Life In 19x19 http://lifein19x19.com/ |
|
How is your positional judgement? http://lifein19x19.com/viewtopic.php?f=15&t=16909 |
Page 2 of 7 |
Author: | Knotwilg [ Wed Sep 11, 2019 9:36 am ] |
Post subject: | Re: How is your positional judgement? |
This is easier than the ones before. Black needs to save 2 cutting stones with an empty triangle. Then White pushes up, cutting off 3 stones from Black's area of development. This game is White's. |
Author: | Joaz Banbeck [ Wed Sep 11, 2019 7:26 pm ] |
Post subject: | Re: How is your positional judgement? |
Black plays K2 and wins. My general assessment is that white has two weak groups, black has one.
I don't claim that the following diagrams are conclusive, nor even accurate. I post them to illustrate the general idea that when white saves one group, the other is in trouble. I'll leave it to LZ to figure out the exact tactics.
|
Author: | Dusk Eagle [ Wed Sep 11, 2019 8:55 pm ] |
Post subject: | Re: How is your positional judgement? |
I prefer white. These black stones will be a liability.
I prefer white. Black has to play , then white has more solid territory as well as sente for the fight on the bottom side. But I would never assign a ~90% win rate to one side in these positions like Elf is so keen to. |
Author: | Knotwilg [ Thu Sep 12, 2019 1:59 am ] |
Post subject: | Re: How is your positional judgement? |
Reacting to Joaz: Quote: Black plays K2 and wins. My general assessment is that white has two weak groups, black has one. I see it differently: one of these two White groups are two low stones, providing useful aji but expendable. The other one, which is the one to save will put pressure on Black's 3 stones, which are not expendable, while impacting Black's only area for growth.
B4 probably captures the stones, but the capture is not big and the stones provide useful aji at A or B. I don't really know how to use WHite's sente either, but I don't think Black can match White' three corners and komi. Probably WHite just has to make shape in the centre to control the game. |
Author: | dfan [ Thu Sep 12, 2019 4:48 am ] |
Post subject: | Re: How is your positional judgement? |
Dusk Eagle wrote: But I would never assign a ~90% win rate to one side in these positions like Elf is so keen to. To put "90%" in terms some of us are more comfortable with, KataGo estimates the score of the first position Bill posed (the only one I fed it) at around +5 for the side that's ahead, and puts the win rate in the 60s (as the other bots do). |
Author: | Knotwilg [ Thu Sep 12, 2019 7:19 am ] |
Post subject: | Re: How is your positional judgement? |
edited to include the answer to Joaz' fine inclusion of a probe (don't know if Bill still likes it ) |
Author: | mhlepore [ Thu Sep 12, 2019 7:20 am ] |
Post subject: | Re: How is your positional judgement? |
dfan wrote: Dusk Eagle wrote: But I would never assign a ~90% win rate to one side in these positions like Elf is so keen to. To put "90%" in terms some of us are more comfortable with, KataGo estimates the score of the first position Bill posed (the only one I fed it) at around +5 for the side that's ahead, and puts the win rate in the 60s (as the other bots do). Is it possible that there is a hardware issue at play? (For example, if the elf evaluation is on a machine without a GPU, but the others are using a GPU) Otherwise it is headscratching to see strong bots vary that much. |
Author: | Kirby [ Thu Sep 12, 2019 7:46 am ] |
Post subject: | Re: How is your positional judgement? |
Elf usually has pretty high variation in winrates compared to other bots, so it’s not surprising to me that LZ is around 60 and Elf is around 90. I would be more surprised if Elf were around 70 and LZ were at 40, because that’d be a different interpretation of who had the lead. |
Author: | dfan [ Thu Sep 12, 2019 8:08 am ] |
Post subject: | Re: How is your positional judgement? |
mhlepore wrote: dfan wrote: To put "90%" in terms some of us are more comfortable with, KataGo estimates the score of the first position Bill posed (the only one I fed it) at around +5 for the side that's ahead, and puts the win rate in the 60s (as the other bots do). Is it possible that there is a hardware issue at play? (For example, if the elf evaluation is on a machine without a GPU, but the others are using a GPU) The KataGo evaluation was with approximately the same number of playouts as the ELF evaluation. Quote: Otherwise it is headscratching to see strong bots vary that much. ELF is very, very opinionated. It is not unusual for it to give a position a 90% win rate while other strong bots give it a win rate more like 65%. |
Author: | Bill Spight [ Thu Sep 12, 2019 8:41 am ] |
Post subject: | Re: How is your positional judgement? |
dfan wrote: mhlepore wrote: Otherwise it is headscratching to see strong bots vary that much. ELF is very, very opinionated. It is not unusual for it to give a position a 90% win rate while other strong bots give it a win rate more like 65%. Since, as players approach perfection, their winrate estimates will approach 100% or 0%, we might say that Elf is very, very self-confident. |
Author: | Kirby [ Thu Sep 12, 2019 8:46 am ] |
Post subject: | Re: How is your positional judgement? |
Facebook knows everything, after all.. |
Author: | Joaz Banbeck [ Sun Sep 15, 2019 3:13 pm ] |
Post subject: | Re: How is your positional judgement? |
Bump. Bill, are we going to get an 'official' answer to #4? |
Author: | Bill Spight [ Sun Sep 15, 2019 5:09 pm ] |
Post subject: | Re: How is your positional judgement? |
Game 4. Oteai game between Fujisawa Kuranosuke, 9 dan, and Takagawa Kaku, 7 dan, 1950-04-12a. Fujisawa was the first non-Meijin 9 dan; later he changed his name to Hosai. Fujisawa played White, with no komi. After Elf gives White an estimated winrate of 88% (144.2k playouts), assuming, as always, 7.5 komi and area scoring. With no komi, Black, Takagawa, won by 2 pts. Takagawa was only 2 years from his winning streak of 8 Honinbo titles, and was known for his skill at the opening, so I expected that he would be ahead, even with no komi, but Fujisawa cleaned Takagawa's clock. He was a 9 dan, after all.
Fujisawa played at 40. Through Elf played as Dieter did. If is at 36, is at 38. Then at a, at 35. In the main variation from there the result is a ko that White takes first. White ignores Black's threat at b, allowing Black to descend to c. Elf rates at 36 as 89% for White (29.6k playouts). The 1% difference is within Elf's margin of error, I believe.
---- OC, since this is a no komi game, we cannot assume that Takagawa played badly. And even by Elf's reckoning he made only two sizable errors up to .
was joseki, but Elf does not like it. Elf also considers the extensions to be minor mistakes.
was also a joseki play at the time, but I believe that - are new to the bot era.
is a thick play. The result is the sequence up to . Elf considers to lose 14%. Whether that is so with no komi is another question.
This way Black gets territory and White gets thickness. Surely playable by Black with no komi. |
Author: | Bill Spight [ Sun Sep 15, 2019 5:34 pm ] |
Post subject: | Re: How is your positional judgement? |
Game 5. After Elf estimates a winrate of 90% (26.8k playouts, area scoring, 7.5 komi) for one player. Which player is it?
Enjoy! |
Author: | Bill Spight [ Tue Sep 17, 2019 7:19 am ] |
Post subject: | Re: How is your positional judgement? |
Game 5. Discussion hidden out of courtesy. Miyashita Shuyo, 7 dan played White agains Takagawa Kaku, 7 dan, in a preliminary round of the Honinbo tournament on August 23, 1950 (GoGoD 1950-08-23a). Komi was 4½ pts. After Elf, assuming area scoring and 7½ pt. komi, estimates Black's winrate at 90% (28.4k playouts). Takagawa won by resignation after Black 275. Takagawa won the Honinbo title in 1952 and held it for nine straight years. He was known for his mastery of the opening and for his positional judgement.
Like Dieter and Leela Zero, Elf likes the crosscut here. But Elf's path immediately diverges.
Elf shows this variation with 1% worse winrate estimate than its play, and with fewer playouts (3.1k vs. 14.3k). Elf ends this variation here. OC, the difference could be noise.
After the crosscut, White plays to sacrifice the stones.
White pushes into the center, enclosing the stones for a furikawari. After Black's lead looks apparent. The actual game continuation is of some interest, I believe.
Takagawa did not fancy the sacrifice and played . Elf regards this as losing 2%. Then he cut at 37. Elf regards this as losing 3%. Both plays are within Elf's margin of error, but they do add up. But then Miyashita returns the favor with , losing 5½%.
Elf recommends the atari at , followed by the hanging connection. The result is a furikawari similar to Elf's mainline, but with White 5% better off.
This variation emphasises the flexibility of today's top bots. After , marked, Elf switches gears and, instead of saving the stones, sacrifices them with sente. Then Black invades the top side. (In other variations Elf plays at a.)
is a probe. is the blunder. Black tries to save everything in a complicated fight in which he is outnumbered and has too many weaknesses. Elf says that loses 25%. Then loses 10% more. Elf thinks it should be at 48, taking away a White dame. A few moves after this diagram Black jumps out at a and then White gets to play the hane at b. Things went from bad to worse for Takagawa, and by it is White who has a winrate estimate of 90%. Backing up, Takagawa could have avoided the complications.
Elf recommends the underneath hane for instead of the cut. Then Black does not have a second weak group to worry about. In fact, can attack White's weakness. Now a look at the earlier play.
The splitting play on the side is one of the hoariest stratagems in go, going back to the earliest extant game records. In this case Miyashita probably played it to avoid sanrensei by Black. Before the AI revolution both the sanrensei and the splittinng play had begun to be questioned. In his 21st century go writings, Go Seigen downplayed the splitting play, but was plainly ambivalent. There is even one position on which he took opposite opinions on it on two different occasions. Nowadays, the bots generally disapprove of the splitting play. BTW, is on R-09 instead of R-10 to give less space for Black to approach while extending his bottom side moyo,
= OC, at the time this 3-3 invasion was almost anathema. The hane-and-connect on the second line was considered joseki until the AI era. But even now it is hard for me to shake the feeling that Black does not gain by this sequence (taking into account White's extra play, OC). Surely, it seems, Black's fourth line stones in the wall are worth more than 2 pts. of territory each, especially with the Black position on the bottom side.
Elf's second choice is the top left enclosure. Note that Black does not play sanrensei, but makes the "enclosure" of the bottom right corner, extending the Black moyo only a little. Then White invades the top right corner. Note that Black blocks on the top side. the opposite of what we might expect from human ideas about the direction of play. {shrug}
ELf regards Takagawa's (next diagram) as losing 9%. Instead Black approaches the top left corner. The kick joseki there is a favorite of Elf's. Note that Black approaches from the bottom, anyway. As for it has only 1.5k playouts, but I suspect that it would be Elf's choice in that corner, anyway, because it prevents Black from playing the block there in response to the 3-3 invasion.
You can see how would have been considered ideal at the time. It was a long extension from the top right 4-4 stone and it invited , which Black kicked with , resulting in a position for White was and still is, considered a bit overcentrated. Then is an ideal development of the Black bottom side. Nonetheless, Elf prefers the previous diagram for Black by 10%. When I was learning go the textbooks decried without an enclosure of the top left corner. Elf does, too. regarding it as losing 11½%. However, it and other such extensiions without an enclosure have a long history in top level play. Perhaps Miyashita wanted to prevent a kind of double wing formation by Black from the top right 4-4 stone if he enclosed the top left.
Elf, of course, would enclose the top left corner. Then Black would not extend far on the top side, but simply enclose the top right corner. Then White would play the shoulder blow to reduce the Black moyo on the bottom side. ---- Below in progress.
Elf thinks that loses 18½%. So what is the point of ? Elf offers the following variation.
White builds a wall and then invades the right side. What does Elf recommend for instead?
= Black takes kikashi against White on the left side and secures his group in the top left with sente. White ignores and invades the right side. In the actual game, up to White made three sizable errors: the splitting play on the right side, the extension on the top side, and the wrong plan on the left side. |
Author: | bernds [ Tue Sep 17, 2019 7:24 am ] |
Post subject: | Re: How is your positional judgement? |
Bill Spight wrote: Is this one too complicated? Well, none of these are actually easy - even 90% from ELF is probably just a few points difference. But I'll go ahead and say Black is better - not sure what White has achieved on the left side, and I have a feeling the AI wouldn't like L17. With Black to move it seems like B should be able to achieve a good result at the bottom. |
Author: | Bill Spight [ Wed Sep 18, 2019 3:54 am ] |
Post subject: | Re: How is your positional judgement? |
In keeping with doing two per week, here is the next game. I think it may be better suited to human evaluation than some of the others, as there is no obvious complex fight going on. Game 6. As always, assume area scoring and 7½ pt. komi. After Elf estimates the winrate of one player as 87½% (36.6k playouts). Which player is it?
Enjoy! |
Author: | Knotwilg [ Wed Sep 18, 2019 6:50 am ] |
Post subject: | Re: How is your positional judgement? |
Game 5 no LZ black crosscuts and leads. ANalysis without LZ Game 6 no LZ Analysis without LZ. white caps and leads (shoulder hit too heavy) edit - with LZ LZ agrees with my taking sides and my move proposals. Her variations of course quickly diverge: Game 5 Game 6 |
Author: | Bill Spight [ Thu Sep 19, 2019 10:15 am ] |
Post subject: | Re: How is your positional judgement? |
Game 6. Discussion hidden out of courtesy.
GoGoD 1951-02-01a, 4½ pt. komi Assuming area scoring and 7½ pt. komi, after Elf reckons White to be ahead (87½% winrate, 36.6k playouts). However, Black, Kitani, won by resignation.
In the actual game Maeda played at a. Elf prefers the 3d line side attachment by 1% (attachment for sabaki, 66.9k playouts). Black plays the hane, , and White :42: replies with the 2d line counter-hane. plays the two step hane and connects.
is a kind of driving tesuji. switches to the top right corner, preventing Black from playing the jump attachment at a.
Black makes sabaki in the top right, starting with the top attachment, . Black's early errors
68 years ago, Kitani played , the "AI attachment". But then, after the hane, , he played the crosscut, losing 18½% by comparison with the counter-hane at a.
= After White plays and sacrifices two stones.
loses 11½%.
stands. Then Black makes four stones and sacrifices them. Then starts to make sabaki for Black in the top right corner.
switches to the 4th line side attachment against the top left enclosure.
After the exchange in the top left, runs out. |
Author: | Knotwilg [ Thu Sep 19, 2019 12:04 pm ] |
Post subject: | Re: How is your positional judgement? |
Bill, it's been a long time since I enjoyed Go so much. Making my own positional judgment, then check with Lizzie and then see what Elf does ... the times we live in! |
Page 2 of 7 | All times are UTC - 8 hours [ DST ] |
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |