Life In 19x19

Dare I say using KataGo?

The best for both diagram after White has refuted Black's trick play, has this interesting end position. Black captures 3 stones on a large scale. White gets territory and influence at the top plus sente. KG finds this about 1.5 better for White, so it should be playable.

https://senseis.xmp.net/?34PointHighApp ... mentNoseki

I'm struggling with the interpretation as Black looks larger. Perhaps sente makes the difference - this was a Black corner to begin with - but there's aji in the loose capture or at least a few ko threats. A and B come to mind.

Click Here To Show Diagram Code

[go]$$W Best for both
$$ ------------------
$$ | . . O . . . . . .
$$ | . O O O O O . . .
$$ | . X O X O X O 3 .
$$ | . X O X X X X 1 .
$$ | . X X O X . 2 . .
$$ | . . . O O . . a .
$$ | . . X . . . 4 . .
$$ | . . . . X . . . .
$$ | . . . . b . . . .[/go]

I didn't check KataGo just now but I think I had the same or similar position in a practice game with KataGo some months ago, it was confusing.

First thing is that black maybe isn't that much bigger than white. Maybe 7 points and the direction of black's or white's shape could go either way.

Maybe the main thing then, which I vaguely recall, is that it is awkward for black to lean onto white. See diagrams.

Click Here To Show Diagram Code

[go]$$B Black is awkward, liberties are short.
$$ +------------------------
$$ | . . O . . . . . . . . .
$$ | . O O O O O . . . . . .
$$ | . X O X O X O O . . . .
$$ | . X O X X X X O . , . .
$$ | . X X O X . X 1 2 . . .
$$ | . . . O O . . . 3 4 . .
$$ | . . X . . . X . . . . .
$$ | . . . . X . . . . . . .
$$ | . . . . . . . . . . . .
$$ | . . . , . . . . . , . .[/go]

Click Here To Show Diagram Code

[go]$$B This doesn't make much sense in the detail, I think, but maybe for illustrating a point. 11 at half black point.
$$ +------------------------
$$ | . . O . . . . . . . . .
$$ | . O O O O O . . . . . .
$$ | . X O X O X O O . . . .
$$ | . X O X X X X O . , . .
$$ | . X X O X . X 2 1 5 . .
$$ | . . . O O . 9 3 4 . 7 .
$$ | . . X . . A X 8 6 . . .
$$ | . . . . X . 0 . . . . .
$$ | . . . . . . . . . . . .
$$ | . . . , . . . . . , . .[/go]

I think there are similar issues with other moves that cover or lean on white.

What I think I recall is that in my particular game the white suji was eventually not to peep but this keima. Maybe the threat is stronger than the execution, like they say in chess.

Click Here To Show Diagram Code

[go]$$W Good suji?
$$ +------------------------
$$ | . . O . . . . . . . . .
$$ | . O O O O O . . . . . .
$$ | . X O X O X O O . . . .
$$ | . X O X X X X O . , . .
$$ | . X X O X . X . . . . .
$$ | . . . O O . . . 1 . . .
$$ | . . X . . . X . . . . .
$$ | . . . . X . . . . . . .
$$ | . . . . . . . . . . . .
$$ | . . . , . . . . . , . .[/go]

I might misremember but it makes sense to me now that if white has good suji to block black and black doesn't have anything as natural then that is a big shape thing. This reminds me of something I read recently in a book called Shape Up!, by someone that hasn't been seen on L19 for awhile, about "gain lines". Which isn't defined in the book, as far as I can tell, but it is the contact line between two groups that you must cross if you wish to push the other side back.

At first it would appear as if the gain line is diagonal, the first diagram, but it appears that white already has the upper hand, second diagram, due to black's move here being awkward.

The way I draw the lines probably doesn't make much sense, it might overemphasis the difference

Click Here To Show Diagram Code

[go]$$W Apparent gain line?
$$ +------------------------
$$ | . . O . . . . . . . . .
$$ | . O O O O O . . . . . .
$$ | . X O X O X O O . . . .
$$ | . X O X X X X O . , . .
$$ | . X X O X . X . . . . .
$$ | . . . O O . . . . . . .
$$ | . . X . . . X . . . . .
$$ | . . . . X . . . . . . .
$$ | . . . . . . . . . . . .
$$ | . . . , . . . . . , . .
$$ {AR H15 M12}[/go]

Click Here To Show Diagram Code

[go]$$W Real gain line?
$$ +------------------------
$$ | . . O . . . . . . . . .
$$ | . O O O O O . . . . . .
$$ | . X O X O X O O . . . .
$$ | . X O X X X X O . , . .
$$ | . X X O X . X . . . . .
$$ | . . . O O . . . . . . .
$$ | . . X . . . X . . . . .
$$ | . . . . X . . . . . . .
$$ | . . . . . . . . . . . .
$$ | . . . , . . . . . , . .
$$ {AR H15 K10}[/go]

How about this as an explanation?

Sente is secondary. Difference of numbers of played stones is a primary aspect. Do not forget local prisoners!

kvasir wrote:if white has good suji to block black and black doesn't have anything as natural then that is a big shape thing

Makes a lot of sense. Thanks!

All these sente aji exploitations must not distract from Black having the better local result because of having played one excess stone! Suppose your Katago assessment of White's 1.5 global lead is correct (we cannot know yet because you have not stated numbers of playouts), it means that locally Black should be better by 7 (equal to komi to compensate the one excess stone) - 1.5 = 5.5 points for excess territory and territoty potential of the local stones.

RobertJasiek wrote:All these sente aji exploitations must not distract from Black having the better local result because of having played one excess stone! Suppose your Katago assessment of White's 1.5 global lead is correct (we cannot know yet because you have not stated numbers of playouts), it means that locally Black should be better by 7 (equal to komi to compensate the one excess stone) - 1.5 = 5.5 points for excess territory and territoty potential of the local stones.

Yes, black has one more stone and his position should look more impressive. Also, if we are talking about KataGo evaluation then it would be about the total effect on the whole board. Black could need 7 points in total effect, something between 6 and 8 seems reasonable. But we don't have any real way of estimating the total effect precisely enough, except using a computer.

So I pointed out one reason, or plausible reason rather, why black's position isn't as overwhelming as it could look at first. I think there are other examples of aji and suji that could be invoked. One could also try tewari.

What I don't understand is how one would reach an accurate and precise enough estimate of the total effect in points without accurately treating the follow ups here.

One could of course say
total effect : e := -1.5
moves: t := -7
territory: s := +7
outside: x , it is an unknown
and say total effect is
-1.5 = e = t + s + x = x
therefore outside is -1.5 but then the question is why? "outside" here could also be "everything else", I suppose.

This is somewhat simplistic, probably not exactly what you have in mind. But I have another reason to not trust this kind of formulae in general, which is that the error cascades. If instead t = -6.5, s = 6.5 then x = -2.5, and this problem is magnified by every factor that you add to the simplistic model.

The 7 points to compensate 1 excess stone is the value of my simplifying model in [30] with these idealistic assumptions:

0) 7 is the correct komi.
1) We are still early enough in the game so that the per-move move value is 14 (twice the komi).
2) The excess move is played elsewhere and is independent of the local situation.
3) 50% of the player's move value benefits his immediate territory increment, 0% of the move value benefits his influence for a later increment of his territory and 50% of the move value is influence
hurting his opponent's territory.

kvasir wrote:

RobertJasiek wrote:All these sente aji exploitations must not distract from Black having the better local result because of having played one excess stone! Suppose your Katago assessment of White's 1.5 global lead is correct (we cannot know yet because you have not stated numbers of playouts), it means that locally Black should be better by 7 (equal to komi to compensate the one excess stone) - 1.5 = 5.5 points for excess territory and territoty potential of the local stones.

Yes, black has one more stone and his position should look more impressive. Also, if we are talking about KataGo evaluation then it would be about the total effect on the whole board. Black could need 7 points in total effect, something between 6 and 8 seems reasonable. But we don't have any real way of estimating the total effect precisely enough, except using a computer.

So I pointed out one reason, or plausible reason rather, why black's position isn't as overwhelming as it could look at first. I think there are other examples of aji and suji that could be invoked. One could also try tewari.

What I don't understand is how one would reach an accurate and precise enough estimate of the total effect in points without accurately treating the follow ups here.

One could of course say
total effect : e := -1.5
moves: t := -7
territory: s := +7
outside: x , it is an unknown
and say total effect is
-1.5 = e = t + s + x = x
therefore outside is -1.5 but then the question is why? "outside" here could also be "everything else", I suppose.

This is somewhat simplistic, probably not exactly what you have in mind. But I have another reason to not trust this kind of formulae in general, which is that the error cascades. If instead t = -6.5, s = 6.5 then x = -2.5, and this problem is magnified by every factor that you add to the simplistic model.

I do not understand this formula e = t + s + x
I guess e takes into account the komi but I do not see the komi in t + s + x. t looks like the komi but if I understand correctly 2t is the temperature of the board not the 2*komi value. Assume a game in which black player is known as far better than white player. In this case the players might agree to use for example a komi = 20. That does not change t but e is very different isn't it?
Could you clarify the meaning of each term of the formula?

Gérard TAILLE wrote:I do not understand this formula e = t + s + x
I guess e takes into account the komi but I do not see the komi in t + s + x. t looks like the komi but if I understand correctly 2t is the temperature of the board not the 2*komi value. Assume a game in which black player is known as far better than white player. In this case the players might agree to use for example a komi = 20. That does not change t but e is very different isn't it?
Could you clarify the meaning of each term of the formula?

I thought I did state the meanings.

If we want the formula to have the following form

e = t + s + x

then one could use these (hopefully clearer) definitions

e : estimate total effect of playing a move, when using katago to evaluate it is the residual of the evaluation before the move and after the move.

t : estimate cost of having played more moves in the position, if the value of one move is k, the count of black and white moves are cB and cW, then t = k * (cW - cB), note that it is negative when black played more moves. Note that k here could also be called a value of a half move, but not in the way we are counting moves in this definition

s : estimate territory score, that is the difference of black and white territory, if we say estimated black and white territory are tB and tW then s = tB - tW

x : in my case I called it outside or everything else, I think that is something some people call influence, basically it should estimate the residual e - (t + s), however you go about it.

I think you can estimate t, s and x to some degree directly, but if you prefer you could write

e = t + s + x = k * (cW - cB) + tB - tW + x

I'm not recommending this or similar formulas as a way to evaluate positions, I don't think it is easy to estimate the factors that go into such a model accurately enough. They may have other uses as models, for example to communicate about which factors that make up the whole. Communicating was my goal and it is often a good idea to avoid delving into every nitty-gritty detail when trying to communicate

I hope this clarifies something.

kvasir wrote:

Gérard TAILLE wrote:I do not understand this formula e = t + s + x
I guess e takes into account the komi but I do not see the komi in t + s + x. t looks like the komi but if I understand correctly 2t is the temperature of the board not the 2*komi value. Assume a game in which black player is known as far better than white player. In this case the players might agree to use for example a komi = 20. That does not change t but e is very different isn't it?
Could you clarify the meaning of each term of the formula?

I thought I did state the meanings.

If we want the formula to have the following form

e = t + s + x

then one could use these (hopefully clearer) definitions

e : estimate total effect of playing a move, when using katago to evaluate it is the residual of the evaluation before the move and after the move.

t : estimate cost of having played more moves in the position, if the value of one move is k, the count of black and white moves are cB and cW, then t = k * (cW - cB), note that it is negative when black played more moves. Note that k here could also be called a value of a half move, but not in the way we are counting moves in this definition

s : estimate territory score, that is the difference of black and white territory, if we say estimated black and white territory are tB and tW then s = tB - tW

x : in my case I called it outside or everything else, I think that is something some people call influence, basically it should estimate the residual e - (t + s), however you go about it.

I think you can estimate t, s and x to some degree directly, but if you prefer you could write

e = t + s + x = k * (cW - cB) + tB - tW + x

I'm not recommending this or similar formulas as a way to evaluate positions, I don't think it is easy to estimate the factors that go into such a model accurately enough. They may have other uses as models, for example to communicate about which factors that make up the whole. Communicating was my goal and it is often a good idea to avoid delving into every nitty-gritty detail when trying to communicate

I hope this clarifies something.

Thank you very much kvasir. I see more clearly your point. Now I understand that the komi used is not relevant here but I have another issue.
In your post https://lifein19x19.com/viewtopic.php?p=280063#p280063 you used the value
moves: t := -7
where I expected the value t := -14
Can you clarify why you use this value moves: t := -7 ?

Gérard TAILLE wrote:Thank you very much kvasir. I see more clearly your point. Now I understand that the komi used is not relevant here but I have another issue.
In your post https://lifein19x19.com/viewtopic.php?p=280063#p280063 you used the value
moves: t := -7
where I expected the value t := -14
Can you clarify why you use this value moves: t := -7 ?

It is a matter of convention.

The way I was counting moves was that if black played cB times and white cW times the difference of moves played (which I think is sometimes called the tally) is cB - cW. Let's say we have a position with a given t and then add a move for black in one case and for white in another:
tB = (cB + 1) - cW = t + 1
tW = cB - (cW + 1) = t - 1

Now tB - tW = 2 and just for the purpose of the original post I call that a two move difference, you could also say it is a one move difference. Depending on your choice you will have to scale other values but I found it natural in this case to work with what others may call half-moves.

For example you could say

t = K * (cW - cB) / 2

and use K = 2 * k.

Now that I think about it one must be careful to match this with the KataGo evaluation function Do you think I made a mistake in that regard? I was of course trying to make a conceptual point more than anything else but I think I matched it correctly, but there are at least two conventions.

kvasir wrote:

Gérard TAILLE wrote:Thank you very much kvasir. I see more clearly your point. Now I understand that the komi used is not relevant here but I have another issue.
In your post https://lifein19x19.com/viewtopic.php?p=280063#p280063 you used the value
moves: t := -7
where I expected the value t := -14
Can you clarify why you use this value moves: t := -7 ?

It is a matter of convention.

The way I was counting moves was that if black played cB times and white cW times the difference of moves played (which I think is sometimes called the tally) is cB - cW. Let's say we have a position with a given t and then add a move for black in one case and for white in another:
tB = (cB + 1) - cW = t + 1
tW = cB - (cW + 1) = t - 1

Now tB - tW = 2 and just for the purpose of the original post I call that a two move difference, you could also say it is a one move difference. Depending on your choice you will have to scale other values but I found it natural in this case to work with what others may call half-moves.

For example you could say

t = K * (cW - cB) / 2

and use K = 2 * k.

Now that I think about it one must be careful to match this with the KataGo evaluation function Do you think I made a mistake in that regard? I was of course trying to make a conceptual point more than anything else but I think I matched it correctly, but there are at least two conventions.

I do not know if if you made a mistake but we can try to apply the formula to the following example.

Click Here To Show Diagram Code

[go]$$B Black to play - komi 7.5
$$ ---------------------------------------
$$ | . . . . . . . . . . . X . X O . a . O |
$$ | . . . . . . . . . . . X X X O O . . O |
$$ | . . . O . . . . . . . X . X O O O O O |
$$ | . . . , . . . . . , . X X X X X X X X |
$$ | . . . . . . . . . . . X X X X X X X X |
$$ | . . . . . . . . . . . . . O O O O O O |
$$ | . . . . . . . . . . . . . O . . . . . |
$$ | . . . . . . . . . . . . . O O O O O O |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

According to katago (with not a lot of playouts ) the position is about equal.
What is the value of territory in the upper right corner? It is not very difficult : if black plays first and kills white then the black territory is +21, and if it white plays first to live then the black territory is -7. As a consequence, for this initial position I consider the black territory is ((+21) + (-7)) / 2 = +7
What is the value of the influence of the upper right corner? Considering the other corner do not give a significant advantage to a player and because the global position is equal that means that the influence of the upper right corner cancels the territory in the corner. So the value of this inflence is about equat to -7.

Now let's try to estimate the total effect of playing a black move at "a", killing white stones in the upper right corenr.
Your formula is
e = t + s + x
According to katago, killing the white stones is a good move => e = 0
We can see that killing the white stones do not change the influence of the upper right corner => x = 0
The territory in the corner becomes +21 instead of +7 => t = 14
e = t + s + x => 0 = +14 + s + 0 => s = -14
I may be wrong but if you consider s = -7 then the formula should be
e = t + 2s + x

Gérard TAILLE wrote:I do not know if if you made a mistake but we can try to apply the formula to the following example.

I think there is a misunderstanding here. To evaluate this position using a using katago evaluation and a formula such as the one mentioned you need to take into account that the local position has certain number of moves and the global position has in effect one more or less move for each player, depending on who is to play. We can't talk about mapping the whole board evaluation into a local evaluation unless we basically agree on a value for the right to move on the whole board.

While it is interesting it is also an expansion of scope of the discussion we were having. If I get back to you on this then it wouldn't really be in connection to what was said above BTW I do get the values that I expect I'm just not sure it is worthwhile to copy the calculations here without breaking this cycle of what was an example of a model, that I criticized, becoming a model that you seem to expect me to expand up on Maybe I can create a new topic on it if I'm in a good mood sometime

Gérard TAILLE wrote:I do not know if if you made a mistake but we can try to apply the formula to the following example.

You can see how I apply the formula in https://lifein19x19.com/viewtopic.php?p=280132#p280132