It is currently Wed Feb 20, 2019 8:13 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 9 posts ] 
Author Message
Offline
 Post subject: MikeKyle analyses Hoshi, low approach, low 1 space pincer
Post #1 Posted: Thu Feb 14, 2019 4:52 pm 
Dies in gote

Posts: 67
Liked others: 7
Was liked: 15
Rank: EGF 2k
KGS: MKyle
I posted a while ago about some komoku, high approach, low pincer patterns. I thought I'd do another deep dive and share it here again for a feeling of accountability.

Again, I'm aiming at:
  • Having fun trying to understand go better
  • Improve my instinct in the opening
  • Perhaps pick up some good habits (shape, fighting, positional judgement)

I'm investigating and writing primarily for my own benefit, but I'd be thrilled if anyone could offer suggestions, corrections, thoughts, or even if people just read and take something from it.
I'm going to look at the fairly common position:
Click Here To Show Diagram Code
[go]$$B
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . 1 . . . . . .
$$ | . . . . . 2 . 3 . .
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ --------------------[/go]



I will stick to trying to consider some of these questions:

  • Black 3
    • When do professionals play black 3
    • When do AI (mainly ELFv1 unless i find a new toy) play black 3
    • Meaning in relation to other pincers
    • Meaning in relation to other choices
  • What happens next
    • What happens next in joseki books
    • what happens next in pro play
    • What happens next in AI play
    • What happens in special cases
    • What happens next in weak play
    • Can/should deviations be punished

But I've been looking at batch analysis of patterns arising in pro games, so maybe it will be centered around this.


This post by MikeKyle was liked by: sorin
Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #2 Posted: Thu Feb 14, 2019 5:18 pm 
Dies in gote

Posts: 67
Liked others: 7
Was liked: 15
Rank: EGF 2k
KGS: MKyle
I did some automated analysis with Elfv1. I think the whole board position is important, so I wanted to look at the patterns in the context in which they happen.

I sketched out the joseki I knew and a few more from books allong with a couple of new patterns and a couple of things i was curious about in an sgf. I created some scripts that do a pattern search to find board positions where these patterns get played and run them through Elf. All games used are 9p vs 9p games. It then taggs the sgf with some summary stats about what Elf thinks about the move accross the range of board positions.



I classified things as a mistake only if Elf thinks the win rate drops by more than 3%. iqr is the inter-quartile range ie. the central 50% of the data. Since I filtered down to 9p vs 9p games only, some of the patterns don't have so much data.
I was planning to use this as a starting point and have a closser look at the patterns where Elf has strong opinions.


Attachments:
HoshiPincerPatterns.sgf [21.46 KiB]
Downloaded 108 times

This post by MikeKyle was liked by: marvin
Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #3 Posted: Fri Feb 15, 2019 5:52 pm 
Dies in gote

Posts: 67
Liked others: 7
Was liked: 15
Rank: EGF 2k
KGS: MKyle
There's one modern trend here that is very stark and certain and I don't think I really feel i properly understand. Maybe a stronger player could help me see?

In this shape:
Click Here To Show Diagram Code
[go]$$B
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . 7 . . . . . .
$$ | . . 6 1 . . . . . .
$$ | . . 4 5 a 2 . 3 . .
$$ | . . 0 8 9 b . . . .
$$ | . . . . . . . . . .
$$ --------------------[/go]


a has disappeared and b has replaced it.
It seems that Alphago did this in one of the master online games, and I remember it being in the teaching tool, and it's taken off since then. I still expected that it might depend on the whole board and that a might still be appropriate, but it looks like Elf1 thinks that a is always a mistake of at least around 5% (usually about 9%). b is apparently never a mistake in my sample of positions. Pros seem to agree - in my filtered down 9p vs 9p search the pattern with b appears only 7 times before 2017 (starting in 1961 with Go Seigen of course) but then appears 25 times in 2018 alone, whilst the pattern with a is down to only 3 times in the year.

I understand that playing at a leaves white with the move at b to start an attack on the wall, but is that really so bad when the pincer is tight? if it's dangerous then black can connect on top right? looking at the Elfs projected sequences, the AI seems to cash in on the aji fairly soon by just playing a move on the lower side, but surely you concede a lot of forcing moves on the lower side with b too? It seems that both choices leave behind aji/forcing moves, and in any choice like this there is going to be a trade off, but does anyone feel they have a good handle on why humans used to unanimously think that a offered the better deal and now think b does?


This post by MikeKyle was liked by: sorin
Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #4 Posted: Sun Feb 17, 2019 10:05 am 
Dies in gote

Posts: 45
Liked others: 8
Was liked: 8
Rank: 1kyu
KGS: struttnoddy
It always struck me that playing at 'a' looked like particularly bad shape when 'b' was available. It might be as simple as not making the empty triangle being a more efficient way to do the same thing (maintaining/increasing liberties for example).

My slightly more in depth undertanding of it is this, presuming some other stuff happens elsewhere on the board white might have the chance to play like this, black has to play 3 to separate the white groups and then comes under attack:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . 4 . . .
$$ | . . O X 1 O . X .
$$ | . . O O X 2 . . .
$$ | . . . . 3 . . . .
$$ +------------------[/go]


By comparison, if white tries to cut black in this variation, black's outside shape remains solid and white has to have significantly more support to attempt the cut. That's not to say it can't happen but it's harder for white to attack:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . . . . .
$$ | . . O X 2 O 3 X .
$$ | . . O O X 1 . . .
$$ | . . . . . . . . .
$$ +------------------[/go]



In addition, if black tenukis before connecting at all black can even sacrifice on the 2nd line to maintain good shape for relatively little gain on white's part:

Click Here To Show Diagram Code
[go]$$W
$$ | . . O . . . . . .
$$ | . . . X . . 8 . .
$$ | . . O X . 6 . . .
$$ | . . O X 1 O 4 X .
$$ | . . O O X 2 3 . .
$$ | . . . . 7 5 . . .
$$ +------------------[/go]


Alternatively white could descend looking for fewer points but keeping holes in black's shape, but black could easily defend at a, b or c.

Click Here To Show Diagram Code
[go]$$W
$$ | . . O . . . . . .
$$ | . . . X . a . . .
$$ | . . O X . b c . .
$$ | . . O X . O . X .
$$ | . . O O X 1 . . .
$$ | . . . . . . . . .
$$ +------------------[/go]


At this point if white playing to cut off the black stone on the 2nd line has become that good a move, either you have reached the early endgame (obviously it's still a huge move), or the black stones on the outside have already been neglected too much.

I think just looking at the shapes that arise from white attempting to cut through black in either variation it should be relatively obvious which is better locally. The only hesitation I have is that if black has come under heavy attack already it may be easier to defend like this if black had already connected solidly instead, starting to make eyeshape and getting out at a/b/c/d/e:

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


There may also be less aji in the corner for white when black doesn't connect solidly and likewise the 1st line endgame may end up being better for white, later on moves at 'a' or 'b' might be sente for white:

Click Here To Show Diagram Code
[go]$$W
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . . . . .
$$ | . . O X . O . X .
$$ | . . O O X X . . .
$$ | . . . a b . . . .
$$ +------------------[/go]


That's my two cents anyway! I think it's probably one of this instances where connecting underneath is locally better but it can also be highly situational


This post by Struttnoddy was liked by 2 people: MikeKyle, sorin
Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #5 Posted: Sun Feb 17, 2019 3:07 pm 
Lives with ko

Posts: 249
Liked others: 208
Was liked: 128
MikeKyle wrote:
In this shape:
Click Here To Show Diagram Code
[go]$$B
$$ | . . . . . . . . . .
$$ | . . . . . . . . . .
$$ | . . . 7 . . . . . .
$$ | . . 6 1 . . . . . .
$$ | . . 4 5 a 2 . 3 . .
$$ | . . 0 8 9 b . . . .
$$ | . . . . . . . . . .
$$ --------------------[/go]


a has disappeared and b has replaced it.


Even more surprisingly, when this shape is played without black's pincer present on the board, b is preferred to a nowadays, I have seen something like this:

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


Presumably, the idea is to make white heavier before connecting.

(Most recently, this occurred in the Sumire (soon to be 1p) game vs Choi 9p.

_________________
Sorin - 361points.com


This post by sorin was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #6 Posted: Sun Feb 17, 2019 5:18 pm 
Lives in sente

Posts: 1310
Liked others: 605
Was liked: 398
Rank: AGA 4k KGS 2k
GD Posts: 61
KGS: dfan
Struttnoddy wrote:
My slightly more in depth undertanding of it is this, presuming some other stuff happens elsewhere on the board white might have the chance to play like this, black has to play 3 to separate the white groups and then comes under attack:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . 4 . . .
$$ | . . O X 1 O . X .
$$ | . . O O X 2 . . .
$$ | . . . . 3 . . . .
$$ +------------------[/go]

By comparison, if white tries to cut black in this variation, black's outside shape remains solid and white has to have significantly more support to attempt the cut. That's not to say it can't happen but it's harder for white to attack:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . . . . .
$$ | . . O X 2 O 3 X .
$$ | . . O O X 1 . . .
$$ | . . . . . . . . .
$$ +------------------[/go]

The main thing I notice when comparing these diagrams (besides the fact that White has one more move coming to them in the second one) is that White has six liberties in the first and two in the second.

Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #7 Posted: Sun Feb 17, 2019 5:52 pm 
Lives with ko

Posts: 249
Liked others: 208
Was liked: 128
dfan wrote:
Struttnoddy wrote:
My slightly more in depth undertanding of it is this, presuming some other stuff happens elsewhere on the board white might have the chance to play like this, black has to play 3 to separate the white groups and then comes under attack:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . 4 . . .
$$ | . . O X 1 O . X .
$$ | . . O O X 2 . . .
$$ | . . . . 3 . . . .
$$ +------------------[/go]

By comparison, if white tries to cut black in this variation, black's outside shape remains solid and white has to have significantly more support to attempt the cut. That's not to say it can't happen but it's harder for white to attack:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . . . . .
$$ | . . O X 2 O 3 X .
$$ | . . O O X 1 . . .
$$ | . . . . . . . . .
$$ +------------------[/go]

The main thing I notice when comparing these diagrams (besides the fact that White has one more move coming to them in the second one) is that White has six liberties in the first and two in the second.


That's why black would not descend in the first diagram quoted above, but would play hane with 3 in the following diagram, at least that would be my default answer locally:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . 7 . . .
$$ | . . O X 1 O . X .
$$ | . . O O X 2 . 6 .
$$ | . . . . 5 3 4 . .
$$ +------------------[/go]


Of course, the descend is a viable move too, depending on the overall position.

_________________
Sorin - 361points.com


This post by sorin was liked by: dfan
Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #8 Posted: Tue Feb 19, 2019 5:32 am 
Dies in gote

Posts: 67
Liked others: 7
Was liked: 15
Rank: EGF 2k
KGS: MKyle
Thanks for answers struttnoddy, sorin, dfan.
I always forget about endgame considerations - I guess if we assume that there is territory on both sides (certainly doesn't seem a given for black) then I think b is two points better in swing value (the point at a + going from sente endgame for white under a, to double gotte under b. My technical endgame is terrible, so please correct me if this is wrong)

When thinking about the aji left if black plays a I wasn't really thinking about playing b directly because I thought white would rather keep good endgame and can cash in on the aji less directly. However I looked at some of the follow ups played by Elfv2 and it looks as though playing b is often the right move. (the hane underneath does seem to be better than the descent in all the games i analysed)

I took the batch of games that were analysed at the start, I cropped the sgfs at the choice between a and b and i used twogtp to get elfv2 to play out the rest of the games under each choice to see how the aji was used. I captured some of the interesting or recurring sequences in this sgf. It suffers badly from lacking the whole board context that guides the sequences!


a lot of the same sequences seem to be played either way, but maybe black can be a little bit bolder when connected underneath - if nothing else then it's still possible to sacrifice something.

I thought a fairly big difference is that this move seems much less intimidating if you're connected underneath:
Click Here To Show Diagram Code
[go]$$W
$$ | . . d . c . . . .
$$ | . . O 1 . . . . .
$$ | . . . X . . . . .
$$ | . . O X . . . . .
$$ | . . O X . O . X .
$$ | . . O O X X . . .
$$ | . . . . . . . . .
$$ +------------------[/go]


In the games where black chose the other move, black would often find time to play 'c', which is a nice move to play as it has nice follow up at d (if white is feeling slow, white might even protect at d) but perhaps it's more of a protection from white taking the nice forcing sequence on that side.


Attachment:
aji00108.sgf [2.26 KiB]
Downloaded 27 times


This post by MikeKyle was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: MikeKyle analyses Hoshi, low approach, low 1 space pince
Post #9 Posted: Tue Feb 19, 2019 12:10 pm 
Dies in gote

Posts: 45
Liked others: 8
Was liked: 8
Rank: 1kyu
KGS: struttnoddy
sorin wrote:

That's why black would not descend in the first diagram quoted above, but would play hane with 3 in the following diagram, at least that would be my default answer locally:

Click Here To Show Diagram Code
[go]$$B
$$ | . . O . . . . . .
$$ | . . . X . . . . .
$$ | . . O X . 7 . . .
$$ | . . O X 1 O . X .
$$ | . . O O X 2 . 6 .
$$ | . . . . 5 3 4 . .
$$ +------------------[/go]


Of course, the descend is a viable move too, depending on the overall position.


I would definitely agree the hane seems best in this situation, but I think the point stands that if white has any strength nearby black is likely to come under attack after white descends, hane or not. Black probably does have more eye shape and liberties with the hane, but the situation could have been avoided entirely by playing the other variation, albeit opening the door to a whole different set of possibilities!

Additionally if black hanes it removes something of the corner aji by giving white the decent at 'a' in sente later, which if black has sealed the white corner group in could be very useful:

Click Here To Show Diagram Code
[go]$$W
$$ | . . . . . . . . .
$$ | . . Y . . . . . .
$$ | . . Y . Y . . . .
$$ | . Y O . . . . . .
$$ | . Y O X . . . . .
$$ | . O O X . . . . .
$$ | . . O X X O . X .
$$ | . . O O X 1 . . .
$$ | . . . a . 2 . . .
$$ +------------------[/go]


This might not be unrealistic if the black group had to run up the board somewhat but would of course be a minor loss when compared to saving the outside black group, far be it from me to disagree with Elf!

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 9 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 5 guests


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:  
cron
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group