.5 and that they do not try to invest any extra efforts
> >for winning in view of such circumstances?! No, really, nobody could
> >believe that!
> This is almost troll fodder
You don't agree? Here are arguments:
- There are professionals who change their playing style if komi is changed.
A modest estimate is at least 50%. E.g., almost no professional plays as if
there was 0 komi. The entire fuseki theory has changed to fit komi games.
- When a professional game is in the, say, 5-points endgame, then there is
comparatively little motivation to play so accurately that the result is a
0.5 points loss instead of a 1.5 points loss but every professional would be
motivated to turn a [territory scoring] 0.5 points loss into a 0.5 points win.
Let us modestly estimate that 50% professionals are more eager to play
precisely in the latter case.
Combining both possible reasons, a percentage significantly above 50% of all
professionals can be expected to apply at least either reasoning. If this is
not many professionals, then maybe you define "many" differently:)
--
robert jasiek
ro...@telus.net
<john...@harrowgo.demon.co.uk> wrote:
>"Robert Jasiek" <jas...@snafu.de> wrote in message
>news:3C52DFAA...@snafu.de...
>
>> The right column presumes that no player changes his strategy (even not
>> subconsciously) independendly of knowing at the game start whether the
>> komi is 5.5 or 6.5. This is very unrealistic.
>
>But you've already told us professionals are very weak
That is an outright falsehood. He said only that they were "too weak"
to do something that would require almost god-like strength.
>and have low
>strategic ability,
Likewise.
>I do find the approach of pulling the wings off flies and saying they can't
>fly rather a sterile way of conducting a discussion. Or is it just me
>getting old?
That's one possible explanation.
-- Roy L
Joel Olson
<<
The variance in the second column might indicate that the sample is too
small. Furthermore, it is unclear why periods of greatly different
length can be compared equally.
>>
<
But the samples are not too small for the periods 1980-1999. On average
300 games per year is not a small sample, no matter what the size of the
population is. It is not a random sample though, they are the top games.
>
Perhaps the way to look at the first three columns is as experimental
results. They're certainly not drawn at random from some large
population. Rather they capture nearly all of the group to which any
others aspire.
In this sense, sample size would only be an issue if some of the data
were in error (experimental error = miscounts, erroneous records, etc.).
The third column, is thus not "variance", but is a part of the emperical
results distribution, and could well be flanked by percentages of W
half-pointers, and both B & W 1.5 and 2.5 point differences.
The reason for looking at these is that after effects of the change in
komi have settled down, one would expect a fair komi to produce a
symmetrical distribution - roughly equal percentages on either side. Any
skewness in this distribution would be an opportunity for explication.
As Robert pointed out, W has no exterior reason to minimize an accepted
loss. Nor has Black to maximize a sure win.
There are 2 other eras that could be similarly examined: the change from
no komi to 4.5, and the change from 4.5 to 5.5.
Bill Spight
> There are 2 other eras that could be similarly examined: the change from
> no komi to 4.5, and the change from 4.5 to 5.5.
>
Actually, some statistics about komi were published in the AGA Journal
in the mid 70s (Summer '77, I think). As I recall, for both 4.5 komi and
5.5 komi the median result among Japanese pros was that Black was ahead
on the board by 7 points, and a komi of 6.5 would have given the closest
division of results, slightly favoring Black.
Best,
Bill
Joel Olson
<
Actually, some statistics about komi were published in the AGA Journal
in the mid 70s (Summer '77, I think). As I recall, for both 4.5 komi and
5.5 komi the median result among Japanese pros was that Black was ahead
on the board by 7 points, and a komi of 6.5 would have given the closest
division of results, slightly favoring Black.
>
Let me first say that my response above was made before reading the "New
Komi" thread. I apologize to any posters who feel slighted by the
remarks on experimental error, and well recognize the difficulty in
collecting accurate data. My own inclination is to think primarily in
terms of Japanese scoring.
Summer '77: By luck I have it!
<<
The Value of the First Move by Roger Clegg
In January 1975, the AGJ published an article of mine which argued
that the traditional value of 5 for the first move was too low. Recently
I acquired a large enough random sample to solve the problem
statistically. Games selected by editors can't be used as there is a
tendency to choose the more exciting ones ending in resignations. My
sample consists of 1) 358 final games of tournaments (excluding TV) from
the 1975 Kido Yearbook (to page 181) and the 1976 Kido Yearbook (to page
178); unlike previous yearbooks, the inclusion of preliminary games is
systematic; and 2) the complete professional games of 1961 and 1962,
excluding no-komi and TV games. I adjusted the results to 5 1/2 komi
when necessary and marked a 1/2 point win for Black with 4 1/2 komi as
1/2 point win for white. The total sample is 2404 games, of which white
won 1090 and black won 1314, or 54.66%. The 95% confidence range is
52.6% to 56.7%.
Adjusting komi, white won by 5 1/2 in 31 games, 4 1/2 in 53, 3 1/2 in
56, 2 1/2 in 80, 1 1/2 in in 67, 1/2 in 83. Black won by 1/2 in 74,
1 1/2 in 67, 2 1/2 in 67, 3 1/2 in 68, 4 1/2 in 42, and 5 1/2 in 28.
The ends of the distribution drop off rapidly because of resignations;
and statistical fluctiations make it difficult to see the peak; but it
is between 70 and 80. If it is 75, then a change of 1 point in the value
ofkomi would produce a change of 3.1% in the percentage of games won by
black. Applying this to the figure of 43.66% gives a value for the first
move of 7.0. A peak of 80 gives 6.9 and a peak of 70 gives 7.1. Applying
the earlier 95% confidence limits gives a range of 6.3 to 7.8.
In any game situation there is at least one sequence which is best for
both sides; and the question is, if such sequences are followed
throughout the game, how many points does black win by on the board? The
value of the first move has to be a whole number, and it is plain as a
pikestaff that it is 7. Why has nobody calculated this before; or, if
they have, why haven't they told anyone else about it? Professionals
certainly don't seem to know. All we ever get from "Go Review" are
remarks like, "Of course it is very difficult to know the value of the
first move." It is only two years since the last 4 1/2 komi tournaments
switched to 5 1/2, and a further change to 6 1/2 is obviously required.
If you rearrange the equation, "One handicap stone plus white first
move equals black first move;" you see that a handicap stone has twice
the value of the first move. Therefore a two stone handicap is worth 21
points, a three stone handicap is 35, and a four stone handicap is 49 or
50. The curve may start increasing at this point, as white can no longer
take a corner and the argument breaks down. These figures fit well with
the results of professional handicap games.
Editor's note: Bill Spight of Alcalde, NM pointed out one of the
difficulties of using a statistical analysis to establish the value of
the first move from professional games. He wrote, "Having a komi affects
the play. Before komi it was believed that in a game between evenly
matched contestants, Black, by playing solidly and conservatively could
win by three points or so. The onus was on White to play aggressively
and seek complications.
With a komi (of 4 1/2) the roles were reversed: White played a waiting
game, while Black played enterprisingly. Black's advantage was so great
that 4 1/2 komi is a thing of the past. But Black is still considered
the one who has to show enterprise, not White.
Statistics indicate that 5 1/2 komi may not be enough, but a change in
attitude on the part of White might change everything."
>>
:-)
Bill Spight
Thanks for posting that article. Nostalgia time. ;-)
Roger Clegg:
> I adjusted the results to 5 1/2 komi
> when necessary and marked a 1/2 point win for Black with 4 1/2 komi as
> 1/2 point win for white. The total sample is 2404 games, of which white
> won 1090 and black won 1314, or 54.66%. The 95% confidence range is
> 52.6% to 56.7%.
As I was interested in the effect of changing the komi, I looked at the
raw data and did *not* combine the games played under different komis.
Then I found, to my surprise, that the komi change had made no
difference to the median result on the board.
Best,
Bill
DanGo
wrote:
[]
Very interesting stuff, Joel! thanks for the effort to post this.
In between the lines I can read that Roger Clegg had access to the
complete professional games from 1961 and 1962 (maybe not the full
records, but at least the detailed results). Probably Nihon Ki-in
only. Does anybody have an idea where they came from and where they
are now?
Cheers
Jan van Rongen
s/NOSPAM/xs4all/
Lord Tim Brent
said:
>"Many White players" .... Oh, does white play first in Germany? Does
>he/she give the komi? Yeah, nobody could believe that !
>
>"So you think" ... read what I write, please don't try to guess what
>I think. I gave an example of a pro who in the midst of a game didn't
>know what the komi was. Seo was playing black and "officially
>reversed" of course meant that it was changed from W+0.5 to B+0.5.
>
>Cheers, Jan
Well,if Seo didn't know what the komi was,he deserved it....
Tim
------
Duchy of Grand Fenwick
Thinking is the most unhealthy thing in the world,
and people die of it just as they die of any other
disease. Fortunately, in England at any rate,
thought is not catching.
Oscar Wilde, The Decay of Lying, 1889
Lord Tim Brent
said:
>Just to complete the picture. Pls. Use a fixed width font to see this
>message correctly.
>
>These figures relate to the games in the Jan. 2002 release of the
>GoGoD database. Whether the figures indeed suggest that a 6.5 (or even
>7.5) komi would be more appropriate, I leave to the readers.
>
>(A) Komi [5.5] 7000+ games, excluding some 300+ games with 5 point
>komi. These are mostly games from top tournaments and matches (Japan,
>Korea).
>
> period | B+ % | B+0.5% | What if komi 6.5
>---------------------------------------------
>1941-1979 | 53.1% | 2.7% | 50.4%
>1980-1984 | 53.4% | 2.4% | 51.0%
>1985-1989 | 51.9% | 2.7% | 49.2%
>1990-1994 | 53.8% | 2.9% | 50.9%
>1995-1999 | 54.0% | 3.1% | 50.9%
>2000-2001 | 57.1% | 4.9% | 52.9% (*)
> total | 53.6% | 2.9% | 50.7%
>
>Jan van Rongen
>s/NOSPAM/xs4all/
>
>PS searches performed using Kombilo0.2.
All nice arguments on increasing komi.
Did you consider that maybe it is more the quality of the players than
the komi?
Looking at the overall (54-46 Black),it is not as great IMO an
advantage. After all,a 9-Dan match,you can not say the two players are
100% equal. Since Game 1 is the only one in which colours are drawn,
it is possib;le for the weaker player to take White 4/7 games.
From what I understand,pre-komi,if player a won x percentage of
games,the rule was player B got x 2/3 games.
I am sure in other games the situation is similar,if not greater
advantage for first move than Go.
You can never truly make it 50/50 split. Besides,many players play
better probably with Black than White.
Joel Olson
<<
Roger Clegg:
My sample consists of 1) 358 final games of tournaments (excluding TV)
from the 1975 Kido Yearbook (to page 181) and the 1976 Kido Yearbook (to
page 178); unlike previous yearbooks, the inclusion of preliminary games
is systematic; and 2) the complete professional games of 1961 and 1962,
excluding no-komi and TV games.
point win for Black with 4 1/2 komi as 1/2 point win for white. The
total sample is 2404 games, of which white won 1090 and black won 1314,
or 54.66%. The 95% confidence range is 52.6% to 56.7%.
>>
<
As I was interested in the effect of changing the komi, I looked at the
raw data and did *not* combine the games played under different komis.
Then I found, to my surprise, that the komi change had made no
difference to the median result on the board.
>
The Honinbo Tournament went to the 4.5 point komi in the 1940s; and to
5.5 points in 1975, much later than the other tournaments. So both the
1961-62 games and the 1975-6 games are observations from periods when
komi had been in use for some time.
Given a set of numbers, it is easy to apply the computations for mean,
median, variance, confidence range, etc., and get numerical results. The
validity of the interpretaion of these results, however, rests on the
data having properties assumed by statistical theory, typically that
they are a random sample of a larger population. When the assumptions
are not met, alternative procedures and/or tempered interpretations of
the results are needed. It is good practice to specify one's assumptions
and address these issues at the outset.
In particular, combining numbers from different time periods implies an
assumption that the data are independent of time, i.e. static.
This static model of the games results is a severe abstraction of
reality.
but may justify Clegg's use of confidence numbers. I also think it is
necessary, but perhaps not sufficient to justify adjusting the results.
In other words, for these manipulations to be valid, one must assume
that strategy is fixed.
I think most people would allow the static assumption for a single game,
being willing to consider a player's 'strategy' as broad enough to
contain adaptions to the changing board position.
As a cautious step toward a more realistic model of the game data, the
question is: how short a time period is required for the static model to
hold? How fast is technique/strategy evolving?