It is currently Fri Mar 29, 2024 4:41 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 12 posts ] 
Author Message
Offline
 Post subject: Opening problems for AI: Problems 4 and 5
Post #1 Posted: Sun Jun 14, 2020 1:27 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
With apologies to Dave (ez4u), but as you shall see, I couldn't resist.

This pair of problems comes from a game between Takagawa and Fujisawa Hideyuki (GoGoD 1965-03-11a, 5.5 komi. Takagawa was White). I don't have to worry about whether the human play is textbook or not. These two guys, known as masters of the fuseki in their time, wrote the textbooks. ;)

Problem 4.

Click Here To Show Diagram Code
[go]$$Bc White to play.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Lots of choices, but I think that the bots will probably agree on the same point or next to it. And that they won't like Takagawa's play, by comparison.

Problem 5 is hidden, for when you are ready. :)

Click Here To Show Diagram Code
[go]$$Bc Black to play.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Again, I expect the bots to agree on the same point or next to it, and not to like Fujisawa's play.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #2 Posted: Sun Jun 14, 2020 1:38 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Discussion of Problem 4.

Click Here To Show Diagram Code
[go]$$Wcm10 One space extension. Joseki.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . 1 . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

This is joseki, OC. But what might haven given us pause before the AI era is the fact that the Black hanging connection does not have a strong follow-up against the solid White corner, so the extension is not necessary and may be inefficient. The bots have another idea.

Click Here To Show Diagram Code
[go]$$Wcm10 Enclosure. Elf's main line.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . 2 . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 4 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 0 . . . . . . . . . . . 9 7 . . |
$$ | . . O . . . . . . . . . . . . 3 6 . . |
$$ | . . . , . . . . . , . . 5 . . , 8 . . |
$$ | . . . 1 . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Elf's top choice is to enclose the bottom left corner. OK, that's plainly a good play. What is shocking, however, is that Elf regards the extension in the top left as 21% worse, i.e., a very bad play. :shock: As the mainline variation shows, the top left is non-urgent.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #3 Posted: Sun Jun 14, 2020 1:53 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Discussion of Problem 5.

Click Here To Show Diagram Code
[go]$$Bcm11 Three space extension. Joseki.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . . . . . . . . 4 . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . 6 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . 1 . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . 3 . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . 2 . . . . . 5 . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

The game continued as above. Note that :b13: aims to make two enclosures or to get the mutually desired extension at :b15:.

Click Here To Show Diagram Code
[go]$$Bcm11 3-3 invasion. Elf's mainline.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . . . . . . . . 0 . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 2 . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . 8 6 . . |
$$ | . . O . . . . . . . . . . . . 4 5 . . |
$$ | . . . , 3 . . . . , . . . . . , 7 . . |
$$ | . . 1 . . . . . . . . . . 9 . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Elf regards the bottom left corner as urgent and plays the 3-3 invasion. What follows is its mainline variation. Again, what is shocking is that it regards the 3-3 invasion as 20% better than :b11: in the game. :shock:

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.


Last edited by Bill Spight on Mon Jun 15, 2020 4:52 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #4 Posted: Sun Jun 14, 2020 3:24 pm 
Lives in sente

Posts: 757
Liked others: 114
Was liked: 916
Rank: maybe 2d
KataGo recent stronger unreleased 40 blocks, lots of playouts, analysisWideRootNoise = 0.1 to force evaluation of lots of options. Winrates from black's perspective.

Problem 4
Difference with F15 is about 9% winrate and around 1 point. Difference increases by 1% to about 10% if you play it out and let it get thousands of playouts instead of only 830.
Attachment:
opening4.png
opening4.png [ 1.26 MiB | Viewed 7046 times ]


Problem 5
Difference with C9 is about 12% winrate and around 1.5 points. Difference decreases by about 1% to 11% if you play it out and let it get thousands of playouts instead of only 345.
Attachment:
opening5.png
opening5.png [ 1.25 MiB | Viewed 7046 times ]


This post by lightvector was liked by 2 people: Bill Spight, Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #5 Posted: Sun Jun 14, 2020 3:30 pm 
Lives in sente

Posts: 757
Liked others: 114
Was liked: 916
Rank: maybe 2d
Also for problem 4...

We sat from the heatmap above, that KataGo actually thinks that the peep is another plausible move, and is basically the same winrate as sealing the corner in the bottom left. What is the followup to the peep?

Click Here To Show Diagram Code
[go]$$Wcm10
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . a . . . . . . . . . . . |
$$ | . . O , . 3 . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X 2 1 . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . d , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . b c . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


So interestingly, KataGo considers it fine to leave the lower left alone and respond on the top, but thinks white must play this way rather than playing the human joseki 1-point-jump. I'm not sure whether it usually prefers this particular pattern as the local followup for white given the upper left shape or if it's specific to this board. I think the idea at least in this case is that now black's left stones are a bit heavier and attackable, making it much harder for black to have time to take the lower left corner.

Locally, KataGo expects black 'a' next, but if black does rush to the lower left at 'b' or 'c' next, then white 'd'. :study:


This post by lightvector was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #6 Posted: Sun Jun 14, 2020 3:36 pm 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Corners before sides!


This post by Uberdude was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #7 Posted: Sun Jun 14, 2020 3:58 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
lightvector wrote:
Also for problem 4...

We sat from the heatmap above, that KataGo actually thinks that the peep is another plausible move, and is basically the same winrate as sealing the corner in the bottom left. What is the followup to the peep?

Click Here To Show Diagram Code
[go]$$Wcm10
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . a . . . . . . . . . . . |
$$ | . . O , . 3 . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X 2 1 . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . d , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . b c . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


So interestingly, KataGo considers it fine to leave the lower left alone and respond on the top, but thinks white must play this way rather than playing the human joseki 1-point-jump. I'm not sure whether it usually prefers this particular pattern as the local followup for white given the upper left shape or if it's specific to this board. I think the idea at least in this case is that now black's left stones are a bit heavier and attackable, making it much harder for black to have time to take the lower left corner.

Locally, KataGo expects black 'a' next, but if black does rush to the lower left at 'b' or 'c' next, then white 'd'. :study:


Thanks. :) Very interesting.

OC, the peep was joseki at the time.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #8 Posted: Mon Jun 15, 2020 4:10 am 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Problem 4: I'm not giving away anything by saying that White needs to choose between continuing locally (if so where), or playing elsewhere (if so where).

Click Here To Show Diagram Code
[go]$$W Reference sequence
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . 1 . . . . . . . . 4 . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . 2 . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . 5 . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . 3 . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


Superficially traditionally, :w1: induces :b2: which induces :w3: ; if next Black encloses the lower right, both have an ideal extension, so :b4: encloses the other corner and White approaches the lower right. This looks too good for White, so Black will probably play elsewhere with :b2: preferably enclosing the lower right. So, should White play differently locally ? Or play the lower right immediately?
Click Here To Show Diagram Code
[go]$$W Reference sequence
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . 3 . . . . |
$$ | . . O , 1 . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . 2 . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


:w1: puts more pressure on the left side so :b2: gets more weight but then :w3: is almost compulsory. It's unclear who'll get the lower right now.

Click Here To Show Diagram Code
[go]$$W Reference sequence
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . 1 . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Since Black can't pressure the top left very severely, due to the hanging connection, :w1: can probably go to the important corner right away.

This is my choice.


This post by Knotwilg was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #9 Posted: Tue Jun 16, 2020 6:46 am 
Oza
User avatar

Posts: 2401
Location: Tokyo, Japan
Liked others: 2338
Was liked: 1332
Rank: Jp 6 dan
KGS: ez4u
I think that we need to realize the context here. Katago calculates that :w4: is a mistake worth several % in winrate and a few tenths of a point. I was surprised that GoGoD has 137 examples of this fuseki. The normal human continuation has been to simply enclose the upper right, mainly making the small knight shimari. Katago calculates that this is around 5% worse than playing b at the 3-3 point in the lower left. The approach at :b5: is the second most popular idea in human play. However, there are only two games with Black replying to :w6: with :b7:. Instead Black more often initiates some form of the nadare joseki with c in place of :b7:. Katago calculates that White should reply to :b5: locally. However, after :w8:, Black should switch to b in the lower left. The difference is not large but Black is still better of by being the first to play there. When Black connects instead, White would be better off to enclose the bottom left and when White extends at the top instead, Black should take the last opportunity to invade.

Click Here To Show Diagram Code
[go]$$Bc Black to play.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 2 . 0 . . . . . . . . a . . . . |
$$ | . . 8 c . . . . . , . . . . . , 1 . . |
$$ | . . 6 5 . . . . . . . . . . . . . . . |
$$ | . . 7 . . . . . . . . . . . . . . . . |
$$ | . . . 9 . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . 4 . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . b . . . . . . . . . . . . 3 . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

At this point there are two unfinished corners for Black to choose from. Blue is the large knight's enclosure at the bottom right while the same large knight enclosure in the top right is a close competitor. But there is something else here as well. Katago calculates that White's invasion at a on the left side is a very important point. In the actual game White never plays there. However, looking at katago's analysis the issue persists. This does not appear in human play.

Click Here To Show Diagram Code
[go]$$Bc Black to play.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . . . . . . . . . . . . . |
$$ | . . O . . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . a . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . O . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Policy for White when Black takes the bottom.
Attachment:
Policy for White when Black takes the bottom side 800.jpg
Policy for White when Black takes the bottom side 800.jpg [ 112.14 KiB | Viewed 6906 times ]

If we run a search in GoGoD on the left side of the board in the game. There are 260 hits. Only 7 of the games have the next play at k below. The ten plays that appear more frequently are marked a through j (variously for White and Black, e.g. White's most frequent next play in this area is the extension at a while Black's is the attachment at b). In reality play normally moves elsewhere, often for considerable lengths of time.

Click Here To Show Diagram Code
[go]$$Wc Continuations
$$+ - - - - - - -
$$| . . . . . . .
$$| . . . . . . .
$$| . . . O . O .
$$| . . O , i . f
$$| . . O X . . g
$$| . . X . . . j
$$| . . . X . . .
$$| . . k . . . .
$$| . . . . . . .
$$| . . . , . . .
$$| . . X . . . .
$$| . . . . . . .
$$| . . e . . . .
$$| . . . . . . .
$$| . . O . . . .
$$| . . . , . . h
$$| . . . O b c a
$$| . . . d . . .
$$| . . . . . . .
$$+ - - - - - - -[/go]

_________________
Dave Sigaty
"Short-lived are both the praiser and the praised, and rememberer and the remembered..."
- Marcus Aurelius; Meditations, VIII 21


This post by ez4u was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #10 Posted: Fri Jun 19, 2020 12:53 pm 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Applying the LZ Opening Gospel to problem 4:

No urgent local situations or class -3 to -1.
The presence of the 5-3 stone means it's not an empty corner, but class 1 tells you to play only 4-4 or 4-3 as the first move in corner, and a litle interpretation of the gospel tells you correcting your mistake of playing silly distant corner moves by turning them back into a good 3-4 shimari is somewhere between class 1 (first move in empty corner) and class 2 (shimari/approach a 3-4). So the gospel says white should play:

Click Here To Show Diagram Code
[go]$$Wc White to play.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . , X . . |
$$ | . . O X . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . 1 . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


Gospel gets it right again! (Honestly, I didn't bribe Bill to set this up)

Gospel: 3, Humans: 0, N/A: 1


This post by Uberdude was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #11 Posted: Sat Jul 25, 2020 11:25 pm 
Dies with sente

Posts: 113
Liked others: 11
Was liked: 27
Rank: 1d
Universal go server handle: iopq
Uberdude wrote:
Applying the LZ Opening Gospel to problem 4:

No urgent local situations or class -3 to -1.
The presence of the 5-3 stone means it's not an empty corner, but class 1 tells you to play only 4-4 or 4-3 as the first move in corner, and a litle interpretation of the gospel tells you correcting your mistake of playing silly distant corner moves by turning them back into a good 3-4 shimari is somewhere between class 1 (first move in empty corner) and class 2 (shimari/approach a 3-4).


In this case, because you made a low shimari it's not actually efficient according to fat KataGo.

But having the white stone at 5-4 is JUST as good as 4-4, where the Black win rate is 50.5% or so in both cases. With the stone at 5-4 it likes the shimari, with the stone at 4-4 it likes approaching another 3-4

You may also notice that playing 3-3 vs. a 5-4 stone is good, but not necessarily better than vs. a 4-4 stone


This post by iopq was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Opening problems for AI: Problems 4 and 5
Post #12 Posted: Fri Jun 04, 2021 9:55 am 
Lives in gote

Posts: 470
Liked others: 62
Was liked: 278
Rank: UK 2d Dec15
KGS: mathmo 4d
IGS: mathmo 4d
Speculatively:
Problem 4
Current position:
1. Seems obvious. Shimari at D3.
2. Planning to pincer at D11 doesn't seem effective when D3 is so big. If not D3, I think E14 can be considered.
3. I prefer R7 to P17

Past:
1. Seems pretty even, the lower left shimari should be the large knights, but the hanging connection in the upper left is dubious.

Check:
1. Perhaps once W has a shimari on the lower left, B has miai on the lower right of big moves either side. Hence O16 is better than Q6
2. My feeling was that W P17 remains low on the upper side even if W gets the next move at F17.

Problem 5
Current position:
1. Standard. B should play D3 or C3 of course. I think I prefer C3 here to attack in combo with the hanging connection.
2. Otherwise, I can't choose between O16 and Q6


Check:
1. This steals the lower side from W.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 12 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group