Life In 19x19

Thanks everyone, some comments:

perceval wrote:@dts:
Out of curiosity, and if it is not secret, how does your current tenuki detection algo work today ?

Actually, it's very similar to what you outlined. There were ...a few tweaks to get it to behave reasonably. The algorithm you suggest has a potential problem when joseki are played out late in the game and happen to come near another stone which is near another stone which is near... etc.

John Fairbairn wrote:

Tenuki means to play somewhere else.

Not really. The Japanese is a noun not a verb. The verb usage is te wo nuku or tenuki suru. More important, it means skipping a move. ...

Thank you for that, I've been using the word wrong (surprise). It seems it is related to what a pro said to me after a simul game, "Why didn't you answer me?" Perhaps I'll switch to using the phrase "non-local" or "played elsewhere" or some such.

RobertJasiek wrote:Although, in the strict sense, it depends on defining "local", there are (in my new book) pretty good criteria of conditions when playing elsewhere is attractive. Conditions such as this: "The set of local groups is stable."

Now, I propagate that the player chooses(!) what to consider as groups or belonging to a set of local groups. Hence locality can be derived from those given groups. Well, in principle. In practice one also needs some convention of how to form a local enviroment around a known, given group. E.g., the group plus all its liberties and maybe plus further (empty) intersections "in between". We get a locale. Playing outside the locale is then considered a tenuki.

(Everything needs to be updated every move.)

This is the sort of thing I was hoping for... except I think this may be computationally infeasible for my purposes. I'm not even sure there's a known solution for determining the set of local groups.

Bill Spight wrote:...
Like sente and gote, tenuki depends on the concept of locality. Towards the end of the game, it is possible to define independent regions of the board (except for ko fights, OC), and then tenuki is clear. Early on, however, regions of the board are seldom completely independent. Then what is considered tenuki is fuzzy. If a play is tactically related to the last play, it is not tenuki, no matter how far away it is on the board. If it is not tactically related, it is tenuki, no matter how close it is.

This is a great definition for a human, but "tactically unrelated" is not easy to translate into an algorithm.

-----

Perhaps I should have said a bit more originally on what I'm trying to do-- I potentially need to answer this question several hundred times per move per pro game, in all 65,000+ pro games of GoGoD. I only build my database occasionally, so it doesn't have to be super fast, but it does need to complete in my lifetime. So, basically I'm hoping for a definition that could be applied with reasonably accurate results by a 35kyu...

-----

Alright, well, I'm not sure what I'll end up doing, but a related question: I have considered giving an "urgency" rating to each position, which would be basically "when this position appears in pro games, how many moves on average pass before the players play locally again? So 0.0 would indicate that pros always play locally immediately, and larger numbers indicate less urgent positions (pros play N moves elsewhere before coming back to this position). Does this seem like an worthwhile metric?

Thanks for the input

daniel_the_smith wrote:not even sure there's a known solution for determining the set of local groups.

Set a purpose, then we determine the local groups. E.g., a purpose might be, given a particular string, determine all existing strings of which each considered alone can be connected to the string. Then we take the intersections of the string and those strings as our locale. (For other purposes, other definitions are possible, of course.)

"tactically unrelated" is not easy to translate into an algorithm.

It is easy but finding an algo with a sufficiently low computational complexity might be the essential problem.

I have considered giving an "urgency" rating to each position, which would be basically "when this position appears in pro games, how many moves on average pass before the players play locally again?

You can find a presumably simpler one in Joseki Vol. 2 Strategy, ch. Unrest Model. It distinguishes the unrest levels 0, 1, 2 and greater than 2. No fractions and no pro game databases are needed.

just went on dailyjoseki and i had the position that i tried to reproduced in my earlier post but did not exaclty remenbered:

Click Here To Show Diagram Code

[go]$$c
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 5 . . . . . W . . . . . . b . . |
$$ | . . . . . . . . . W B . . . . . . . . |
$$ | . . . 1 . 4 . 6 9 X . . . . . X . . . |
$$ | . . . . 3 2 . . 7 . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

here the marked stones are not in the joseki and appear after and before ... which is weird ..

I agree that for late joseki you risk having the whole board as a local group. which is not good.
M
You can add (proably did it already) a cutoff wit the distance to the starting move or the quadrant / half board where the action takes place in. maybe here the 3 stones are considerd as tenuki because the play are on the middle of the board ? maybe This limit should go a little further than the middle of the board considering this example ?

another idea:
Maybe you can infer something by comparing games with the same joseki with each other (the approach here is purely within one game whereas each joseki happend at least 3 times in your db) ie if the joseki here happens only when there is a stone at K4 or around then this stone is definitly part of this joseki pattern even if it was there before and any contact move on it are part of the sequence.



i dont know if its worth it too you but can you extract the other games with this pattern ? are they many without the marked exchange (if all games have this echange then it is also an hint that it is part of the sequence regardless of the local context of teh games)


and maybe just drop examples where there are too many adjacent big groups ?
maybe in that case the joseki variation would be too specific to be useful out of context ?

RobertJasiek wrote:

I have considered giving an "urgency" rating to each position, which would be basically "when this position appears in pro games, how many moves on average pass before the players play locally again?

You can find a presumably simpler one in Joseki Vol. 2 Strategy, ch. Unrest Model. It distinguishes the unrest levels 0, 1, 2 and greater than 2. No fractions and no pro game databases are needed.

That's fine (and I am interested in picking up a copy of your book, is it available in the US?), but I want to actually measure things, as opposed to theorize about them (IOW, we almost certainly have different purposes here). Assuming I write my code correctly, a measurement would not be subject to errors of opinion (I would be very surprised if you have a procedure such that a 30 kyu could come up with the same "unrest level" that you would for a given position). I'm asking whether the urgency measurement I described would be useful.

I'm not sure what all is involved in your "unrest level", but if it's a comparable figure this could possibly be a test for it. (i.e., how well does it model pro play?) Which may or may not be something you care about.

@perceval, I'm still thinking about your post...

perceval wrote:just went on dailyjoseki and i had the position that i tried to reproduced in my earlier post but did not exaclty remenbered:

$$c
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 5 . . . . . W . . . . . . b . . |
$$ | . . . . . . . . . W B . . . . . . . . |
$$ | . . . 1 . 4 . 6 9 X . . . . . X . . . |
$$ | . . . . 3 2 . . 7 . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$c
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 5 . . . . . W . . . . . . b . . |
$$ | . . . . . . . . . W B . . . . . . . . |
$$ | . . . 1 . 4 . 6 9 X . . . . . X . . . |
$$ | . . . . 3 2 . . 7 . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

here the marked stones are not in the joseki and appear after and before ... which is weird ..

Just to explain this real quick: the marked stones suddenly appear because in all 8 games where black played (the bad shape) 9, those stones were already on the board.

Although, I find white's reported next move very odd:

daniel_the_smith wrote:

RobertJasiek wrote:Although, in the strict sense, it depends on defining "local", there are (in my new book) pretty good criteria of conditions when playing elsewhere is attractive. Conditions such as this: "The set of local groups is stable."

Now, I propagate that the player chooses(!) what to consider as groups or belonging to a set of local groups. Hence locality can be derived from those given groups. Well, in principle. In practice one also needs some convention of how to form a local enviroment around a known, given group. E.g., the group plus all its liberties and maybe plus further (empty) intersections "in between". We get a locale. Playing outside the locale is then considered a tenuki.

(Everything needs to be updated every move.)

This is the sort of thing I was hoping for... except I think this may be computationally infeasible for my purposes. I'm not even sure there's a known solution for determining the set of local groups.

There is a paper about the possible omission number of a group (i. e., how many times you can tenuki), with an algorithm to determine the number, based upon the judgement of a Japanese amateur 6 dan. OC, you have to be able to say what a group is before you can determine its PON. I'll look it up.

Bill Spight wrote:...
Like sente and gote, tenuki depends on the concept of locality. Towards the end of the game, it is possible to define independent regions of the board (except for ko fights, OC), and then tenuki is clear. Early on, however, regions of the board are seldom completely independent. Then what is considered tenuki is fuzzy. If a play is tactically related to the last play, it is not tenuki, no matter how far away it is on the board. If it is not tactically related, it is tenuki, no matter how close it is.

This is a great definition for a human, but "tactically unrelated" is not easy to translate into an algorithm.

Well, it is possible to program fuzziness. (Not that I have ever done it, but I have a couple of books. ) The problem there is that you need to get somebody's judgement about what is tenuki. (Your own is fine for your purposes, I am sure. But involving humans takes time.)

Perhaps I should have said a bit more originally on what I'm trying to do-- I potentially need to answer this question several hundred times per move per pro game, in all 65,000+ pro games of GoGoD. I only build my database occasionally, so it doesn't have to be super fast, but it does need to complete in my lifetime. So, basically I'm hoping for a definition that could be applied with reasonably accurate results by a 35kyu...

I think that your own judgement, coupled with a joseki book, would be good enough for that.

Alright, well, I'm not sure what I'll end up doing, but a related question: I have considered giving an "urgency" rating to each position, which would be basically "when this position appears in pro games, how many moves on average pass before the players play locally again? So 0.0 would indicate that pros always play locally immediately, and larger numbers indicate less urgent positions (pros play N moves elsewhere before coming back to this position). Does this seem like an worthwhile metric?

Thanks for the input

That sounds a lot like the PON. Which is a better metric, perhaps, than how long before play returns to the locale, since the latter depends upon the whole board.

Question: can't you just call a move that shows up fewer than X times in the database a tenuki?

This has the disadvantage that occasionally when a move on the other side of the board shows up in a well-known fuseki, DJ will test us on the whole-board sequence rather than the corner sequence; and that some more obscure variations of the more obscure joseki will be lost; but on the whole, that's not a huge problem, right?

BTW, the concept of joseki is not set in stone. Apparently joseki that cover more than a single corner are being developed now. But that is not a new thing. Here is an ancient joseki. Is there a tenuki?



Edit: I have not really studied the book, but I do not think that it uses the term, tenuki. Often it seems like plays like are intended as tenuki, but. . . .

daniel_the_smith wrote:
Just to explain this real quick: the marked stones suddenly appear because in all 8 games where black played (the bad shape) 9, those stones were already on the board.

So i think this is a good illustartion of my proposal:
if several games have the same "tenuki" sequence then its proably not a tenuki .. because the common thing thsoe game have in common is the joseki at hand. (thisis true where you are a bit later in the game but this is where the issue arise anyway : early on divinfg the board in 4 quandrats is enough)
here the marked sequence is played in the 8 games between 7 and 9 and is part of joseki: you have more info if you aggreagate data from sevral games

Not sure it works often though

Here is a reference to the PON paper:

Estimating the Possible Omission Number for Groups in Go by the Number of n-th Dame
by: Morihiko Tajima, Noriaki Sanechika

COMPUTERS AND GAMES
Lecture Notes in Computer Science, 1999, Volume 1558/1999, 265-281, DOI: 10.1007/3-540-48957-6_18

Springer charges for a download, but maybe you can contact one of the authors. You might also check on the computer go mailing list.

John Fairbairn wrote:

Tenuki means to play somewhere else.

Not really. The Japanese is a noun not a verb. The verb usage is te wo nuku or tenuki suru.

I think that English grammar allows the use of tenuki as a verb or a noun. As in, "if it's only worth 15 points, tenuki".

The point is (under the Japanese definition), it occurs when you do not respond to a threatening move just played by the opponent. Playing ine one area and ending in sente then playing somewhere else is not a tenuki. There is a sense of bravado in a tenuki, as in the proverb "if it's only worth 15 points, tenuki" (or variations on the number).

An important point. If Black plays on the 3-4 in an open corner and White plays in another corner, that is not tenuki.

Click Here To Show Diagram Code

[go]$$Wcm10
$$ --------------------
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . O . X 3 . .
$$ | . . X X O . 1 . . ,
$$ | . . . O X O . . . .
$$ | . . . . X . . . . .
$$ | . . 2 . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . , . . . . . ,[/go]

Likewise, for Black to play somewhere else after is not tenuki.

However, the threat can be very weak.

Click Here To Show Diagram Code

[go]$$B
$$ --------------------
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . 5 3 4 . . . .
$$ | . . 1 , 2 . . . . ,
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . , . . . . . ,[/go]

For to tenuki would be bad, I think, but for to tenuki is not uncommon. The cut is not such a strong threat. Bravado overstates the case, I think.

Click Here To Show Diagram Code

[go]$$B
$$ --------------------
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . 2 . . . . .
$$ | . . 1 , . . . 3 . ,
$$ | . . . . . . . . . .
$$ | . . . . 4 . . . . .
$$ | . . 5 . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . , . . . . . ,[/go]

Likewise, in this case tenuki is normal. Black's threatened attack is not severe.

I think that in general amateurs do not tenuki enough. And I think that the reason has less to do with caution or bravado than with tunnel vision.

Bill, I'm not convinced that your last two examples would normally be described as tenuki in Japanese, though of course they could be. The point is, here, you are not so much ignoring the opponent's last move (as you say the threat is weak) but rather deferring your response while you make a probe elsewhere. Although it doesn't always happen, I think you have a strong intent of returning soon in cases like the above, and so are not really ignoring the opponent. I do think the key nuance tends to be "ignore". Japanese has other ways of saying "play elsewhere" when "ignoring" does not apply (e.g. hoka ni tenjiru), and these appear often in go texts.

For Daniel's purpose, the message I take from that is that "tenuki = ignoring a threat" is not really a common joseki concept. When an otherwise common move is not played, it is usually deferral. If so, Daniel's idea of counting how long a move is deferred seems to be absolutely on the right track. And, while being careful not to overtax my brain, I'm not sure deferral in joseki has much to do with locales or the price of fish either. Which justifies his idea even more.

daniel_the_smith wrote:your book, is it available in the US?)

It needs to be ordered from Europe. E.g., from me postage is EUR 3.45.

but I want to actually measure things

I think my symbolic levels work as a measurement, if your application is simply to determine whether to tenuki at all or which of two very different plays elsewhere is more urgent. Similar levels are not distinguished because I think that other considerations (of other concepts, terms) would have to be taken into account then.

(I would be very surprised if you have a procedure such that a 30 kyu could come up with the same "unrest level" that you would for a given position).

I think if we say 18 kyu, then the the chance is reasonable that he and I would find the same level, once the reader has made himself familiar with the concepts stability, urgency, investment.

I'm not sure what all is involved in your "unrest level"

One must be able to notice that a group is unsettled, that a local sente play is available, which major development directions there are from a group, whether a group is important or can be sacrificed etc. Since you want to do it by a program (I think), it must be able to cut its tactical reading by noticing that something is (rather) quiet or stable.

Then you get a few conditions like "add 2 to the level for each weak, unsettled, important group".

how well does it model pro play?

I have not done statistics because (besides lack of time) many other decisions can play a role. (E.g., when basically one would tenuki, one could still consider playing preliminary sente forcing moves.)

Bill Spight wrote:There is a paper about the possible omission number of a group (i. e., how many times you can tenuki), with an algorithm to determine the number, based upon the judgement of a Japanese amateur 6 dan.

Now this is very related to my book's terms n-connected and p-alive, which define exactly that, see
http://senseis.xmp.net/?NConnection

OC, you have to be able to say what a group is before you can determine its PON.

No. One simply chooses some stones / strings as a group! (Only if you want to do that particularly meaningfully would you want to acquire some greater ability for the selection.:)
)

I have a couple of books.

What do they say about tactical relation and the like?

RobertJasiek wrote:

Bill Spight wrote: I have a couple of books.

What do they say about tactical relation and the like?

They are about fuzzy logic and programming.