Values of moves

daal · **#21**

Bill Spight wrote:

daal wrote:

Bill Spight wrote:

I edited the diagrams to show the bitter end. The value of the position lies between 4 and -4, both values from Black's point of view.

Ok, the value of that position is 0. So what does the value of the position tell us about whether to move there or not? Does it make sense to compare it to the value of another position, for example this one:

Not that one, but this one.

Click Here To Show Diagram Code: [go]$$B After White plays gote $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . W . . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . . . . | $$ | . . , X . . , . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

Playing in the top left, because the three moves (a, b, and c) and positions are assumed to be independent (even if that's not quite right).

For instance, let's guess that this position is a White sente.

Click Here To Show Diagram Code: [go]$$W White sente $$ +-------------------+ $$ | . . . 3 . C C C C | $$ | 6 5 1 2 W . C C C | $$ | . 4 7 X . O C C C | $$ | S 8 . X . O C C C | $$ | . . X . , . O . . | $$ | . . . . X . . . . | $$ | . . , X . . , . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

Then we can make a ball park estimate that this position is worth 12 pts. to White. That means that we can estimate that :wc:

has gained 12 pts. Like Tami said.

Ok. So you are saying that the value of a position gives us a reference from which to judge changes in that position. Right? In the above example are you showing how one determines a value for a move at a in order to be able to compare it to other moves such as b or c, Right? If so, I don't get it. Don't we have to assume that black answers the move at a?

Bill Spight · **#22**

daal wrote:

Ok. So you are saying that the value of a position gives us a reference from which to judge changes in that position. Right? In the above example are you showing how one determines a value for a move at a in order to be able to compare it to other moves such as b or c, Right?

Yes, as long as you realize that the biggest move is not always best.

Quote:

If so, I don't get it. Don't we have to assume that black answers the move at a?

If so, don't you also have to assume that White answers Black's move at a? In that case, a move at a is a free lunch. You don't have to compare it to anything, just play it.

Or maybe you think you have to compare it to other plays to decide whether to answer it or not. In that case, you can't assume that it is a free lunch.

daal · **#23**

I get that it's a free lunch. If it is sente for both, just play it. That makes sense. What I don't see is what the diagram means where you say that w gets 12 points. Since we are assuming a play at a is sente, why would that diagram take place? Wouldn't black just answer a?

Bill Spight · **#24**

daal wrote:

I get that it's a free lunch. If it is sente for both, just play it. That makes sense. What I don't see is what the diagram means where you say that w gets 12 points. Since we are assuming a play at a is sente, why would that diagram take place? Wouldn't black just answer a?

You're assuming that it is a free lunch, I'm not.

Click Here To Show Diagram Code: [go]$$Bcm11 White to play $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . 1 . . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . X . O . . . | $$ | . . , X . O , . . | $$ | . . . . . . . . . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

TANSTAAFL :cool:

RobertJasiek · **#25**

For understanding some basics of evaluation, starting with one local endgame that is a whole board (life and death) problem is maybe not the simplest:)

EdLee · **#26**

Quote:

a whole board (life and death) problem...

daal, Robert,

Does it help (a bit) if we remove L&D concerns... ?

Click Here To Show Diagram Code: [go]$$Wc (ignoring dimmed area) $$ ------------------------- $$ | ? ? ? . . . . . . . . | $$ | ? ? ? . . . a . . . . | $$ | ? ? ? . . X . O , . . | $$ | ? ? ? . . X . O . . . | $$ | ? ? ? . . X . O . . . | $$ | ? ? ? . . X . O . . . | $$ | ? ? ? . . . . . O . . | $$ | ? ? ? . . . X . b . . | $$ | ? ? ? . . X . . , . . | $$ | ? ? ? . . . c O . O . | $$ | ? ? ? . . . . . . . . | $$ -------------------------[/go]

RobertJasiek · **#27**

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . . . . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . 1 . . | $$ | . . , X . . 2 . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

A minimum amount of tactical reading is also necessary.

daal · **#28**

Alright, I give up. it was a stupid question.

RobertJasiek · **#29**

We are told that there are no stupid questions but maybe too optimistic expectations.

Basic endgame theory can be value-less but as soon as we always want to compare different moves well, move values and other values come in. If they shall be meaningful by comparing well, calculation and comparison of numbers cannot be avoided.

There have been attempts of simplifying that: no follow-ups, only one follow-up move, no fractions, only occasional trivial fractions, rounding, only positive numbers, no arithmic operation, only addition, only gote, deemphasised sente etc. All such simplifications have quickly led to insufficient scope of application or frequent mistakes in evaluation.

Some kinds of simplifications can make sense if used so that they do make sense: rounding if the rounding error does not cause wrong comparisons, stopped deeper iterative analysis of follow-ups if such rounding is guaranteed, considering an ensemble of moves with large values and replacing all smaller moves by some environment model, ignoring the last move before the microendgame etc.

Traditional endgame theory has the simplification of initially avoiding calibration of move values and the complication of initially possibly having uncomparable move values. Modern endgame theory has the complication of immediately calibrating move values and the simplification of already having comparable move values.

There is no such thing as consistently good endgame without effort of value calculation and comparison. Quite like there is no consistently good life and death without effort of tactical reading, no consistently good positional judgement without effort of counting and analysis etc.

The endgame affects circa half of the moves so why would you expect to play good endgame if spending less than half of your go study on it? You play a scored game but expect to succeed without calculating numbers? Fight for your life and you find yourself in counting liberties instead of endgame values. However, good fighting for resignation relies on assessing its necessity by good positional judgement and predicted endgame in the case of a peacefully continued game.

Learn to love calculations!

Bill Spight · **#30**

Tami wrote:

daal wrote:

You said you will not have time to calculate during a game. I guess the usual thing is to calculate examples outside your games, and by that train your intuition to better and better guess the values quickly during a game. It's like with tsumego. The more you do them the better you get at guessing how to kill or live in a fast game. I wouldn't mind doing this, if there were a list of 10 common positions and their values, but there doesn't seem to be such a list, so I wouldn't know where to start.

There is the Nihon Kiin Yose Small Dictionary. ISBN 978-4-8182-0437. There is a section giving values for typical endgame plays. It has a lot of other good things, too, and you only need minimal Japanese to get the gist of it.

Just don't trust it about the free lunch (double sente). It's whole board example with several so-called double sente is just wrong.

Also, check out the Sensei's Library miai value lists, as light vector said.

Bill Spight · **#31**

daal wrote:

Alright, I give up. it was a stupid question.

What was a stupid question? Certainly not your original one.

Maybe the easiest way to get a ballpark figure is harder than you had hoped, but it still is within your capabilities.

But the place to start is with evaluating local positions. (At least for humans.

) Endgame books show how to do that.

Once you can make ballpark estimates of local positions, it is not difficult to make ballpark estimates of plays. And even if you don't make ballpark estimates of plays, you can make ballpark estimates of the results of your reading.

Bill Spight · **#32**

RobertJasiek wrote:

There is no such thing as consistently good endgame without effort of value calculation and comparison.

Well worth repeating.

Bill Spight · **#33**

daal wrote:

What is difficult is that even Big and Not Big depend on sente and followup.

The difficulty of figuring out sente was the main reason, IMHO, the go programmers failed to come up with a good evaluation function in four decades of collective effort. If they had, I doubt if today's bots would be using winrates.

However, humans understand the concept of sente, if fuzzily. Sente is important for evaluating local positions, because after a sente is played with sente, the resulting position has the same value as the initial position, and is easier to count. If you start with evaluating local positions, you will develop your sense of sente.

Quote:

One question though: How would you go about deciding between a sente move and a big gote move?

There's good news and bad news. Let's start with the bad news.

Bad news 1). You may want to save the sente for a ko threat.

Silver lining. If you are even thinking about playing the sente, that is probably not the case.

Bad news 2). You can only be sure that the big gote is right if it is quite big. Using the common way of counting — which I do not recommend, but which you probably are already using — the gote normally has to be bigger than both the sente move plus its threat. :shock:

Silver lining. A gote that big has probably already been played.

Now for the good news.

Good news 1). If you are not saving the sente as a ko threat (probably the case) and the threat of the sente is bigger than the big gote, so that your opponent should answer the sente, play the sente.

Good news 1a). That is very likely to be the case.

These are the basics. If we consider the rest of the board, other considerations apply.

There is a rule of thumb about doubling the size of a sente (using, cough, the common way of counting) in order to compare it with a gote. This actually applies to reverse sente. You can normally save a sente to use as a possible ko threat until just before your opponent will play the reverse sente. That's why it is useful for sente, as well.

Bill Spight · **#34**

zermelo wrote:

You said you will not have time to calculate during a game. I guess the usual thing is to calculate examples outside your games, and by that train your intuition to better and better guess the values quickly during a game. It's like with tsumego. The more you do them the better you get at guessing how to kill or live in a fast game.

I agree with the idea of non-play practice. Especially reviewing your own endgames.

But as for calculation, parrots can "count" up to 6 at a glance, i. e., tell the difference between 5 items and 6. So can humans. Even if we're not Rainman. :cool:

And if all we want is ballpark estimates, getting up to 30 in a second or two is not hard. We can also multiply. E. g., there's a territory bounded by stones on the third line up to the side star point. 2x10 = 20. Bingo!

Bill Spight · **#35**

EdLee wrote:

Quote:

It's better to make a hanging connection after the hane:

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . 3 9 5 7 8 . | $$ | . . . . 1 2 6 0 . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . . . . | $$ | . . , X . . , . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

Assuming the other player doesn't want to risk the ko, then you get more points.

Hmm... something seems off ? :scratch:

And

tenuki (above variation) causes different calculations.

Click Here To Show Diagram Code: [go]$$B var 2 $$ +-------------------- $$ | . . . . 5 3 4 . . | $$ | . . . . 1 2 6 . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . . . . | $$ | . . , X . . , . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +--------------------[/go]

If

simply replies with descend, then locally B has fewer points than B's hane & solid connect in var 2.

Click Here To Show Diagram Code: [go]$$B var 3 $$ +-------------------- $$ | . . . 3 . 4 . . . | $$ | . . . . 1 2 . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . . . . | $$ | . . , X . . , . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +--------------------[/go]

This

tiger's mouth is after :b3:

hane, but leads to yet different calculations
( but locally B still seems worse off than var 2 ):

Click Here To Show Diagram Code: [go]$$B var 4 $$ +-------------------- $$ | . . . 5 6 3 4 . . | $$ | . . . . 1 2 . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . . . . | $$ | . . , X . . , . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +--------------------[/go]

When making an estimate during play, I reckon the ko to be Black sente (unless I know better) and reply to it thus.

Click Here To Show Diagram Code: [go]$$B var 5 $$ +-------------------- $$ | . . . 5 . 3 4 . . | $$ | . . . . 1 2 6 . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . . . . | $$ | . . , X . . , . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +--------------------[/go]

This is obviously no worse for White than the hanetsugi.

Pio2001 · **#36**

daal wrote:

What is the easiest way to determine a rough value of a move in order to compare alternatives. I am more interested in "ballpark" than "correct." For starters, how about this: black to play: What are a, b and c worth?

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . a . . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . b . . | $$ | . . , X . . , . . | $$ | . . . . c O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

Hi Daal,
A first easy rule is to play first the moves that are very sente, then the reverse sente, then the double gote.

To know if a move is sente, you have to look where the sequence can stop. For example :

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . . . . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . a b . | $$ | . . , X . . . . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

In the above diagram, thinking that a is sente because White must answer b is a mistake !... because then, Black must answer White b.

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . a . . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . 1 2 . | $$ | . . , X . . 3 . . | $$ | . . . . . O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

Here, Black 3 is locally sente. But is it really sente ? White could ignore Black 3 and take a instead.

On the other hand, for me, a is really sente, because, in the diagram below, after move 2, black can stop there and play b

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . 1 2 . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . b . . | $$ | . . , X . . . . . | $$ | . . . . c O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

c is reverse sente, so it can be played next.

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . 1 2 . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X . 3 4 . | $$ | . . , X . . 5 6 . | $$ | . . . . 7 O . O . | $$ | . . . . . . . . . | $$ +-------------------+[/go]

If you are counting continuations, when black and white continuations are both gote, you can assume that the continuation is halfway between them. If one of the continuations only is sente, assume that it will be the one being played.

Another trick : if you have prisoners that need to be counted in parallel, you can switch to area counting (just count the number of intersections that change colour if White starts instead of Black, stones included). I'm not sure of the implications. Robert Jasiek talks about it in his book Endgame 2, but I have not yet read this chapter.

Gomoto · **#37**

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . . . . . . | $$ | . . . . 1 . . . . | $$ | . . , X . O , . . | $$ | . . . X . O . . . | $$ | . . X . , . O . . | $$ | . . . . X 5 3 6 . | $$ | . . , X 8 . 4 . . | $$ | . . 7 2 0 O . O . | $$ | . . . 9 . . . . . | $$ +-------------------+[/go]

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . 5 3 4 . . | $$ | . . . 9 X 2 6 . . | $$ | . . , X 8 O , . . | $$ | . . . X . O . . . | $$ | . . X . , 7 O . . | $$ | . . . 1 X X X O . | $$ | . . , X O . O . . | $$ | . . X O O O . O . | $$ | . . . X 0 . . . . | $$ +-------------------+[/go]

Click Here To Show Diagram Code: [go]$$B $$ +-------------------+ $$ | . . . . X X O . . | $$ | . . . X X O O . . | $$ | . . , X O O , . . | $$ | . . . X 2 O 4 . . | $$ | . . X . 3 X O . . | $$ | . . . X X X X O . | $$ | . . , X O . O . . | $$ | . . X O O O . O . | $$ | . . 1 X O . . . . | $$ +-------------------+[/go]

daal · **#38**

Bill Spight wrote:

daal wrote:

Alright, I give up. it was a stupid question.

What was a stupid question? Certainly not your original one.

Maybe the easiest way to get a ballpark figure is harder than you had hoped, but it still is within your capabilities.

But the place to start is with evaluating local positions. (At least for humans.

) Endgame books show how to do that.

Once you can make ballpark estimates of local positions, it is not difficult to make ballpark estimates of plays. And even if you don't make ballpark estimates of plays, you can make ballpark estimates of the results of your reading.

Yes, indeed harder than I hoped. I have started once again to look into miai counting. I watched [url="https://www.youtube.com/watch?v=ZbgQ9jvhZS0"]a video[/url] that was made during the 2012 go congress, and it explained some of the basic ideas, but I think it takes a lot of practice to be able to do during a game. Do you perhaps know of a set of exercises with easy problems? I think that might be a good place to start

RobertJasiek · **#39**

The youtube videos on miai counting or Sensei's miai values list are not an easy start. If you want problems rather than theory with easy examples, easy problems rather than kos and whatnot, more than the few problems in O's book and no Japanese book (I think John mentioned some, but I do not know whether the problems and answers are easy), you have two options: dig for Bill's problems here (not that many, not always easy, often with difficult extras such as value trees) - or wait a few months. If you cannot wait, what prevents you from reading examples as if they were problems?

Kirby · **#40**

daal wrote:

I'll tell you anyway. I am not good at calculating, and am even worse at calculating under pressure. I am 100% positive that any method that promises a correct evaluation is too hard for me to use in a game situation. For this reason, I am looking for something that I am able to use, with the hope that it would provide a right answer more often than not, and that this would be better than just guessing.

What sticks out to me here is the reference to look for a (quick) way to get to the "right answer".

For quite some time, I was depressed about AlphaGo. Regardless of study, talent, or effort, anyone can download the latest go ai and be capable of getting the right answer for a given position. What's the point then? Isn't coming up with the answer the essence of go?

I maintain that it is.

But for awhile, I missed something: it's not the answer that's the essence of go. It's COMING UP with the answer.

In other words, the mental process of evaluating and identifying a move is enjoyable and gives go meaning, independently of where you happen to place your stone.

So I suggest that, while it may be possible to come up with a shortcut to give you a semi-accurate answer, it may not give the same satisfaction as thinking deeply about a position, and coming to a solution- even if it's wrong!

Just my thoughts about it at the moment. YMMV

Values of moves

Who is online