Go problems don't bring any result?

Bill Spight · Post by **Bill Spight** » Thu Dec 15, 2016 12:41 pm

Seeing potential eyes

One skill for life and death problems is identifying possible eyes. There is research that compared the eye movements of good life and death solvers with the eye movements of others. (The usual research design would be to divide subjects into two equal groups, based on their ability to solve the problems.) The eyes of the weaker group focused on potential plays; the eyes of the stronger group focused on potential eyes. (OC, it is unlikely that there was no overlap.) Let's take another look at the problem from that standpoint.

Click Here To Show Diagram Code: [go]$$W Where are the eyes? $$ --------------------- $$ | . . . . . . . X . . $$ | O O . . . O O X . . $$ | O X O O O X X X . . $$ | X X X X X . . . . , $$ | . X . . . . . . . . $$ | . . . . . . . . . . $$ | . . . . . . . . . .[/go]

Well, there certainly seems to be an eye in the corner. The rest of the eyespace is fairly amorphous, though.

Click Here To Show Diagram Code: [go]$$W Potential eyes $$ --------------------- $$ | . . . . . . . X . . $$ | O O C 1 C O O X . . $$ | O X O O O X X X . . $$ | X X X X X . . . . , $$ | . X . . . . . . . . $$ | . . . . . . . . . . $$ | . . . . . . . . . .[/go]

But if White plays

the marked points are potential eyes.

Which did I see first, the move or the eye points? Hard to say.

Bill Spight · Post by **Bill Spight** » Fri Dec 16, 2016 1:51 am

Sub-problems, chunking, and other problem solving strategies

As the research on eye movements suggests, simply building a game tree is not the only way to solve go problems. Still using our example problem, let me discuss some other strategies.

It looks like there is an eye in the very corner, but let's take a closer look.

Click Here To Show Diagram Code: [go]$$B Corner eye? $$ --------------------- $$ | C 2 1 . . . . X . . $$ | O O 3 . . O O X . . $$ | O X O O O X X X . . $$ | X X X X X . . . . , $$ | . X . . . . . . . . $$ | . . . . . . . . . . $$ | . . . . . . . . . .[/go]

After

encloses the marked point, making an eye, but White cannot hold onto it because

puts the White stones in atari. OC, Black can only do so because

and

have two dame. So the original impression of an eye in the corner turns out to be something of an illusion. That eye is not secure.

Is this one of those silly variations? Not really.

is not played as the first move of the solution. Rather it is played as the first move in the attempt to take away the potential eye at the 1-1 point. And it succeeds, under current conditions.

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

That is, even if

and

are captured, as in this diagram, so that White gets an eye, it is not the distinct eye on the 1-1, and enlarging it may not allow White to live.

Click Here To Show Diagram Code: [go]$$B Corner eye? Var. 2 $$ --------------------- $$ | . X X O 6 7 9 X . . $$ | O O O 8 5 O O X . . $$ | O X O O O X X X . . $$ | X X X X X . . . . , $$ | . X . . . . . . . . $$ | . . . . . . . . . . $$ | . . . . . . . . . .[/go]

In fact, the throw-in,

, kills. After

or

at 8,

and

take away any potential second eye.

Click Here To Show Diagram Code: [go]$$B Corner eye? Var. 3 $$ --------------------- $$ | . X B O 8 6 7 X . . $$ | O O O 9 5 O O X . . $$ | O X O O O X X X . . $$ | X X X X X . . . . , $$ | . X . . . . . . . . $$ | . . . . . . . . . . $$ | . . . . . . . . . .[/go]

To prevent the second throw-in,

descends to the second line. However, that play succumbs to the atari,

, because of damezumari. As I used to joke,

was played with great foresight.

Well, it's late at night and I will have to pick this discussion up later. The first four moves of these variations, especially

, may be silly, but seeing the kill with

is pretty cool.

Oh, yeah!

does not have to be a throw-in, as you may have noticed, because of damezumari. One little factoid to file away.

It is a common problem solving strategy to break a problem down into subproblems. Here we have seen two subproblems: first, to take away the distinct eye on the 1-1, and second, to kill in the position after

. Neither of these arose as part of reading out a game tree. And, in fact, it may turn out that neither is relevant to the solution. However, both help us to understand the problem position better, to see its possibilities. And in general that will help us to solve the problem.

Another question arises, OC: Which is more important, solving the problem or understanding the position?

RobertJasiek · Post by **RobertJasiek** » Fri Dec 16, 2016 2:41 am

Bill, how do you use your three diagrams to solve the problem without tree reading? In particular, please explain this for different white replies. OC, White dies if he helps Black to kill the white group.

We know that there are different ways of solving problems, incl. e.g. guessing a successful move already achieving an aim within one move. Research on human eye movement is not needed to come to this conclusion. However, the question remains what ways do always work (in principle) and what ways do you apply when you use a way that does not always work as soon as you find out this when analysing a particular problem? Answer: you must fall back to a way that always works, such as Regular Reading. All talk about other ways is incomplete ALA you do not also specify some default, general way.

What does the mentioned research say when other observed ways do not work?

JFTR, a solution consists of the correct answer and the verification why it is correct. The process of solving might also go through unsuccessful analysis before. How do the other ways provide the necessary verification?

uPWarrior · Post by **uPWarrior** » Fri Dec 16, 2016 2:59 am

Robert: In your effort to show that Bill was wrong, you failed to even understand the broad points of his message.
You don't need to disagree with every post you read.

RobertJasiek · Post by **RobertJasiek** » Fri Dec 16, 2016 3:30 am

uPWarrior, I ignore the meta-discussions about no necessity of disagreeing with every post, your claim of my failure to understand parts of Bill's post and your claim of my effort to show that Bill was wrong.

You say that Bill's post has broad points and apparently you think they are worth pointing out. Can you please point out them so that those of us that might not have been fully aware of them yet can better appreciate them?

uPWarrior · Post by **uPWarrior** » Fri Dec 16, 2016 4:27 am

For instance, you immediately disregarded the research on eye movement and underlined that one has to fallback to reading as "any other way is incomplete".

Bill exemplified two alternatives that imply that one should start by understanding the problem better:

understanding the eye shape of the problem and understanding its key points;
chunking the problem into smaller pieces.

Can you get this understanding if you simply read out the tree? Of course not, reading and evaluating end (or intermediate) positions won't give you that.
This is particularly important given the anecdotal evidence that good tsumego solvers DO focus on understanding the problem first (i.e., their eyes move to the eye shape) instead of focusing on reading killing sequences (i.e., their eyes don't focus on potential plays).

This means that the time we spend staring at a problem should be translated into a better understanding of the position, not on a higher number of variations discarded.

Cassandra · Post by **Cassandra** » Fri Dec 16, 2016 4:49 am

uPWarrior wrote:... one should start by understanding the problem better ...

In my opinion, "understanding the problem better" (i.e. "identifying its shape issues and vital points") usually is strongly underestimated when discussing the "proper" way of solving tsume-go.

Starting with SHAPE will benefit the process of choosing candidate MOVEs for beginning your solution sequence.

RobertJasiek · Post by **RobertJasiek** » Fri Dec 16, 2016 6:03 am

uPWarrior, I have not disregarded the research on eye movement but commented that we do not need a research on eye movement to conclude that there are different ways of solving problems. Research on eye movement can be useful for some kinds of studies. If it is used for tracking a visual aspect of go problem solving, there must be proof on a relation between eye movement, focus on part of the board, view of the same or another part of the board and accounting of aspects of problem solving not represented by these aspects. In particular, focussing on one particular intersection does not exclude the possibility of viewing several intersections simultaneously and possibly with an asymmetric view around the focussed intersection.

Focussing eyes can be done at a rest point while the brain performs reading possibly independent of a currently focussed intersection.

I have not said that any other way was incomplete. Besides I do not say that Regular Reading was the only way of reading (a breadth first search reading would also be possible in principle but require much more temporary storage than the depth first search of Regular Reading). E.g., another complete way would be computer-like sampling of randomly chosen move sequences without repeating any already walked sequence. Although such ways can be imagined, they have serious disadvantages: a) very large numbers of sequences might have to be studied (often extremely many more than possible within reasonable time) and b) much more temporary storage is needed. I can imagine such alternative, general ways of solving problems but I have yet to hear of any (for tactical problems with a given, then static aim) that would be as applicable in general as Regular Reading.

Referring to research on eye movement tempts one into a wow factor but there is no indication that eye movement alone would be general enough for solving all problems in principle. Problem solving is not just one-layer pattern recognition. Eye movement does not correlate as identity between focus and thought about intersections (or stones).

Starting by understanding a problem better is not always the most efficient way. For some problems, it helps. For other problems, it wastes time, and starting right away with Regular Reading is the most efficient way if understanding a problem consists of the decision-making during Regular Reading and these decisions cannot be efficiently represented by other means, such as techniques.

Understanding the eye shape of the problem is not always the most efficient way to start. More specifically, it is not always the shape of eyespace that gives useful information but it can be more relevant to identify the initial eyespace and the potential eyespaces at all. Doing so is part of ignoring unrelated intersections and sequences to prepare reading to a manageable size. This is also very useful for Regular Reading, ALA one does not overlook relevant moves; in case of doubt, include them.

Understanding key points is not always the most efficient way to start. If there are obvious key points, it can be useful to start with them. If there are too many or none, then looking for key points is a waste of time and instead starting directly with Regular Reading is more efficient.

Chunking the problem into smaller pieces is not always the most efficient way to start. In particular, there are (ca. 50%) problems that cannot be (easily or at all) split by a regional or sequence-process-local way. Instead, another way is (only!) sometimes useful: reduction of an initially difficult problem to a known kind of problem (e.g., a nakade) in a currently studied subtree.

I have not suggested that other ways (such as techniques) would be useless. They are sometimes faster than Regular Reading, provided they permit verification wihout or with less Regular Reading than if only using it.

I thought that good problem solvers are neural nets that sample so much that reading is emulated but we really do not understand in detail how the programs think... I.e., we do not even know if the NNs do understand anything - we only know if they are successful empirically.

In general, there must be an early choice of whether to explore Regular Reading or other ways. Regular Reading automatically provides verification. Other ways might suggest a result but verification is often not automatically included; in this case, one also needs Regular Reading or highly sophisticated, specialised means as a means of verification.

Bill Spight · Post by **Bill Spight** » Sat Dec 17, 2016 2:17 am

Well, it's late at night again. Just a brief note.

RobertJasiek wrote:Bill, how do you use your three diagrams to solve the problem without tree reading?

Please note that solving the two sub-problems yields a portion of a depth first search tree.

However, the question remains what ways do always work (in principle) and what ways do you apply when you use a way that does not always work as soon as you find out this when analysing a particular problem? Answer: you must fall back to a way that always works, such as Regular Reading.

I suppose that by Regular Reading you mean a brute force depth first search of the game tree. In practice, that does not always work. In Gateway to All Marvels John Fairbairn points out that some lines of play were erroneously believed to be solutions to some of the classical go problems for centuries before they were refuted.

Elsewhere you bring up the question of efficiency and memory usage. Chunking addresses both of those. Also, consider this little problem.

Click Here To Show Diagram Code: [go]$$B Black to play $$ ------------------------------- $$ . . . . O . . . . . . . . . . . $$ . . . X O O . O O O . . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

Very easy. Not many branches of the game tree to read. Practical depth only 9 moves.

Yes, but only if Black solves the sub-problems of preventing a second eye and preventing seki. Otherwise brute force search suffers a combinatorial explosion. This is not atypical.

Also, it is efficient to do tree search in parallel. Humans are capable of that, but only subconciously. That's one reason I talk so much about seeing.

John Fairbairn · Post by **John Fairbairn** » Sat Dec 17, 2016 2:57 am

Yes, but only if Black solves the sub-problems of preventing a second eye and preventing seki. Otherwise brute force search suffers a combinatorial explosion. This is not atypical.

Excellent example, Bill. I've seen this sort of thing many times because I tried writing a tsumego program once. It was very frustrating because there were so many problems (the vast majority, in fact) that were trivial to my way of human thinking but hard for the computer. I was using test cases smaller than yours yet typically nothing less than 11-ply would work for brute search.

More than that, though, I think we need to hold RJ's other claims up to scrutiny. He says only 50% of problems are amenable to being solved by a known technique. I challenge that on the grounds that it can at best only be true for the start of the problem. But many problems involve techniques that are applied deeper in the tree. In fact they can't be done in practical terms without a technique. The most obvious is the visual inspection of a nakade shape. For humans that is trivial. For brute search nakade shapes add an Amazonian forest of bushes to the tree.

RJ also makes much of the argument that looking for a technique wastes time if no technique is available, and so is inefficient. That seems preposterous. That claim can only stand up for an individual problem where there is no technique available anywhere (and very, very few cases of this exist, as I have just said). The loss of time can only be very small, often milliseconds. And as you rightly say, there is a major difference here between 'looking for' and 'seeing'. But taking problems as a corpus, in the vast majority of cases a technique does exist and the saving in time and effort that makes is enormous. So even if the very questionable 50% figure is correct, the loss on the negative half is hugely outweighed by the gains on the positive half. Which means we shouldn't be talking about 50% in the first place, really. A ship that's keeling over with 50% of people on the port side and 50% on the starboard side still has much more than a 50% chance of going to Davey Jones' locker.

Interesting to see that the meaning of Regular Reading wasn't obvious to you, either. I took it to mean a Quotidian Quotient - doing a tsumego every day. A natural laxative

RobertJasiek · Post by **RobertJasiek** » Sat Dec 17, 2016 3:08 am

OC, Regular Reading is not brute force reading.

Regular Reading is tactical reading for a given start position and aim, explores variations, ignores obvious failures and obviously inferior moves (in case of doubt, do not ignore), at a branch goes backwards if a) a successful next move has been found or b) all interesting (neither obvious failure nor obviously inferior) next moves have been identified as failures, else go forwards with a not previously explored interesting next move to the next branch or a terminal position, a position is terminal if it is obvious whether the aim is fulfilled there, propagate collected information backwards, stop when the start position's branch is decided. Regular Reading can be refined by accelerating techniques, such as using prior status knowledge about non-terminal positions or identifying reversion to previously explored positions. The method and simplifying techniques I have described elsewhere in great detail.

EDIT: Answer to John: http://www.lifein19x19.com/forum/viewto ... 17&t=13879

Cassandra · Post by **Cassandra** » Sat Dec 17, 2016 4:21 am

In my opinion, identifying the so-called "obvious" features of a tsume-go problem requires doing some SHAPE analysis beforehand.

Click Here To Show Diagram Code: [go]$$B Black to play $$ ------------------------------- $$ . . . . O . . . . . . . . . . . $$ . . . X O O . O O O . . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

S h a p e A n a l y s i s

Click Here To Show Diagram Code: [go]$$B White's potential eye space $$ ------------------------------- $$ . . . . O C C C C C . . . . . . $$ . . . X O O C O O O . . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

White's potential eye space consists of six empty points.

Click Here To Show Diagram Code: [go]$$B Dead eye shape (1) $$ ------------------------------- $$ . . . . O T S T . . . . . . . . $$ . . . X O O T O O O . . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

At the left, there is a dead triangle four-point eye shape visible which central point is still unoccupied.

Click Here To Show Diagram Code: [go]$$B Dead eye shape (1) $$ ------------------------------- $$ . . . . O s S s . . . . . . . . $$ . . . X O O s O O O . . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

If White occupied that point, she would divide her eye space into three parts, all of which are sure eye points.

Click Here To Show Diagram Code: [go]$$B Dead eye shape (2) $$ ------------------------------- $$ . . . . O . T S T . . . . . . . $$ . . . X O O . O O O . . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

A dead three-point eye shape in the centre.

Click Here To Show Diagram Code: [go]$$B Dead eye shape (3) $$ ------------------------------- $$ . . . . O . . T S T . . . . . . $$ . . . X O O . O O O . . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

Another dead three-point eye shape at the right.

Click Here To Show Diagram Code: [go]$$B Danger of seki $$ ------------------------------- $$ . . . . O . S S S . ? . . . . . $$ . . . X O O ? O O O ? . X . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

Occupying all (at least three) vital points of your opponent's eye potential space might not always lead to success.

Click Here To Show Diagram Code: [go]$$B Looking forward $$ ------------------------------- $$ . . . . O . . . . M M . . . . . $$ . . . X O O . O O O . . Y . . . $$ . . X . X O O X O X . X . . . . $$ . . . . X X X X X X . . X . . . $$ . . . . . . . . . . . . . . . . $$ . . . . . . . . . . . . . . . .[/go]

Black's marked stone at the right is looking forward to supporting any friendly stone at the crossed intersections.

RobertJasiek · Post by **RobertJasiek** » Sat Dec 17, 2016 5:25 am

Cassandra, identifying obvious failures does not necessarily involve shape analysis. E.g., if there are several connected moves on the outside from the same direction of potential reduction, the moves further to the outside can be identified without shape analysis. Instead, it is a matter of functional analysis combined with identifying greater impact on the inside.

John Fairbairn · Post by **John Fairbairn** » Sat Dec 17, 2016 5:31 am

Regular Reading is tactical reading for a given start position and aim, explores variations, ignores obvious failures and obviously inferior moves (in case of doubt, do not ignore), at a branch goes backwards if a) a successful next move has been found or b) all interesting (neither obvious failure nor obviously inferior) next moves have been identified as failures, else go forwards with a not previously explored interesting next move to the next branch or a terminal position, a position is terminal if it is obvious whether the aim is fulfilled there, propagate collected information backwards, stop when the start position's branch is decided. Regular Reading can be refined by accelerating techniques, such as using prior status knowledge about non-terminal positions or identifying reversion to previously explored positions. The method and simplifying techniques I have described elsewhere in great detail.

I still can't see what is regular about that. More to the point, I can't see anything there that strong players (or disciplined players) can't do already (they may not always do it for reasons such as time constraints or fatigue). That sort of description exists in the literature for chess, shogi and go, and is obviously well known in games programming (with many more pruning refinements, of course). Are you claiming something novel, or are you just trying to be systematic?

But the bigger question is: for amateurs who simply want to reach a modest strength (e.g. daal's shodan quest) which method is better? Brute force with pruning is unnatural and therefore hard for a human, and since amateurs usually don't want to invest the sort of time needed to train that ability to a worthwhile level, it seems obvious to me that shape and technique training is the way to go for them.

What I would add, though, is that go is a game that is very easy to play to a modest standard on shape principles alone. That is why blitz play is so perniciously bad in go compared with chess, where your mistakes are found out sooner and in a more obvious way (tactical blunders). Choosing candidate moves then doing a tiny bit of brute force (even 2-ply helps a lot) as a check makes a huge difference across the amateur spectrum. Tsumego can help with that.

RobertJasiek · Post by **RobertJasiek** » Sat Dec 17, 2016 5:53 am

Regular about regular reading is that it is the regular, standard method of tactical reading. I might also have called it standard reading.

In fact, the very purpose of describing regular reading is describing what every above-beginner player should usually be doing as tactical reading when exploring variations for the sake of a given aim.

What might be novel about my detailed description elsewhere is a) being so detailed and clear about what is the method for, AFAIK, the first time for go players (some have said that mine would be clearer than the one for chess players), b) stating and explaining the inherent and other major means of simplification clearly, c) being systematic for go players (rather than computer go theorists and prorgrammers with their different, algorithm-orientated simplification needs). Nothing is new, but having written down the theory clearly for go players is novel. Not even test reading is new, I would guess, but OTOH, I did not see it being mentioned in the go world before.

Brute force NEVER is good in practice. Regular reading is very much more restricted in reading volume than brute force with just some pruning.

Your suggestion towards shape and technique is very bad ALA you do not at least also include regular reading as central. The stronger I have become, the less useful shapes have become. For up to ca. 5 kyu, shapes are more useful as temporary means ALA knowledge and reading skill are still weak. Shapes are bad excuse not to read and verify assumptions. The major remaining use of shapes is early, familiar terminal positions with known status for known side conditions.

I do not buy it that go would be played at modest standard solely on shape principles. Those trying such all(!) fail because of making big beginner mistakes and failing to read at least the absolutely necessary simple variations.

Life In 19x19

Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?

Re: Go problems don't bring any result?