Values of moves

General conversations about Go belong here.
Post Reply
User avatar
daal
Oza
Posts: 2508
Joined: Wed Apr 21, 2010 1:30 am
GD Posts: 0
Has thanked: 1304 times
Been thanked: 1128 times

Re: Values of moves

Post by daal »

John Fairbairn wrote: ...there are a few pointers that couch-potato go players need to bear in mind.
A lovely introduction to the topic!
3. We could stop there, but again with O's method it's so easy to try a few more steps. Here's one. How much territory do we count for Black in the corner?
Thanks for the very nice and simple example. I got it right right away, probably because O Meien's method as you describe it is exactly what Kyle Blocher described in his video. In the video, which was made at the 2012 US go congress, he is asked whether pros use this method, and he answers that they don't. Was he wrong, or is O basically an outlier?

My recommendation would be to ignore follow-ups for a while until you are at home with the method, and then add just one level of follow-up. This not only saves on flea powder but it also accords with O Meien's advice not to worry about the sixteenths.
This does seem in line with my wish to get a ballpark figure. I suppose there are a few halfpoint games that I'll just have to lose.
Patience, grasshopper.
mitsun
Lives in gote
Posts: 553
Joined: Fri Apr 23, 2010 10:10 pm
Rank: AGA 5 dan
GD Posts: 0
Has thanked: 61 times
Been thanked: 250 times

Re: Values of moves

Post by mitsun »

Position values and move values are not really that complicated as concepts. For simplicity, assume we have a position where either side may play first with equal probability. In an actual game, this can be a horrendous assumption, but it allows positions to be treated as independent.

The value of a complex position can be calculated by averaging the values of positions resulting from further play. Position value = 1/2 [ (value of resulting position if B plays first) + (value of resulting position if W plays first) ].

The value of a move in a position can be calculated by differencing the values of positions resulting from further play. Move value = 1/2 [ (value of resulting position if B plays first) - (value of resulting position if W plays first) ].

Of course these equations can be mixed if desired. Position value = (value of resulting position if B plays first) - (move value).

From this point of view, the value of a position is more fundamental than the value of a move. However, as has been pointed out, if you just want to find the value of a move, it may not be necessary to calculate the value of the position.

For the mathematically inclined, the value of the move is the derivative, with respect to the next play, of the value of the position.
Bill Spight
Honinbo
Posts: 10905
Joined: Wed Apr 21, 2010 1:24 pm
Has thanked: 3651 times
Been thanked: 3373 times

Re: Values of moves

Post by Bill Spight »

daal wrote:I got it right right away, probably because O Meien's method as you describe it is exactly what Kyle Blocher described in his video. In the video, which was made at the 2012 US go congress, he is asked whether pros use this method, and he answers that they don't. Was he wrong, or is O basically an outlier?


Except for the use of multiples, which O does not do, everybody calculates the values of sente, gote, simple ko, and nidan (two stage) ko positions the same, going back at least 200 years. Where O (along with Blocher, Berlekamp, Wolfe, Mueller, myself, et al.) differs from the usual pro practice is in utilizing miai counting (Absolute™ counting) for plays and in taking whole board temperature into account (although he does not use that term). He also innovates by utilizing an error term of ¼ of the temperature.

Does Blocher talk about komaster? Pros have not caught on to that, unfortunately, even though Berlekamp published the idea back in 1994. :(
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.
bernds
Lives with ko
Posts: 259
Joined: Sun Apr 30, 2017 11:18 pm
Rank: 2d
GD Posts: 0
Has thanked: 46 times
Been thanked: 116 times

Re: Values of moves

Post by bernds »

Bill Spight wrote:
bernds wrote:]If I can compute the value of a move, why would I care about the value of a local position?
But you are not able to compute the value of a move without computing the value of local positions. daal computed the value of the move by computing the value of two positions, the one where Black had 5 pts. and the one where White had 1 pt. He did so by counting the local scores.

However, he cannot rely upon the local positions he uses to evaluate plays to be scorable. So he needs to be able to calculate the value of non-final (non-scorable) positions, as well.
I guess I'm still confused, as always when looking at this miai counting business. Let's take the first example from Sensei's library:

Click Here To Show Diagram Code
[go]$$B Example diagram (gote)
$$ . X X X X X X O O O O O .
$$ . X . . . . X O . . . O .
$$ . X . . . . . . . . . O .
$$ -------------------------[/go]
SL page wrote:In Example 1[#2], the count is 2 (Black has 2 points more than White).
And I have to say that makes little sense to me. At the moment, no one has points. If we assume that everyone connects against a hane, then I can agree Black has two more points if we average the two possibilities, but in that case we could just look at a smaller section of the problem and say with the same justification that the count is zero.

The way it's stated it sounds like it should be important, I just don't see it. You're saying I can't just count up the final positions of a sequence of plays - what's the smallest example where that is the case?
Bill Spight
Honinbo
Posts: 10905
Joined: Wed Apr 21, 2010 1:24 pm
Has thanked: 3651 times
Been thanked: 3373 times

Re: Values of moves

Post by Bill Spight »

daal wrote:This is a fine idea. Where might one find more such easy examples?


You might take a look at this post and the following in This 'n' That ( viewtopic.php?p=194704#p194704 ). I see that I post a problem pretty soon, but maybe it's not too difficult. :)
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.
John Fairbairn
Oza
Posts: 3724
Joined: Wed Apr 21, 2010 3:09 am
Has thanked: 20 times
Been thanked: 4672 times

Re: Values of moves

Post by John Fairbairn »

I got it right right away, probably because O Meien's method as you describe it is exactly what Kyle Blocher described in his video. In the video, which was made at the 2012 US go congress, he is asked whether pros use this method, and he answers that they don't. Was he wrong, or is O basically an outlier?


Both right and wrong, and yes and no. O Meien said he was surprised his method was not used in Japan, but it was normal in China. He probably meant Taiwan specifically, but I imagine they got it from mainand China anyway.


Position values and move values are not really that complicated as concepts. For simplicity, assume we have a position where either side may play first with equal probability. In an actual game, this can be a horrendous assumption, but it allows positions to be treated as independent.

The value of a complex position can be calculated by averaging the values of positions resulting from further play. Position value = 1/2 [ (value of resulting position if B plays first) + (value of resulting position if W plays first) ].

The value of a move in a position can be calculated by differencing the values of positions resulting from further play. Move value = 1/2 [ (value of resulting position if B plays first) - (value of resulting position if W plays first) ].

Of course these equations can be mixed if desired. Position value = (value of resulting position if B plays first) - (move value).

From this point of view, the value of a position is more fundamental than the value of a move. However, as has been pointed out, if you just want to find the value of a move, it may not be necessary to calculate the value of the position.

For the mathematically inclined, the value of the move is the derivative, with respect to the next play, of the value of the position.


I'm sorry, but this is the kind of thing I meant when I said seeing it makes me lose the will to live. I understand every word but I have no idea what it means. Mathematicians actually speak mathlish. It may look like English, it may sound like English, but only other mathlanders understand it. And like the Eskimos and their 400 kinds of snow, I expect the go-playing mathlanders also have 400 kinds of sente.

Bill: De-iri is traditional counting. Since miai became mainstream only after I was born, I'm not sure I'm ready to accept that as traditional yet - and I know you're not young enough yourself to be an uberdude!
Bill Spight
Honinbo
Posts: 10905
Joined: Wed Apr 21, 2010 1:24 pm
Has thanked: 3651 times
Been thanked: 3373 times

Re: Values of moves

Post by Bill Spight »

bernds wrote:
Bill Spight wrote:
bernds wrote:]If I can compute the value of a move, why would I care about the value of a local position?
But you are not able to compute the value of a move without computing the value of local positions. daal computed the value of the move by computing the value of two positions, the one where Black had 5 pts. and the one where White had 1 pt. He did so by counting the local scores.

However, he cannot rely upon the local positions he uses to evaluate plays to be scorable. So he needs to be able to calculate the value of non-final (non-scorable) positions, as well.
I guess I'm still confused, as always when looking at this miai counting business. Let's take the first example from Sensei's library:

Click Here To Show Diagram Code
[go]$$B Example diagram (gote)
$$ . X X X X X X O O O O O .
$$ . X . . . . X O . . . O .
$$ . X . . . . . . . . . O .
$$ -------------------------[/go]
SL page wrote:In Example 1[#2], the count is 2 (Black has 2 points more than White).
And I have to say that makes little sense to me.


Like nearly every yose text aimed at non-mathematicians, the SL text is imprecise. To be precise, Black has 2 more points than White on average.

At the moment, no one has points. If we assume that everyone connects against a hane, then I can agree Black has two more points if we average the two possibilities, but in that case we could just look at a smaller section of the problem and say with the same justification that the count is zero.


You mean like so?

Click Here To Show Diagram Code
[go]$$B Example diagram 2
$$ X X X O O O
$$ . . X O . .
$$ . . . . . .
$$ -----------[/go]


Yes, we could say that the count of this region is 0. Is that of any interest?

The way it's stated it sounds like it should be important, I just don't see it.


Players are generally interested in who has more (expected) territory.

You're saying I can't just count up the final positions of a sequence of plays - what's the smallest example where that is the case?


Click Here To Show Diagram Code
[go]$$B Small position (all stones alive)
$$ . X X X X O .
$$ . X . . . O .
$$ -------------[/go]
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.
Calvin Clark
Lives in gote
Posts: 426
Joined: Thu Aug 13, 2015 8:43 am
GD Posts: 0
Has thanked: 186 times
Been thanked: 191 times

Re: Values of moves

Post by Calvin Clark »

bernds wrote:Let's take the first example from Sensei's library:

Click Here To Show Diagram Code
[go]$$B Example diagram (gote)
$$ . X X X X X X O O O O O .
$$ . X . . . . X O . . . O .
$$ . X . . . . . . . . . O .
$$ -------------------------[/go]
SL page wrote:In Example 1[#2], the count is 2 (Black has 2 points more than White).
And I have to say that makes little sense to me. At the moment, no one has points. If we assume that everyone connects against a hane, then I can agree Black has two more points if we average the two possibilities, but in that case we could just look at a smaller section of the problem and say with the same justification that the count is zero.

The way it's stated it sounds like it should be important, I just don't see it. You're saying I can't just count up the final positions of a sequence of plays - what's the smallest example where that is the case?


It's arbitrary, but you have to start somewhere in order to compare the values of later positions to the starting position. So imagine that :bc: :wc: are already there:

Click Here To Show Diagram Code
[go]$$B Example diagram (gote)
$$ . X X X X X X O O O O O .
$$ . X a . . . X O . . . O .
$$ . X a . . . B W . . . O .
$$ -------------------------[/go]



Then we count 8 - 6 = 2. But we don't have to do that. We could as easily think that the points marked 'a' are not likely to be affected as we study this, they will always be black's unless black is killed, which is not the assumption of this problem. So if you don't count the 'a' points, then the score is 0, which is easier to remember than 2.

Indeed, if the diagram shown were like this, I would probably only count the a's for black and the b's for white.

Click Here To Show Diagram Code
[go]$$B black has a lot of points not in play
$$ . X X X X X X X X X X O O O O O .
$$ . X . . . . . . a a X O b b . O .
$$ . X . . . . . . a a . . b b . O .
$$ ---------------------------------[/go]


0 is easy to remember.

(Edit: Fix Diagram to put in clear line. Thanks, Bill.)
Last edited by Calvin Clark on Sat Sep 22, 2018 11:24 pm, edited 1 time in total.
Bill Spight
Honinbo
Posts: 10905
Joined: Wed Apr 21, 2010 1:24 pm
Has thanked: 3651 times
Been thanked: 3373 times

Re: Values of moves

Post by Bill Spight »

John Fairbairn wrote:Mathematicians actually speak mathlish. It may look like English, it may sound like English, but only other mathlanders understand it. And like the Eskimos and their 400 kinds of snow, I expect the go-playing mathlanders also have 400 kinds of sente.


Mathlanders have only one or two kinds of sente, but I have redefined one of them as ambiguous. ;) It's regular go players who have multiple meanings of sente, often without realizing it.

Bill: De-iri is traditional counting. Since miai became mainstream only after I was born, I'm not sure I'm ready to accept that as traditional yet - and I know you're not young enough yourself to be an uberdude!


Sorry for not being clear. O's method of calculating the value of a position is traditional. That's what I was talking about. :)
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.
RobertJasiek
Judan
Posts: 6272
Joined: Tue Apr 27, 2010 8:54 pm
GD Posts: 0
Been thanked: 797 times
Contact:

Re: Values of moves

Post by RobertJasiek »

Counts (values of local positions) are used for accurate positional judgement, calculation of moves values, calculation of gains, distinctions of gote/sente, identification of good moments of tenuki and determination of correct move orders. After a move, earlier calculated counts of earlier follow-ups are still useful. For all these reasons, counts are not throw-away objects. They are as important as move values.

Although beginners of endgame evaluation should indeed start with simple local endgames without follow-ups, later follow-ups can easily be relevant especially if they are large, or make some branch larger than other branches. Not just for the sake of evaluation itself but also for finding the more valuable moves.
Calvin Clark
Lives in gote
Posts: 426
Joined: Thu Aug 13, 2015 8:43 am
GD Posts: 0
Has thanked: 186 times
Been thanked: 191 times

Re: Values of moves

Post by Calvin Clark »

Click Here To Show Diagram Code
[go]$$W okay, maybe I have to count
$$ . X X X X X X X X X X O O O O . .
$$ . X . . . . . . . . X O . . O . .
$$ . X . . . . . . 1 . O . . . O . .
$$----------------------------------[/go]


Yes, as Robert states, eventually one may have to leave the parrot zone, when black may chose to allow :w1:, which is more like a real game situation.
Last edited by Calvin Clark on Sat Sep 22, 2018 11:25 pm, edited 1 time in total.
Bill Spight
Honinbo
Posts: 10905
Joined: Wed Apr 21, 2010 1:24 pm
Has thanked: 3651 times
Been thanked: 3373 times

Re: Values of moves

Post by Bill Spight »

The material in this post ( viewtopic.php?p=195326#p195326 ) and the following are pretty easy, I think, but not traditional. ;)
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.
Bill Spight
Honinbo
Posts: 10905
Joined: Wed Apr 21, 2010 1:24 pm
Has thanked: 3651 times
Been thanked: 3373 times

Re: Values of moves

Post by Bill Spight »

@ Calvin

You can add the edges of the board to diagrams, comme ça.

Click Here To Show Diagram Code
[go]$$W okay, maybe I have to count
$$ . X X X X X X X X X X O O O O . . |
$$ . X . . . . . . . . X O . . O . . |
$$ . X . . . . . . 1 . O . . . O . . |
$$ ---------------------------------[/go]
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.
bernds
Lives with ko
Posts: 259
Joined: Sun Apr 30, 2017 11:18 pm
Rank: 2d
GD Posts: 0
Has thanked: 46 times
Been thanked: 116 times

Re: Values of moves

Post by bernds »

Bill Spight wrote:Yes, we could say that the count of this region is 0. Is that of any interest?
In the sense of causing confusion along the lines of "so why does it state that you start by assigning a count, when the number is arbitrary and therefore unimportant?" In Daal's original example, you gave a value of 0 for a move at A, and chose not to discuss the size of the play. Once again that seems to give priority to the question of the value of a position (and you also said "The value of positions is basic"), but I struggle to understand how it is relevant to the question of where to play.

A lot of material both here and on SL seems unnecessarily opaque, it sometimes feels like it's trying to generate mystique rather than trying to be helpful.

Click Here To Show Diagram Code
[go]$$B Small position (all stones alive)
$$ . X X X X O .
$$ . X . . a O .
$$ -------------[/go]
So... what's special about it? Uncertainty whether A is sente for White or not? So how do you go about assigning a count, and how does it help you decide the size of a move at A?
dfan
Gosei
Posts: 1598
Joined: Wed Apr 21, 2010 8:49 am
Rank: AGA 2k Fox 3d
GD Posts: 61
KGS: dfan
Has thanked: 891 times
Been thanked: 534 times
Contact:

Re: Values of moves

Post by dfan »

By the way, if anyone is looking for simple counting examples, and doesn't mind using deiri counting rather than miai counting, Lee Chang-Ho's Endgame Techniques Volume 1 (unfortunately only available in the GoBooks app, I forget why) has tons of examples with very clear explanations, starting from the trivial and working up to ones that are complex enough that I wouldn't bother to work it all out even in a correspondence game. Volume 2 is also very nice but is about endgame tesuji.

For anyone here who happens to already be a member of the Yunguseng Dojang, In-seong Hwang has a bunch of very nice endgame-counting problem-solving lecture videos in which the problems progress nicely over the course of each lecture, again starting with very simple ones.
Post Reply