It is currently Tue Nov 13, 2018 10:23 pm

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 58 posts ]  Go to page Previous  1, 2, 3
Author Message
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #41 Posted: Thu Aug 30, 2018 1:55 am 
Lives in gote
User avatar

Posts: 556
Location: Carlisle, England
Liked others: 196
Was liked: 339
IGS: Reisei 1d
Online playing schedule: When I can
John Fairbairn wrote:
Hi Tami

Since you found Yang's classification style useful. I thought you may like a recommendation for a follow-up.

The book I recommend is a fairly new one (2015) and is called "Surrounding territory efficiently: the four basic points" by Kimu Sujun (ISBN 978-4-8399-5504-5).



Thanks for that. I'll look into that one, too.

The funny thing about the classification system is that it makes more sense to me in light of the second and third chapters of the book, and when one starts to diverge from and play freely it.

What I mean is the realisation that a move is only big (or not big) because of its relationship with what has been played before.

Now, corners are obviously Class 1 - experience, logic and AI all show that you cannot do better than start the game by playing in an empty corner. So it's the next lot of moves - "Class 2" and downwards that need the thinking.

I can tell you that if one tries to stick to Yang's ranking strictly, then it just won't work in real games. I've tried it, and the wheels come off pretty quickly. The reason is that the little table of moves on page 18 (of the edition I have, at any rate) assumes that the opponent is going to "play nicely" and not interfere with your own plans too soon. If your opponent were just to let you build away happily, making first your enclosures, then your happy little extensions in sides with facing corners, then your other extensions, etc., etc., you would indeed produce a pretty powerful opening. This is the ideal world.

But once the opponent decides to stick his oar in, then we enter the real world. However, this is what I think I've learned: the better you understand the reasons for the each kind of move's prioritisation in the ideal world, the better you can adapt to living in the real world. Because I've seen how to build cathedrals in peace, I feel better able to improvise fortifications in war.

_________________
Learn the "tea-stealing" tesuji! Cho Chikun demonstrates here:


This post by Tami was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #42 Posted: Thu Aug 30, 2018 9:13 pm 
Lives with ko

Posts: 170
Liked others: 126
Was liked: 97
Uberdude wrote:
Side extensions from 4-4s used to be played before knight enclosures from them (e.g. see few-decades-old Japanese games in Go World). That's what I learned 10+ years ago. Maybe top pros had already abandoned that idea back then, but at least in the 2010s early knight enclosure from 4-4s have become much more popular (this was before AlphaGo, maybe Korean-driven) and I picked this up from watching pro games and teaching from others with more up to date knowledge.


Very accurate observation about the knight enclosure, Uberdude!

Here are some graphs showing the frequency of three extensions from the 4x4 point (based on data from my personal pro game collection).

Old years are grouped together in larger intervals, so the x-axis is not linear. Also, if one wants to compare the relative frequencies of patterns to each other, the y-axis has different ranges for different patterns, but otherwise each graph is a good indication of how the popularity of a given pattern changed through history.

Attachment:
mid_low_hoshi_extension.PNG
mid_low_hoshi_extension.PNG [ 164.41 KiB | Viewed 1239 times ]


Attachment:
mid_high_hoshi_extension.PNG
mid_high_hoshi_extension.PNG [ 166.42 KiB | Viewed 1239 times ]


Attachment:
keima_hoshi_extension.PNG
keima_hoshi_extension.PNG [ 164.19 KiB | Viewed 1239 times ]

_________________
Sorin - 361points.com


This post by sorin was liked by 3 people: Bill Spight, explo, Uberdude
Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #43 Posted: Fri Aug 31, 2018 12:17 am 
Judan

Posts: 5276
Location: Cambridge, UK
Liked others: 294
Was liked: 2771
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Thanks sorin, nice to see quality data to confirm my impressions. Also interesting the difference between low and high middle extensions, with low having its peak in the first half of the 20th century (shin fuseki?) whilst high had a boom around 1990. I suspect the latter could be the influence of Takemiya and other san-ren-sei fans.

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #44 Posted: Fri Aug 31, 2018 4:29 am 
Judan

Posts: 7807
Liked others: 2140
Was liked: 2739
Thanks, sorin. :)

What about the ogeima? Thanks.

And while we're at it, the ikken tobi? If it's not too much trouble. :)

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #45 Posted: Fri Aug 31, 2018 5:22 pm 
Lives with ko

Posts: 170
Liked others: 126
Was liked: 97
Bill Spight wrote:
Thanks, sorin. :)

What about the ogeima? Thanks.

And while we're at it, the ikken tobi? If it's not too much trouble. :)


Gladly, but I will only be able to do that in 10 days or so, when I will be back at my computer.

_________________
Sorin - 361points.com


This post by sorin was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #46 Posted: Fri Aug 31, 2018 8:59 pm 
Judan

Posts: 7807
Liked others: 2140
Was liked: 2739
sorin wrote:
Bill Spight wrote:
Thanks, sorin. :)

What about the ogeima? Thanks.

And while we're at it, the ikken tobi? If it's not too much trouble. :)


Gladly, but I will only be able to do that in 10 days or so, when I will be back at my computer.


Thanks, sorin. :) When I was learning go those were the main shimaris for a 4-4 corner. The keima and iron pillar were sometimes used.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #47 Posted: Sun Sep 02, 2018 5:13 am 
Judan

Posts: 7807
Liked others: 2140
Was liked: 2739
Another curious AlphaGo Teach position.

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


Before AlphaGo this would have been considered a bread and butter fuseki position. But in the AI era it is hardly surprising to find that Black has lost ground with the now obsolescent joseki in the top left corner. Nor is it hard to predict White's next play, the 3-3 invasion in the bottom right corner. :) But AlphaGo has also played the approach at "b", the usual choice of humans not so long ago.

As with the previous AlphaGo Teach position (see viewtopic.php?p=235635#p235635 ), what is curious is the difference between the winrates of "a" and "b". The Black winrate for "a" is 40.9%, the Black winrate for "b" is 44.7%, for a difference of 3.8%. :shock: Playing through the variations indicates that the difference is not a typo. Did AlphaGo make a mistake when it played "b"?

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #48 Posted: Sun Sep 02, 2018 5:59 am 
Lives in sente

Posts: 1304
Liked others: 312
Was liked: 353
Rank: 5d
GD Posts: 1000
I assume that komi is factored in to the winrates. Correct? Otherwise I would expect Black to have a winrate of over 50%, unless the unfavorable joseki choice in the upper left has made a major change in Black's prospects.

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #49 Posted: Sun Sep 02, 2018 8:05 am 
Judan

Posts: 7807
Liked others: 2140
Was liked: 2739
gowan wrote:
I assume that komi is factored in to the winrates. Correct? Otherwise I would expect Black to have a winrate of over 50%, unless the unfavorable joseki choice in the upper left has made a major change in Black's prospects.


Oh, yes, AlphaGo assumes a komi of 7.5 pts. by area scoring. Black's initial winrate is 47.1% for a 4-4 play.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #50 Posted: Sun Sep 02, 2018 7:32 pm 
Oza
User avatar

Posts: 2054
Location: Tokyo, Japan
Liked others: 1854
Was liked: 1146
Rank: Jp 6 dan
KGS: ez4u
I definitely believe that our perceptions have been changed by the AI successes. However, we have to be a little more discerning with the data in building a story of the effect on humans playing our game. The situation is more complex than indicated by Sorin's graphs up to now. Sorry I don't have time to build some alternatives right now. :sad:

Basically the extensions were mainly played by Black while the small-knight shimari was originally mainly played by White (popularity based on the "large komi" effect). The Black extensions shown above on and under the star point were displaced by the Chinese fuseki more than by the shimari. An analysis based on 10x10 corners that does not distinguish between black and white positions will be misleading to some extent.

I haven't been keeping my database up to date this year so I cannot check the situation in 2018. However, up to late 2017 the combination of the various "extensions" clustered around the side star point remained much more common for Black than the shimari. I put "extensions" in parenthesis because the side stone in the Chinese fuseki may be a low pincer against a white approach stone as well (for example).

_________________
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: Yilun Yang: The Fundamental Principles of Go
Post #51 Posted: Sun Sep 09, 2018 6:13 pm 
Lives with ko

Posts: 170
Liked others: 126
Was liked: 97
Back from vacation - I can address now Bill's request for more hoshi extension patterns.

Attachment:
ogeima_hoshi_extension.PNG
ogeima_hoshi_extension.PNG [ 169.87 KiB | Viewed 835 times ]


Attachment:
ikken_tobi_hoshi_extension.PNG
ikken_tobi_hoshi_extension.PNG [ 170.9 KiB | Viewed 835 times ]


I was quite surprised that ikken-tobi is such a rare pattern (appears at #180 in my ordering, based on 10x10 patterns with min 2 and max 10 stones, I had to re-run my program since initially I was producing only 100 patterns and I couldn't find it :-) )

Looking at the y-axis scales (which tells us the relative frequency of a pattern among all other found patterns, for a given time interval): the keima extension from hoshi has max frequency 0.012 which is about 3X higher than the 0.0035 for ogeima, which in turn is 5X higher than the 0.0007 for ikken-tobi!

Here is another hoshi extension that I would not have thought to be more frequent than ikken-tobi extension, but at a peak of 0.0016 is more than twice more popular than ikken-tobi, for some reason:

Attachment:
ooogeima_hoshi_extension.PNG
ooogeima_hoshi_extension.PNG [ 171.13 KiB | Viewed 835 times ]


About ez4u comments: I did not think of that, but you are right: the patterns for black vs white would be different, I suspect even more so in the older, non-komi games! My procedure doesn't currently account for that, and the color of the stones in the diagrams is just a random choice. It would be interesting to differentiate between black vs white patterns indeed, time permitting :-)

_________________
Sorin - 361points.com


This post by sorin was liked by 2 people: Bill Spight, Waylon
Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #52 Posted: Sun Sep 09, 2018 8:33 pm 
Judan

Posts: 7807
Liked others: 2140
Was liked: 2739
Many thanks, sorin. :)

Before the keima became popular, I would have thought that the ogeima would have been popular, followed by the ikken tobi.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #53 Posted: Sun Sep 09, 2018 10:30 pm 
Lives with ko

Posts: 170
Liked others: 126
Was liked: 97
Bill Spight wrote:
Many thanks, sorin. :)

Before the keima became popular, I would have thought that the ogeima would have been popular, followed by the ikken tobi.


Now that I think about it - I guess the reason for the ikken-tobi enclosure's lack of popularity (the way I measure it, in an otherwise empty 10x10 board quadrant) has to do with the fact that it's usually used more "contextually", in relation to other stones, starting to build a moyo.

While the keima enclosure (at least in recent years) seems to be played more "out of the blue", without any obvious connection to building a moyo, but as a solid development move, and also preventing the opponent from having the freedom to choose the side for an approach move.
Of course an ikken-tobi enclosure can be used for the same purpose, but since it's open on the side (and the opponent can invade and live in 3x3 unconditionally, unlike after a keima enclosure), it is less preferred (?).

_________________
Sorin - 361points.com

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #54 Posted: Mon Sep 10, 2018 1:47 am 
Judan

Posts: 7807
Liked others: 2140
Was liked: 2739
sorin wrote:
Bill Spight wrote:
Many thanks, sorin. :)

Before the keima became popular, I would have thought that the ogeima would have been popular, followed by the ikken tobi.


Now that I think about it - I guess the reason for the ikken-tobi enclosure's lack of popularity (the way I measure it, in an otherwise empty 10x10 board quadrant) has to do with the fact that it's usually used more "contextually", in relation to other stones, starting to build a moyo.


Right. I wasn't thinking of an ikken tobi out of the blue.

_________________
The Adkins Principle:

At some point, doesn't thinking have to go on?

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #55 Posted: Mon Sep 10, 2018 1:59 am 
Oza

Posts: 2216
Liked others: 14
Was liked: 3175
To understand the fascinating data sorin is showing (which in turn is part of our attempt to understand AI moves), we seem very much hampered by the traditional vocabulary.

Extension (hikari) is such a wishy-washy word. In the best-case scenario when one player has, say, a strong corner, 'extension' has a meaning we can relate to - moyo-building. In a slightly more complex fuseki, an extension may be designed to limit the opponent's expansion and we tend to recognise that by using a special term: a tsume or check, or checking extension. Using the same train of thought, we should perhaps be calling a hikari from a strong position an 'expansion' (which to me includes a 'swelling' nuamce). That is, we should be trying to convey its purpose rather than just describing its look.

The reason for making that observation is that the sorts of extensions in sorin's data are just extensions-by-look. The name tells us nothing about their purpose. It's too early for them to be moyo-building expansions or tsume moves. They are certainly not shimaris. What are they doing? I'm not really sure. The best I can come up with is a 'developing' move but even that is wishy-washy.

More important, a single word for all the types of extension shown here is unlikely to exist. A move like the centre play in a Chinese fuskei position or a sanrensei obviously has some element of connection with a stone in the other corner. Ideally that should be acknowledged with a word that means something like fuseki-connector. A move like a keima has a hugely different feel to it. In fact we tacitly recognise that already by sometimes calling it a shimari, but that is quite wrong. So we need a new word for that category, but what? Again I'm not sure myself. In this case that's mainly because I don't really understand what these keima moves are really trying to achieve. Yes, they have some element of protecting the corner, but - judging by old Chinese games where long extensions were quite rare and connected groups were a major goal - they are also seen by pros as moves with some connection impulse.

Any suggestions for what each kind of hitherto called extension is trying to achieve, and what we might call them to convey that? If we try to get away from the extension idea and call them 'marker posts' or some such word that a builder might use when laying out (which is what fuseki refers to) a building site, what are the posts marking and where do they point?

For reference, a shimari is a move that 'closes the door' and locks it shut, puts the bolt in, posts a Keep Out sign. Entry is only by housebreaking through a side window, not via a door and Welcome mat. It does not leave the door ajar or on the sneck. It's not even strictly a 'corner enclosure'. The position in the upper left below is therefore not a shimari (shifted one line to the left, though, it would be). The triangled move below is a shimari. This 'house' just happens to be a big one owned a rich person, but he still needs locks and bolts. This non-corner usage of shimari is explicitly mentioned by Hayashi Yutaka, but I have seen it myself (both on the sides and in the centre). So what do we say the upper left position is doing? It seems more like setting up an open cage as a trap than trying to seal off the corner. Can we see the notorious early 3-3 invasion as grabbing the bait before the trap is set? Has anyone noticed whether AIs play 3-3s once a trap is set? I haven't noticed any such behaviour.

Such a lot of questions to answer and so little time...



This post by John Fairbairn was liked by: sorin
Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #56 Posted: Tue Sep 11, 2018 2:23 am 
Oza

Posts: 2216
Liked others: 14
Was liked: 3175
I thought the following game from yesterday (10 Sept, Liao Xingwen vs Zhang Tao) was a good example of the new questions thrown up about extensions here as pros try to play like AIs.



There are several aspects here. One is the currently fashionable (though far from new) omission of the connection at A. Playing the connection presumably commits Black to extending on the upper side and loses plays on the right side. Omitting it gives him the choice to go either way. Why and how does he make that choice.

Another is the position below. Black played the squared move on the left side making a sort of extension (with what goals?) and letting White get to the lower side first. A Black play on the lower side would have been some sort of extension/connection, but perhaps more accurately a kakoi (surrounding move). Calling it that makes it easier for me to understand - you try not to play kakoi moves in pure gote. But the AI-influenced shape in the lower right must be a factor. For want of a better word it might be called a high shimari, but it's not a shimari really. It's more of a come-on-in-and-get surrounded shape, and that would work well if Black gets the opportunity to exercise his option of playing first on the right side in the upper right corner. But, still, Black is accepting a hard time for his two stones on the lower left side and any White strength acquired from attacking those must impact on the lower right Black thinness.

A final aspect is that Black has several extensions or potential extensions. White has none.

What on earth is going on?

Here is the full game so you can see how it panned out. Note that White did not get any extensions after all. Black did get a big corner on the lower right but elsewhere he just had long strings of stones. White was compact and well-balanced everywhere. White won. Are extensions something to be wary of?


Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #57 Posted: Tue Sep 11, 2018 6:24 am 
Lives in sente
User avatar

Posts: 964
Location: Earth
Liked others: 301
Was liked: 157
A is just a bad move.



Let us discuss it in another thread: forum/viewtopic.php?f=48&t=16061

Top
 Profile  
 
Offline
 Post subject: Re: Yilun Yang: The Fundamental Principles of Go
Post #58 Posted: Tue Sep 11, 2018 7:22 am 
Lives in sente
User avatar

Posts: 964
Location: Earth
Liked others: 301
Was liked: 157
The squared move is part of an interesting pattern (not played out in this game, although it would have been an good option).

Let us have a look here: forum/viewtopic.php?f=48&t=16062

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 58 posts ]  Go to page Previous  1, 2, 3

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