Miai values

[snorri]

I still don't know how this is tied to optimal play, but I guess it'll be more efficient to try to figure out myself than ask questions here.

For local tally differences of 1 (one-sided sente) or 2 (double gote) one can muddle along with deiri values fine. I like the fact you can often add miai values, but that's not enough to sell most people. But I think where they really start to be of interest is when the local tally difference is some other number, like 3 or 4, which often happens in ko positions.

See quickly sections 8 and 9 in Charles Matthews' Setpiece Ko series, part 6 on Ko Accountancy. It shows something practical.

For beginners, just knowing that a half-point ko is not a half a point can help build better habits in the late endgame. A beginner can gain much more than that through usual forms of study, playing a lot, etc., but I don't think it's so wrong to invest a little in the future. That's the way I think of this stuff. Not critical now, but an investment in the future, when improvement will be so hard to come by that every little thing counts.

[John Fairbairn]

There is a conceptual difference between counting to assess the present value of an unsettled position and counting to assess the value of the best move(s) in that position. (A mathematician would probably dispute this distinction; once you know how to perform one count, you can derive the other.) Knowing how to calculate the value of a move is arguably more important to a Go player than knowing how to calculate the value of a position. For position evaluation, miai counting is necessary. But for move selection, I believe deiri counting is more widely (exclusively?) used by professionals.

Thank you. You seem largely to be confirming my intuition except for this phrase: "For position evaluation, miai counting is necessary."

I dispute that. If the mathematicians would kindly resist the urge to toss in caveats such as "A mathematician would probably dispute this distinction" - the kind of mud-creating utterance which is really aimed at other hippos and not curious meerkats like me - what I see, perching as high as I can on my little hind legs, is a near total void of miai counting in pro position evaluation.

To check my intuition, I have just whizzed through the book that started all the recent talk of boundary plays, Yi Ch'ang-ho's "How I evaluate positions", which is obviously 100% about position evaluation. As far as I can see there is not a single mention of miai counting. There is a brief chapter on deiri counting, but in the many walkthroughs that show how his methods apply to actual full-board positions, even deiri counting does not get much of a look in.

Since, as I observed before (and have not so far been contradicted), miai counting is a modern(ish) construct by mathematical go amateurs and does not appear to be part of mainstream thinking among pros even now, I see this as confirmation that miai counting is not necessary, for position evaluation or anything else. I might add that it's also a concept that belongs to Japanese amateurs, and Korean and Chinese go appear not to be "afflicted" in the same way.

It is true that O Meien wrote about miai counting in the guise of "absolute counting", a better term. It is true that he devotes the first part of his book to an example where (he claims) using absolute counting is used to refute the usual play by amateurs in a complex ko position, but (a) it is a very heavy handed example and poorly written, (b) there is no indication that pros ever make the same mistake, and (c) since he claims to have invented absolute counting, we can assume pros hadn't used it anyway (and he doesn't tell us what they did before he came along).

For the curious, if Yi's book does not use miai counting at all and barely uses deiri counting, what does he use? The one constant theme is marking off boundaries with Xs and then simply counting the prospective territories inside. (The word 'prospective' is very important.) What makes Yi's X-marking very different from ours (including Jasiek's) is that he is more accurate, in two ways. One is down to simple reading - he sees the tesujis and need for repair moves that you and I might miss. The other is, slightly paradoxically, is to abandon numbers and to rely on experience. What you see, therefore, in this book are comments along the lines of "that area looks like it's worth 8 points but as it's too easy to take away the base and chase it into the centre, we count it as zero." (The latter step is much more important, because if you make mistakes with e.g. repair moves, these probably balance out between Black and White.)

The next step, once Xs have been marked and prospective territories counted up, is to decide whether a strategy of increasing territory or increasing thickness is called for. If the former, the extra territory will often be in the centre, and this is where Yi seems to have a special skill. None of this process depends on miai counting. (Slightly tongue in cheek, I'd say you can get a good approximation of Yi's centre counts by taking your own and halving them, but of course the real skill lies in creating a dynamic flow that ensures you get the extra points without causing collateral damage and, ideally, keeping sente, and as I pointed out in yet another thread, quoting Sugiuchi, the trick there is to get the sequentially right order of moves (as opposed to moves of the right order of size).

So, again I say, for all practical purposes miai counting seems to be a total waste of time and ranks with rules discussions as one of the two major unnecessary distractions of amateur go.

[Bill Spight]

I still don't know how this is tied to optimal play, but I guess it'll be more efficient to try to figure out myself than ask questions here.

I responded to your question here: http://www.lifein19x19.com/forum/viewto ... =15&t=8832

For local tally differences of 1 (one-sided sente) or 2 (double gote) one can muddle along with deiri values fine. I like the fact you can often add miai values, but that's not enough to sell most people. But I think where they really start to be of interest is when the local tally difference is some other number, like 3 or 4, which often happens in ko positions.

See quickly sections 8 and 9 in Charles Matthews' Setpiece Ko series, part 6 on Ko Accountancy. It shows something practical.

Unfortunately, Matthews relies in part upon Tavernier, but Tavernier made a mistake. He did not take the rest of the board, aside from the ko and threats, into account. You can do that through the concept of temperature.

(Historical footnote: I had worked out the answer, including the use of temperature, although I did not know the term, in the 1970s. In the 1990s Tavernier sent a paper to Professor Berlekamp, who passed it on to me. I wrote to Tavernier, telling him that he should take account of temperature, apparently to no effect. )

Edit: (Footnote 2. One way to take account of temperature is shown in the ko threat scaffold in the Berlekamp, Mueller, Spight paper : http://citeseerx.ist.psu.edu/viewdoc/su ... .1.34.6699 )

[jts]

John, these are the two differences I see between miai counting and deiri counting. Perhaps these are actually not present in the professional conception of deiri counting, but are only extrapolations by western amateurs. Any guidance you can offer would be very useful.

1. In deiri counting you simply count the swing in points. Whether a branch on the tree is gote or sente is irrelevant to the magnitude of the swing. In miai counting, points gained in gote are worth less than points in reverse sente, and points gained in ko are worth less than points gained in gote.

2. Sente moves have a deiri value, but are worth 0pts in miai value (because points are tracked relative to a baseline where everyone plays their own sente moves).

Now, I'm happy to believe that pros count swing values exclusively. But I have trouble believing that pros are unaware of the difference between gote and reverse sente, and that they fail to take this into account in their play. And certainly professional ko fights show better than anything else that pros understand how to count the value of ko moves. And the professional commentaries I've read talk about the endgame in a way that strongly implies that the commentating pro expects each player to get his own sente moves, and counts loss of points in the endgame relative to that baseline.

So I grant you that tesuji and attack are more more fun and useful than counting. But what is the force of your claim that pros prefer deiri counting? Either they use "deiri" in a different way than we do (such that their version of deiri includes the useful features of miai or absolute counting), or they are taking the deiri values and doing mental arithmetic to turn them into miai values on a case-by-case basis.

[daal]

For the curious, if Yi's book does not use miai counting at all and barely uses deiri counting, what does he use? The one constant theme is marking off boundaries with Xs and then simply counting the prospective territories inside....

The next step, once Xs have been marked and prospective territories counted up, is to decide whether a strategy of increasing territory or increasing thickness is called for.

That ol' Yi Ch'ang-ho. Always taking the easy way out.

[Bill Spight]

what I see, perching as high as I can on my little {meerkat} hind legs, is a near total void of miai counting in pro position evaluation.

Me, too.

Since, as I observed before (and have not so far been contradicted), miai counting is a modern(ish) construct by mathematical go amateurs and does not appear to be part of mainstream thinking among pros even now, I see this as confirmation that miai counting is not necessary, for position evaluation or anything else. I might add that it's also a concept that belongs to Japanese amateurs, and Korean and Chinese go appear not to be "afflicted" in the same way.

It also seems to me that pros do not make much use of miai counting. However, I do not believe that it is not part of their mainstream thinking. Back when you and I learned to play go, yose books typically included a section on miai counting. Even if those books were ghost written by amateurs, I do not believe that they snuck miai counting in under the radar of the nominal pro authors. It is true that deiri counting is customary, but it is in the guise of doubling the miai values for everything except regular gote (without saying that that is what is going on). Evaluating a simple ko as 2/3 of the swing is an example of that.

It is true that for static evaluation of go positions, neither miai counting nor deiri counting is necessary. Problems arise, however, with the "anything else". You may recall that, at the International Conference on Baduk that we both attended some years ago, there was a presentation that was canceled at the last minute. I do not remember exactly what it was about, and I do not have the proceedings handy to check, but it was in the realm of play evaluation, and the author made a mistake that I do not think that he would have made if he had understood miai counting. The problem with deiri counting, in a nutshell, is that unless it is understood in terms of miai counting, it is conceptually incoherent. And that leads to mistakes with the "anything else". When I read the guy's paper I was in a quandary. Both Berlekamp and I would have been in the audience, and one or the other of us would have had to point out the error, which would have been embarrassing, to say the least. Fortunately, the presentation was canceled. A good example of the problems that arise with deiri counting is in the otherwise excellent book, "All About Ko". See my review: http://senseis.xmp.net/?BillSpight%2FReviewOfAllAboutKo

As, I suppose, a charter member of the League of Cruel Mathematicians, I went on a campaign during the 1990s of introducing miai counting to Western go. My motivation was not to help professionals, but amateurs. What I saw, and had seen over the years, was Western amateurs making mistakes based upon deiri counting. More about that later, perhaps, but I do not have much time now.

[John Fairbairn]

John, these are the two differences I see between miai counting and deiri counting. Perhaps these are actually not present in the professional conception of deiri counting, but are only extrapolations by western amateurs. Any guidance you can offer would be very useful.

1. In deiri counting you simply count the swing in points. Whether a branch on the tree is gote or sente is irrelevant to the magnitude of the swing. In miai counting, points gained in gote are worth less than points in reverse sente, and points gained in ko are worth less than points gained in gote.

2. Sente moves have a deiri value, but are worth 0pts in miai value (because points are tracked relative to a baseline where everyone plays their own sente moves).

Now, I'm happy to believe that pros count swing values exclusively. But I have trouble believing that pros are unaware of the difference between gote and reverse sente, and that they fail to take this into account in their play. And certainly professional ko fights show better than anything else that pros understand how to count the value of ko moves. And the professional commentaries I've read talk about the endgame in a way that strongly implies that the commentating pro expects each player to get his own sente moves, and counts loss of points in the endgame relative to that baseline.

So I grant you that tesuji and attack are more more fun and useful than counting. But what is the force of your claim that pros prefer deiri counting? Either they use "deiri" in a different way than we do (such that their version of deiri includes the useful features of miai or absolute counting), or they are taking the deiri values and doing mental arithmetic to turn them into miai values on a case-by-case basis.

As I have said, deiri is simply what I see over and over again in commentaries wile I never see miai referred to. Bill has more or less confirmed that, and I can add another snippet: In his "Encyclopaedia" Hayashi gives a longish description of deiri counting with a couple of diagrams. There is no mention of or cross-reference to miai counting in that description. Under the separate entry for miai counting he simply says, "A method of calculating boundary plays by counting while comparing [miai] areas of the same value and offsetting them against each other." I imagine that definition would make a Cruel Hippo shudder, and as you can see it gives no indication of the method's application or worth. As Bill remarks, books on boundary plays do tend to have a section on miai counting, but in my view it is usually self contained, and maybe there only to provide ballast for the mathematically minded. My impression is that if you skip that chapter, you rarely miss out on much in practice.

None of that means that pros do not recognise the need for qualification of their deiri counts using sente and gote. My point is simply that they seem not to treat miai counting as a full-blown method or theory (unlike, say, SL), but instead they just apply a couple of tweaks as and when needed. Pros are nothing if not practical. (Even in the case of books, those that have a section on miai couting often have a separate and more practical section on counting kos.)

On the specific point of reverse sente, although the usual Japanese term nowadays is gyaku yose, I have seen about 5 or 6 rather different terms for it, especially in pre-modern times. This suggests to me, again, that there has been no all-encompassing theory, but instead players have been taught about it ad hoc, and different schools have used different terms.

At the risk of going off at a tangent, I haven't said anything about tesuji and attack. I referred to reading, but that was (most often) at a rather simple, sub-tesuji level - remembering that repair moves are needed, for example. I also mentioned a dynamic strategy stage, but I also said that a decision is made here between going for more territory (which may or not imply attack) or more thickness (solidity, or defence).

A reminder that Genan's early 19th century primer on deiri counting appears in New In Go, item 21. I vaguely recall that the online version is only a partial version of what is on the GoGoD CD, but it will be enough to show the practical approach pros seem to use (e.g. adding a wee bit to counts for aji or ko threats).

[snorri]

Unfortunately, Matthews relies in part upon Tavernier, but Tavernier made a mistake.

John is right. You are cruel. Not for the reason he states, because now there's a danger you'll ruin my vacation. Or maybe that is the reason he states?

What a rabbit hole go has been. A co-worker a long time ago told me not to learn it, but did I listen? No. Now what junk fills my head! SGF Parsers. Obsolete josekis. Poorly understood bits of phrases in 3 different Asian languages. Araban's ever-changing avatars. KGS ratings math. Now, sadly, CGT.

Mixing metaphors, the blue pill doesn't work any more if I take it now, does it?

[Bill Spight]

Unfortunately, Matthews relies in part upon Tavernier, but Tavernier made a mistake.

John is right. You are cruel. Not for the reason he states, because now there's a danger you'll ruin my vacation. Or maybe that is the reason he states?

What a rabbit hole go has been. A co-worker a long time ago told me not to learn it, but did I listen? No. Now what junk fills my head! SGF Parsers. Obsolete josekis. Poorly understood bits of phrases in 3 different Asian languages. Araban's ever-changing avatars. KGS ratings math. Now, sadly, CGT.

Mixing metaphors, the blue pill doesn't work any more if I take it now, does it?

For a reasonable approximation, just remember to have each player make the same number of plays.

E. g., let K stand for a move in the ko, Wi for a White threat, Bi for a Black threat, and T for the temperature (assumed to stay constant).

Line #1: Black takes ko, White plays threat, Black wins ko, White completes threat.

Result: 2*K - 2*W0 (White's largest threat).

Line #2: Black takes ko, White plays threat, Black answers threat, White takes ko, Black plays threat, White wins ko, Black completes threat, White plays elsewhere. (That last play is what Tavernier forgot about.)

Result: 2*B0 - K - T

Comparison: 3*K + T >?< 2*B0 + 2*W0

T, of course, is the miai value of the largest play besides the ko and threats. All of the variables are miai values.

[mitsun]

John, these are the two differences I see between miai counting and deiri counting. Perhaps these are actually not present in the professional conception of deiri counting, but are only extrapolations by western amateurs. Any guidance you can offer would be very useful.

1. In deiri counting you simply count the swing in points. Whether a branch on the tree is gote or sente is irrelevant to the magnitude of the swing. In miai counting, points gained in gote are worth less than points in reverse sente, and points gained in ko are worth less than points gained in gote.

2. Sente moves have a deiri value, but are worth 0pts in miai value (because points are tracked relative to a baseline where everyone plays their own sente moves).

1) Sente and gote are important in deiri counting. The evaluation of a move is not complete without a qualifier like "5 points in gote" or "3 points in sente". Also, reverse sente plays are considered to be worth twice their point swing value.

2) The end result is the same -- when the position is counted, the points are assigned to the side who can get them in sente.

[jts]

1) Sente and gote are important in deiri counting. The evaluation of a move is not complete without a qualifier like "5 points in gote" or "3 points in sente". Also, reverse sente plays are considered to be worth twice their point swing value.

2) The end result is the same -- when the position is counted, the points are assigned to the side who can get them in sente.

That's certainly how I understand deiri counting - you have to mentally adjust it to miai counting to make it useful. (You don't mention ko, but again, I assume pros are perfectly good at adjusting deiri values to account for moves in ko.) I'm not saying there's anything wrong with using deiri counting if you're really good at mental math, or you learned that way and it's not worth changing. But if miai counting is just an easier way to tally up the same arithmetic, why does JF see it as meaningful if the pros use deiri counting? And the answer seems to be that if professionals don't see any value in having a systematic theory of the endgame, there must not be anything valuable to be found there. (Sort of like the story of the economist who refused to pick a twenty off the sidewalk: "If it were really a twenty dollar bill, someone else would have taken it already!")

[John Fairbairn]

Jts : I have a feeling you may subconsciously be trying to make what happens in pro play fit your understanding of deiri, while your conscious brain is making an effort to get at the facts - a common enough problem of having to unlearn something that we all face.

Assuming that's somewhere near the truth, my further guess is that you may be overlooking the process Yi gave under the three ways of counting, and listed under the precise method. As I understand it, he is effectively making the qualifications you are seeking by encapsulating them there - and in a very practical way, I think.

[Shinkenjoe]

Unfortunately, Matthews relies in part upon Tavernier, but Tavernier made a mistake.

John is right. You are cruel. Not for the reason he states, because now there's a danger you'll ruin my vacation. Or maybe that is the reason he states?

What a rabbit hole go has been. A co-worker a long time ago told me not to learn it, but did I listen? No. Now what junk fills my head! SGF Parsers. Obsolete josekis. Poorly understood bits of phrases in 3 different Asian languages. Araban's ever-changing avatars. KGS ratings math. Now, sadly, CGT.

Mixing metaphors, the blue pill doesn't work any more if I take it now, does it?

Just keep on drivinng and enjoy!

[Bill Spight]

For local tally differences of 1 (one-sided sente) or 2 (double gote) one can muddle along with deiri values fine. I like the fact you can often add miai values, but that's not enough to sell most people. But I think where they really start to be of interest is when the local tally difference is some other number, like 3 or 4, which often happens in ko positions.

See quickly sections 8 and 9 in Charles Matthews' Setpiece Ko series, part 6 on Ko Accountancy. It shows something practical.

Unfortunately, Matthews relies in part upon Tavernier, but Tavernier made a mistake. He did not take the rest of the board, aside from the ko and threats, into account. You can do that through the concept of temperature.

My apologies to Charles Matthews. Taking another look I see that he did point out the flaw with Tavernier's model. Somehow I had gotten the impression years ago that he had not pointed it out.