White to play.
White's textbook play loses 15% to par by Elf's reckoning. Assuming komi, OC.
Enjoy!
| Life In 19x19 http://www.lifein19x19.com/ |
|
| Opening problems for AI: Problem 32 http://www.lifein19x19.com/viewtopic.php?f=15&t=17651 |
Page 1 of 2 |
| Author: | Bill Spight [ Thu Jul 16, 2020 5:52 am ] |
| Post subject: | Opening problems for AI: Problem 32 |
White to play. White's textbook play loses 15% to par by Elf's reckoning. Assuming komi, OC. Enjoy!
|
|
| Author: | pwaldron [ Thu Jul 16, 2020 12:36 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
| Author: | RobertJasiek [ Thu Jul 16, 2020 11:27 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Continuing the upper left has relatively small value. Approaching the lower left corner should be slightly less valuable than occupying the empty corner but this is very hard to explain. The most interesting question is whether to press in the upper right in expected sente before taking the empty corner. Since Black can push once more to take away sente and so the empty corner, White should first take the empty corner himself. Afterwards, White has similarly attractive follow-ups in the lower left (approach) and upper right corners (press). Which move in the empty corner? Since White can press in the upper right, Black avoids a move in the lower right aiming at building a big moyo on the right side, i.e., avoids Q5, R5, P4 or O4. Hence, Q4, R3, Q3 or P3 (cooperates with E4) are possible but current AI would not play P3. |
|
| Author: | Knotwilg [ Fri Jul 17, 2020 4:04 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
We have 3 corners with respectively 0, 1 and 2 stones Empty corner is the obvious choice. However, AI exceptionally like pincers when they are also an extension of influence. White may forestall it with a 2 space extension. Pressing looks like black may cut instead of walking along. The textbook play may have been the attachment or the kosumitsuke I still go for the empty corner |
|
| Author: | John Fairbairn [ Fri Jul 17, 2020 5:24 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Bill: In the same way that we look for Direct 3-3 and shoulder hits etc if we are told an AI bot played next, I think we can safely say that we now look for a Bill Spight press when we know who set the problem But two points for you to consider before you reveal your answer: 1. Elf encountered the upper-left joseki in a game against a human (Ch'oe Ch'eol-han) and played elsewhere. Golaxy likewise. In other words, in games with 7.5 komi. 2. This sort of question came up in "9-dan Showdown" (Go Seigen vs Fujisawa K) in two games, and of course the surrounding situation is discussed at length there. Fujisawa in both cases chose not to continue the joseki but went for a big-point strategy instead. I seem to recall that he went for this partly on the basis of tedomari considerations, which may well apply here, too. In fact, of all the games that show that upper-left joseki in human games, the human chose NOT to play the prime text-book more than half the time. This is perhaps a case of the joseki not being joseki 
|
|
| Author: | Bill Spight [ Fri Jul 17, 2020 8:23 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
John Fairbairn wrote: Bill: In the same way that we look for Direct 3-3 and shoulder hits etc if we are told an AI bot played next, I think we can safely say that we now look for a Bill Spight press when we know who set the problem That may be the result of looking for sufficiently large mistakes. You may have a move that in itself, or in general, is not so bad, but when there is a press on the board, it is worse by comparison, and therefore I notice it. John Fairbairn wrote: But two points for you to consider before you reveal your answer: 1. Elf encountered the upper-left joseki in a game against a human (Ch'oe Ch'eol-han) and played elsewhere. Golaxy likewise. In other words, in games with 7.5 komi. Further discussion hidden for courtesy. |
|
| Author: | Gomoto [ Fri Jul 17, 2020 1:20 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Fat Kata: |
|
| Author: | Uberdude [ Sat Jul 18, 2020 1:59 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
RobertJasiek wrote: The most interesting question is whether to press in the upper right in expected sente before taking the empty corner. Since Black can push once more to take away sente and so the empty corner, White should first take the empty corner himself. When you want to get sente from the press, there's another idea I've learnt from bots (e.g. in Kobayashi-denied fuseki) which is quite often better than the extra push, shown below for comparison. Pushing means you help white build strength in the centre, and the turn of 7 is juicy. The other idea is the somewhat crude and cramped looking kick / kosumi-tsuki in the corner. If white fighting-spiritly says "Hey, you ignored my press so I will cover!" with 3 that's often not the best choice: if white continues locally building the thickness with 7 then black takes sente for 8 in empty corner or wherever you wanted, and the key difference to above, to my understanding, is the white group here is less thick/strong as the first diagram. By not pushing from behind black avoids giving white the strength and potential eyespace adjacent to the wall; in both variations white has a wall of 2 stones without an extension facing down the right side. Also in this variation black can q16 empty triangle and cut (sometimes should try to get q16 in sente before tenuki, but white may resist). Of course white doesn't need to take gote at the top, but can take the empty corner, and then black can take the juicy hane at 8 (or lower left). If black doesn't take 8 but lower left at a then it's not sente/painful to allow white to spend a gote there: by playing the corner kick black lowers the temperature of this corner, allowing for grabbing the other big opening points. Compare to if white forcefully takes the empty corner in the push an extra time line, now black is pretty much obliged to take 6 here as reverse sente... ... because if black takes the lower left, white extend 7 is sente now because there's no moves big enough remaining to be worthwhile taking and allowing the white turn. Rather than blocking, white can hane at 3, which is looking for black to obediently block inside and then white can get sente, having made p18 a less good follow up for black. Black counters with 4, and then 7 is the toughest move for white to keep maximum pressure on the corner (as now black r18 block doesn't threaten to capture a stone). After 9 black has 2 main choices: - slide to a, white defends at b, and then black has sente for empty corner at c, or - crawl and then jump with d-e-f, plus is black isn't so low, con is white got stronger in centre so can tenuki to empty corner, and black acvtivating the cutting stones or white defending there is a big move that'll come fairly soon, but probably after shimaris/approaches on lower side. |
|
| Author: | Bill Spight [ Sat Jul 18, 2020 5:29 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
| Author: | John Fairbairn [ Sat Jul 18, 2020 7:35 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Quote: The main takeaway, I think, is that the central hane for  is not worth it, in this and probably most cases, and may possibly be considered joseki no longer. Any reason for not mentioning the main joseki, E14? I mean main in the sense of most played, and also (?therefore) the only line given as joseki by Kitani. The hane is mentioned by Kitani but evaluated as "an even result", which is usually for lines not yet played enough to be considered joseki. I just happen to be familiar with this pattern because of "9-dan Showdown". |
|
| Author: | Bill Spight [ Sat Jul 18, 2020 10:20 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
John Fairbairn wrote: Quote: The main takeaway, I think, is that the central hane for is not worth it, in this and probably most cases, and may possibly be considered joseki no longer. Any reason for not mentioning the main joseki, E14? I mean main in the sense of most played, and also (?therefore) the only line given as joseki by Kitani. The hane is mentioned by Kitani but evaluated as "an even result", which is usually for lines not yet played enough to be considered joseki. I just happen to be familiar with this pattern because of "9-dan Showdown". The main reason for not mentioning E-14 is that I choose these problem positions by going over more than 20 games each, on average. I don't decide on the point of the problem ahead of time. If E-14 had shown up and was a large enough error, I would have mentioned it. Or if Elf had recommended it, or even given it more than 1500 rollouts, I would have mentioned it, then, too. When I was around 2 kyu, I recall reading something about this joseki, maybe in a go magazine. Anyway, the author did not recommend the hane, I don't remember why, but he did say that the player with the corner (White in this case) did generally need protection from the atari at E-15, and that the hanging connection with E-14 offered that. His recommendation for Black was not to play that atari, because it was aji keshi, but to play the keima at E-15, since to prevent the atari White would have to make bad shape. Anyway, in the Waltheri database tenuki is the most popular human option at around 36%, with the hane next at around 33%, and the hanging connection third at around 20%. |
|
| Author: | Bill Spight [ Sat Jul 18, 2020 11:05 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
About the hanging connection Here is the first case I found where the hanging connection was played instead of tenuki. GoGoD 1786-04-10, Honinbo Retsugen (W) vs. Hattori Inshuku, my friend A few comments, based upon Elf's calculations.  Within the margin of error. Minor error. Minor error. Once is probably OK, twice is too much. Within the margin of error. Occupy the last empty corner, is a good rule of thumb, though. Ditto. Within the margin of error. a may be better, however. Minor error. Occupy the last open corner. Blunder. Loses 21½% to par. Occupy the last open corner.
|
|
| Author: | John Fairbairn [ Sat Jul 18, 2020 11:38 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Quote: Anyway, in the Waltheri database tenuki is the most popular human option at around 36%, with the hane next at around 33%, and the hanging connection third at around 20%. GoGoD has 210 games, with the hane and the hanging connection at 60 each. Both are still being played, but the hane is ever so slightly the modern preference. Possibly it's seen as more severe in the modern 7.7 komi era? |
|
| Author: | Bill Spight [ Sat Jul 18, 2020 11:45 am ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
John Fairbairn wrote: GoGoD has 210 games, with the hane and the hanging connection at 60 each. Both are still being played, but the hane is ever so slightly the modern preference. Possibly it's seen as more severe in the modern 7.7 komi era? Emphasis mine. Still being played? Early in the game? If so, my guess is that that won't last. Early in the game tenuki looks to be too good. |
|
| Author: | Bill Spight [ Sat Jul 18, 2020 12:01 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Here is another example in early play. The next one I found. GoGoD 1823-12-18b, Hattori Inshuku (W) vs. Inoue Ansetsu. Castle game. Within the margin of error. Minor error. Borderline error. Probably better to play in an open corner or invade the bottom right corner. Minor error. Invade the bottom right corner. Within the margin of error. Good. Minor error. Enclose the bottom right corner. Blunder. Lose 19½% to par. Invade the bottom right corner.
|
|
| Author: | Gomoto [ Sat Jul 18, 2020 12:10 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
What about  
|
|
| Author: | Bill Spight [ Sat Jul 18, 2020 12:19 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Gomoto wrote: What about Within 1% of the 2d line extension. I figure the difference is very likely to be noise. |
|
| Author: | Bill Spight [ Sat Jul 18, 2020 12:25 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
This time I went looking for a modern example later in the game, but still no tenuki. Here it is. GoGoD 1939-04-19, Inoue Ichiro, 4 dan (W) vs. Takagawa Kaku, 4 dan. Oteai.  Mistake. Loses 16% to par. Enclose the top left corner at a.
|
|
| Author: | John Fairbairn [ Sat Jul 18, 2020 3:36 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
Quote: Still being played? Early in the game? How early is early? Earliest decision point with this joseki seems to be move 15. Four contrasting examples in roughly last decade: twoopting for tenuki and two for hanging connection. Hane: in recent times the earliest this was played was move 18 (2005). Cases since then have been rather late in the game (beyond move 60). If you go back another ten years or so, early decision point combined with early hane or hanging connection seems to be the norm (moves 15 to 23). I could imagine pro views on each choice haven't really changed, but we are seeing far fewer examples now simply because takamoku has fallen right out of fashion in the AI age: a sharp decline culminating in just 7 examples in 2019 and none so far in 2020. |
|
| Author: | Bill Spight [ Sat Jul 18, 2020 5:32 pm ] |
| Post subject: | Re: Opening problems for AI: Problem 32 |
John Fairbairn wrote: Quote: Still being played? Early in the game? How early is early? Early enough so that a normal option, such as playing in an empty corner, approaching a 3-4, 5-4, or 5-3, or making a shimari with one of them, or playing a press against a 3-4 is available. There may be positions where the hane or the hanging connection is preferable to one of those, but I would like to see such a position. John Fairbairn wrote: I could imagine pro views on each choice haven't really changed, but we are seeing far fewer examples now simply because takamoku has fallen right out of fashion in the AI age: a sharp decline culminating in just 7 examples in 2019 and none so far in 2020. Good point. I think that there are positions where the 5-4 is playable, but not many where it is preferable to a 4-4 or 3-4.
|
|
| Page 1 of 2 | All times are UTC - 8 hours [ DST ] |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|
, leads to a big fight in the top left quadrant. Below is Elf's mainline variation for 
.
returns to the push in the center. 
extends. a is also playable. Then 
extends on the left side, which is now hot. 
makes a base in the top right. 
encloses the bottom right corner. 
prevents Black b or thereabouts. Then 
and 
secure the top right corner.
plays the press in the bottom right corner. 
replies with the kick and slide.