adequate komi and computer go

[Detlef Müller]

Hi,

recently a go player suggested finding out the "adequate
komi" using a fairly strong computer program (i.e. playing
many games against itself with increasing komi until win rate
exceeds 50%).

Is this already explored by someone (and if so: does the
outcome differ much for different programs)?

greetings,

Detlef

[SP]

Post by Detlef Müller
Hi,
recently a go player suggested finding out the "adequate
komi" using a fairly strong computer program (i.e. playing
many games against itself with increasing komi until win rate
exceeds 50%).

Do you mean, until a certain komi results in a win rate
nearest to 50%?

Greetings,
Sergio

[Detlef Müller]

[...]

Post by SP

Post by Detlef Müller
recently a go player suggested finding out the "adequate
komi" using a fairly strong computer program (i.e. playing
many games against itself with increasing komi until win rate
exceeds 50%).

Do you mean, until a certain komi results in a win rate
nearest to 50%?

yes - probably there is no komi leading to exactly 50% win
rate.

greetings,
Detlef

[Bill]

Post by Detlef Müller
Hi,
recently a go player suggested finding out the "adequate
komi" using a fairly strong computer program (i.e. playing
many games against itself with increasing komi until win rate
exceeds 50%).
Is this already explored by someone (and if so: does the
outcome differ much for different programs)?
greetings,
Detlef

It seems intutitive that the stronger the program the higher the value
of the komi. To help validate this, consider beginning players where
the value of komi is about 0.

Bill

[SP]

Post by Bill

Post by Detlef Müller
Hi,
recently a go player suggested finding out the "adequate
komi" using a fairly strong computer program (i.e. playing
many games against itself with increasing komi until win rate
exceeds 50%).
Is this already explored by someone (and if so: does the
outcome differ much for different programs)?
greetings,
Detlef

It seems intutitive that the stronger the program the higher
the value of the komi. To help validate this, consider
beginning players where the value of komi is about 0.

I see your point and I agree. However Detlef's idea sounds
very reasonable:
- as long as God herself is not going to tell us the ideal
komi for each playing strength,
- and as long as fairly strong computer programs can play
many more games against themselves than the strongest human
players.

So his idea will likely tell us which is the appropriate
komi for games at the highest level, let's say 9 dan and
higher. Then it's up to us mere mortal beings if we had
better choose smaller values for games between amateur dan
players, between one-digit kyu players like myself and
between two-digit kyu players.

Greetings
Sergio, 2 kyu

[Bill]

Post by SP
So his idea will likely tell us which is the appropriate
komi for games at the highest level, let's say 9 dan and
higher. Then it's up to us mere mortal beings if we had
better choose smaller values for games between amateur dan
players, between one-digit kyu players like myself and
between two-digit kyu players.

I believe that reasoning to be completely appropriate. Anything else is
simply unfair.

Post by SP
Greetings
Sergio, 2 kyu

[Rainer Rosenthal]

Post by Bill

Post by SP
So his idea will likely tell us which is the appropriate
komi for games at the highest level, let's say 9 dan and
higher. Then it's up to us mere mortal beings if we had
better choose smaller values for games between amateur dan
players, between one-digit kyu players like myself and
between two-digit kyu players.

I believe that reasoning to be completely appropriate. Anything else is
simply unfair.

The word "value" has two different meaning with respect to komi.
1. the numerical value like 6.5 or the like.
2. the importance for the game result.

In my (platonic) opinion there is just one correct value(1).
The value(2) increases with the strength of the players.
The value(1) therefore cannot easily found out by watching weak players.

It seemed to me as if there was the opinion, that value(1) should
be changed according to the player's strength. But that is no good idea.
The correctness is best checked by statistics over games of highest
quality.

Cheers,
Rainer

[SP]

It seemed to me as if there was the opinion, that value(1) (i.e. the
numerical value of komi) should be changed according to the player's > strength. But that is no good idea.

Why not? IMHO we should at least think about it.

[Rainer Rosenthal]

Post by SP

It seemed to me as if there was the opinion, that value(1) (i.e. the
numerical value of komi) should be changed according to the player's > strength. But that is no good idea.

Why not? IMHO we should at least think about it.

Well, why change something, which is not so important?
It is agreed that komi is not important for beginners, right?

[Bill]

Post by Rainer Rosenthal

Post by SP

It seemed to me as if there was the opinion, that value(1) (i.e. the
numerical value of komi) should be changed according to the player's

Post by Rainer Rosenthal
strength. But that is no good idea.

Why not? IMHO we should at least think about it.

Well, why change something, which is not so important?
It is agreed that komi is not important for beginners, right?

It seems to be agreed that komi should be 0 for beginners.

[Detlef Müller]

Post by Bill

Post by Detlef Müller
Hi,
recently a go player suggested finding out the "adequate
komi" using a fairly strong computer program (i.e. playing
many games against itself with increasing komi until win rate
exceeds 50%).
Is this already explored by someone (and if so: does the
outcome differ much for different programs)?

[...]

Post by Bill
It seems intutitive that the stronger the program the higher the value
of the komi. To help validate this, consider beginning players where
the value of komi is about 0.

Mh, this contradicts my intuition.

I'd expect a greater mean variation of the results (meaning the
territorial difference at the End of the game) for weaker
players, sure.

But my ituition would say that (after lots of games) in average the
wins and losses caused by big errors should even out.
(I agree that this is no reason to bother newcommers with the
Idea of komi - their strength changes too fast, so "the probably
stronger one gets white" might be the best advice).

The good news ist: this can nowadays simply checked out using
the suggestion of the afore mentioned go player (as the strength
of [even weak] go programs does not change, assuming no learning
from game to game is implemented).

If different algorithms fight each other, it might be that one
of them is iferior realizing a big komi (maybe a montecarlo
algh. could play too lax starting with biger komi).

If this method leads to different komis dependent on the
programs (playing itself), this would be interesting (especially
for programmers).

Greetings,
Detlef

Post by Bill
Bill

[d***@gmail.com]

One could ditch the whole concept of komi in favor of a series of games, alternating white and black, with a 0.5 komi to prevent ties, until the stronger player has demonstrated this by achieving a games tally advantage of two games. Like the sudden-death phase at the end of a close game of ping-pong, where the score is either tied, or advantage to one player, and the tournement is won when someone wins from the advantage state.

[SP]

Post by d***@gmail.com
One could ditch the whole concept of komi in favor of a series of games, alternating white and black, with a 0.5 komi to prevent ties, until the stronger player has demonstrated this by achieving a games tally advantage of two games. Like the sudden-death phase at the end of a close game of ping-pong, where the score is either tied, or advantage to one player, and the tournement is won when someone wins from the advantage state.

You never have enough time for that system in normal
tournaments. Perhaps in lightning tournaments with 10
minutes each and no byoyomi.
But I saw a similar idea in Japanese go clubs, where people
can go and play every day.
30 years later I don't remember the details, but the person
introducing you said your grade and you were given a number
of points on the ladder of that particular club, e. g. 1 kyu
= 200 points. 10 points difference = 1 kyu or dan.
Let's say I start with 200 points and play my first game on
even (nigiri for choosing the colour) with a regular club
member who has 203 points.
a) I win. Now I have 201 points and he has 202.
b) I lose. Now I have 199 points and he has 203.
c) We draw. Our scores do not change.