It is currently Sat Feb 23, 2019 5:27 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 21 posts ]  Go to page 1, 2  Next
Author Message
Offline
 Post subject: Wacky centre move training from bots.
Post #1 Posted: Sun Jul 29, 2018 10:19 am 
Judan

Posts: 5590
Location: Cambridge, UK
Liked others: 313
Was liked: 2951
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
One aspect of Go that bots seem to be particularly good at compared to humans is vague centre moves (though LeelaZero wasn't much impressed by Shusaku's famous ear-reddening move). So I thought I'd collect some examples I come across in my bot reviews that stand out to me. Maybe a top pro would also find them, but my feeling is they are pretty wacky even for top humans. I'll hide the answer so you can have a go figuring it out, and once seen can you see some reason why it makes sense?

#1 a variation from Ilya (black) vs Artem in PGETC finals, white to play

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


LZ #157 says
Click Here To Show Diagram Code
[go]$$Wc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . O X X X O X . . . . . . . . . . . . |
$$ | O O O O O X X . X . . . . . . . . . . |
$$ | . O X X X O O . c b . . . X . , X . . |
$$ | . . . . O X O . . . . . . . . . . . . |
$$ | . . O X X X . . . . . . d . . . . . . |
$$ | . . O . . . . a . 1 . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . O . , . . . |
$$ | . . . . . O . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Expects a-d continuation (d was originally one point lower). I can see what these moves are doing. The initial move might end up being move 5 or 7 in the sequence of pulling out the 3 stones in some variation so that could be one way of understanding the shape relationship to existing or potentially existing stones; the 1-2-3 principle in extreme. e.g. this variation from Elf with black resisting for 4 and fancy 5 to make k17 push bad:
Click Here To Show Diagram Code
[go]$$Wc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . O X X X O X . . . . . . . . . . . . |
$$ | O O O O O X X . X . . 8 . . . . . . . |
$$ | . O X X X O O . 7 3 . 5 6 X . , X . . |
$$ | . . . . O X O . . . . 4 . . . . . . . |
$$ | . . O X X X . . . . . . . . . . . . . |
$$ | . . O . . . . 2 . 1 . . . . . . . . . |
$$ | . . . . 9 . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . O . , . . . |
$$ | . . . . . O . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


Elf says (with LZ's as 2nd choice):
Click Here To Show Diagram Code
[go]$$Wc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . O X X X O X . . . . . . . . . . . . |
$$ | O O O O O X X . X . . . . . . . . . . |
$$ | . O X X X O O . . , . . . X . , X . . |
$$ | . . . . O X O . . . . . . . . . . . . |
$$ | . . O X X X . . . . . . 1 . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . d . . X . . |
$$ | . . . . . . . . . . . c . . . . . . . |
$$ | . . O , . . . . . , . . . O . , . . . |
$$ | . . . . . O . . . b . a . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Expects black to tenuki with a-d to follow. Hard for me to understand the one stone just floating there, though I can see it could be helpful in pulling out the 3 stones later.


This post by Uberdude was liked by 2 people: Elom, Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #2 Posted: Sun Jul 29, 2018 11:58 am 
Honinbo

Posts: 8061
Liked others: 2280
Was liked: 2829
Uberdude wrote:
One aspect of Go that bots seem to be particularly good at compared to humans is vague centre moves (though LeelaZero wasn't much impressed by Shusaku's famous ear-reddening move). So I thought I'd collect some examples I come across in my bot reviews that stand out to me. Maybe a top pro would also find them, but my feeling is they are pretty wacky even for top humans.

{snip}

Elf says (with LZ's {"e"} as 2nd choice):
Click Here To Show Diagram Code
[go]$$Wc
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . O X X X O X . . . . . . . . . . . . |
$$ | O O O O O X X . X . . . . . . . . . . |
$$ | . O X X X O O . . , . . . X . , X . . |
$$ | . . . . O X O . . . . . . . . . . . . |
$$ | . . O X X X . . . . . . 1 . . . . . . |
$$ | . . O . . . . . . e . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . d . . X . . |
$$ | . . . . . . . . . . . c . . . . . . . |
$$ | . . O , . . . . . , . . . O . , . . . |
$$ | . . . . . O . . . b . a . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Expects black to tenuki with a-d to follow. Hard for me to understand the one stone just floating there, though I can see it could be helpful in pulling out the 3 stones later.


I expect that John Fairbairn can relate :w1: with Go Seigen. :) It aims, for one thing at the jump attachment, and possibly at an invasion on the right side. I seem to recall a similar move by Hashimoto Utaro (before the New Fuseki era), and a similar move by Maeda in a handicap game, a reduction aiming at two possible invasions.

_________________
There is one human race.
----------------------------------------------------

The Adkins Principle:

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

— Winona Adkins


Last edited by Bill Spight on Sun Jul 29, 2018 12:48 pm, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #3 Posted: Sun Jul 29, 2018 12:41 pm 
Judan

Posts: 5590
Location: Cambridge, UK
Liked others: 313
Was liked: 2951
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Yes, I was thinking that moves like this might have been more common by pros in times past, e.g. shin fuseki era. Also I wonder if pros of today do think about them more than they play them, but chicken out of playing them in serious tournaments because they are unsure if they are good so apply caution and go for a more traditional move they can be more confident is pretty good if not the best, whereas these centre moves might be brilliant genius moves or utter rubbish. Perhaps if they drank alcohol instead of tea whilst they played inhibitions would be lowered and we would see more creative play ;-).


This post by Uberdude was liked by 2 people: eyecatcher, Pangolino
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #4 Posted: Sun Jul 29, 2018 12:54 pm 
Honinbo

Posts: 8061
Liked others: 2280
Was liked: 2829
Uberdude wrote:
Perhaps if they drank alcohol instead of tea whilst they played inhibitions would be lowered and we would see more creative play ;-).

;)

IMO the 21st century so far has been a time of creativity in go, even before the advent of AlphaGo, and modern neural network bots are only adding fuel to the fire. :D

_________________
There is one human race.
----------------------------------------------------

The Adkins Principle:

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

— Winona Adkins


This post by Bill Spight was liked by 2 people: eyecatcher, gowan
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #5 Posted: Thu Aug 02, 2018 12:49 am 
Judan

Posts: 5590
Location: Cambridge, UK
Liked others: 313
Was liked: 2951
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
From Shibano Toramaru 7p vs Huang YiJ 8p (?). White to play.

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


Elf v1 says black's last move was quite a mistake (-10% in elf strong opinions land) and white should reduce/invade here (and then black's best answer is to attach to the right of it!)
Click Here To Show Diagram Code
[go]$$Wc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . O O O . |
$$ | . . . . . O . . . . . . . O . X O X . |
$$ | . . . O . . . . . , . O . . . X O X . |
$$ | . . . . . . . . . . . . . . . X O X . |
$$ | . . . . . . . . . . . X . X . O X O . |
$$ | . . . . . . . . . . . . . . . . X X . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . 1 . . . . . . |
$$ | . . . . . . X . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O X . . , . . . . . , . . . |
$$ | . . . . . O . . X . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


A very close 2nd choice is the n13 peep: it's bad (-8) for black to connect and then white plays the same n7 and is even better. So black should resist like this and get sente to play the same important n7 point, or after more thought n8 (but white is not bad either, I suppose top side is reinforced and peep stone still lively):
Click Here To Show Diagram Code
[go]$$Wc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . O O O . |
$$ | . . . . . O . . . . . . . O . X O X . |
$$ | . . . O . . . . . , 3 O . . . X O X . |
$$ | . . . . . . . . . . . 2 . . . X O X . |
$$ | . . . . . . . . . . . X . X . O X O . |
$$ | . . . . . . . . . . . . 1 . . . X X . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . 4 . . . . . . |
$$ | . . . . . . X . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O X . . , . . . . . , . . . |
$$ | . . . . . O . . X . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


Leela Zero #157 has similar thoughts (not so surprising as it has been trained on Elf games), but prefers the peep. If black connects (only a little minus unlike Elf) then it too wants n7, but thinks black should resist with m13 instead. If black m15 it thinks white should p15 rather than simple L16 extend. Interestingly Elf didn't consider this move (to 25k playouts) but when shown it likes it +4%, another example of Elf not being exploratory enough (so maybe I shouldn't assume Elf is stronger than LZ for these reviews).

Both Elf and LZ think black's m14 was slack and say c14 is best, but if black wants to grow the moyo want to do so in sente to play n7 key point, e.g. m15 attach and if white is soft and just extends (hane better) then both like this for black.
Click Here To Show Diagram Code
[go]$$Bc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . O O O . |
$$ | . . . . . O . . . . . . . O . X O X . |
$$ | . . . O . . . . . , 2 O . . . X O X . |
$$ | . . . . . . . . . . . 1 . . . X O X . |
$$ | . . . . . . . . . . . . . X . O X O . |
$$ | . . . . . . . . . . . . . . . . X X . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . 3 . . . . . . |
$$ | . . . . . . X . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O X . . , . . . . . , . . . |
$$ | . . . . . O . . X . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


Growing moyos with large knight moves is something I seem to recall Kong Jie and Choi Cheolhan liked doing, but reducing with them is a less familiar concept to me, but I suppose your key point is my key point! Now to train my intuition so that n7 glows in positions like this...

Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #6 Posted: Thu Aug 02, 2018 2:19 am 
Honinbo

Posts: 8061
Liked others: 2280
Was liked: 2829
I also got the impression that Elf is not exploratory enough. (And maybe Leela Zero, as well.) Certainly the usual disparity in playouts between its first and second choices gives that impression. If it did not produce such large deltas, I would not have so much confidence in its judgements, because of the small number of playouts. But again, these evaluations are for play, not analysis. To produce good play you don't have to accurately analyze all positions, as long as the ones you analyze less well do not occur very often.

I have also wondered if, at its level of play, depth of exploration is not more important than breadth. For instance, if it is relatively weak on ladders, depth of search may allow it to read a ladder out, and therefore may be more valuable than refining the evaluation of key points.

Anyway, this time I found N-07, taking a hint from the previous position and asking myself where is the best point for a reduction that aims at two invasions.

_________________
There is one human race.
----------------------------------------------------

The Adkins Principle:

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

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #7 Posted: Mon Sep 03, 2018 2:35 pm 
Judan

Posts: 5590
Location: Cambridge, UK
Liked others: 313
Was liked: 2951
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
From a game in the Shuko problem book, black to play. Given this thread it's obviously not answering that odd looking peep just played ;-) . LZ 157 and Elf v0 and v1 all quickly zero in on the move Hane Naoki played in the game, though for Elf v1 it's not #1 for long (but is still good).

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


h12

Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #8 Posted: Tue Sep 04, 2018 4:04 am 
Oza

Posts: 2290
Liked others: 14
Was liked: 3275
Andrew: Some thoughts, but first a small digression.

The position below is not a problem but part of a discussion from a book on how to "surround".



Kimu Sujun says the best move now is for White to play the hanetsugi starting at A. White can live inside Black's area if he starts at B, but Kimu says this is wrong (under the heading "jealousy is bad") and he cites, as the guide to how to surround, the maxim that you must actually make use of your sacrifice stones. The jealous line of course swingeingly damages the White stones in the lower left corner. LZ agrees with his choice.

After that little sequence White turns to the upper right:



Again LZ can apparently accept this move - it displays some alternatives but we are talking a difference of slivers of percentage points.

Kimu presumes Black A next, which is yet again well in the frame for LZ, although it perhaps may end up choosing B or C, again by a smidgeon, but as the book is about surrounding we can't assume Kimu wouldn't have considered and maybe preferred any of these alternative moves in a real game.

Kimu then finishes off his explanation by giving the White move below.



LZ looks at five moves in the this area (as well as A) but once again they are all within a percentage point. So we can say the human pro has done acceptably well so far, and we can presumably follow his advice.

The tricky part comes next. Kimu discusses the White alternative below.



Kimu's reason for rejecting this move was that it is too thin. What I found interesting was that, hitherto, LZ had basically been showing about a single set of traffic lights as it pondered each move, but now it suddenly lit up like a funfair at night, with well over a dozen moves clamouring for attention. That suggests to me some sort of criterion for computer thinking. The human pro generally sees "thinness = bad," the computer sees "thinness = fun!" Both are seeing the same manifestation but interpreting it differently.

The specific problem with the thinness for Kimu is that it invites the good erasing sequence Black A, White B, Black C, and it is good for Black because even after it White still remains thin. And, for the umpteenth time, LZ accepts the human move as one of an elite bunch of moves each almost inseparable from the other, so Kimu is not really open to criticism.

Now we come to the real reason I am posting this example. The erasing move, like the wacky centre moves already discussed and many other similar moves I have seen, involves "sector lines." This concept does not seem to get a mention much nowadays but I am going to suggest it may be a key to understanding AI play. The concept was not discovered by Wilcox (I have a much earlier book by Miyamoto with it in) but Wilcox did discuss it much more and I think we can safely go with his terminology.

However, his understanding of the concept may possibly be questioned. His basic premise was that you play on a sector line to erase. Go any deeper and you get surrounded and cut off. Go any shallower and you are making things easy for the opponent. I wonder whether this is yet another of the traps rationalists fall into - it's logical so it must be true. The real truth is that logic all depends where you start from. In the case of sector lines, the bots seem to start from a different point.

Here, although once more the LZ scores for all the possibilities cluster closely, it marginally prefers D or E, and it seems to me that sector line theory would advise against both.

In the previous examples (and others I have seen) LZ likewise seems to go against our sector line theory and most typically goes one or maybe two steps deeper. What I am arguing for is that (in black box terms) LZ does appear, like us, to follow some sort of sector line theory in that it "sees" sector lines but handles them differently. Sector lines by definition are thin. We worry about that. For a bot that's fun.

Going back to the "surrounding" theme, it's considered a high-level concept. I remember once startling Charles Matthews when I said something about being surprised the concept didn't seem to have filtered through to the west then. Charles's interest was triggered by the fact that Kim Seongjune had told him not long before that it was that concept, new to Charles, that had got him to 6-dan.

Nowadays there is some awareness of the concept here but my impression is that it's still not understood very well. The mindset seems to be that "surrounding" (kakoi, in its strategic technical-term sense) is about making territory tout court. Well, of course, that's involved, but there's a higher level to it. That higher level is "balance". The White keima at Q12 above is praised by Kimu not because it ensures territory but because it ensures balance.

Balance actually has a lot in similar with thinness. The master of balance in go was Takagawa. Bill will remember this well. Mr T was highly successful with that style for a long time, but eventually Sakata, the master of disruption, toppled him. I have a feeling that amateurs tend to think of balance in terms of things settling down, or stability, but the real meaning should be achieving dynamic poise while still on the tightrope.

I believe that was what Kimu was striving for, but even he, as a pro, is not as nimble as LZ on the tightrope. Although we have allowed ourselves to be dazzled by AI which can do fancy tricks on the rope whereas we have so far only been able to achieve a slightly undignified arms-out wobble, I do recall that we felt exactly the same sort of wonder and astonishment when Olga Korbut came along and showed us somersaults and prodigious leaps on the high beam when before we had basically given high scores for just standing up straight with dignity on the beam. And then Nadia Comaneci came along and outdid even Olga with a perfect 10/10!

That seems to me an analogy for what we are seeing today. We humans may think we have achieved some sort of balance in our go. The bots are confounding that. For us to regain poise we need perhaps to make wackiness more mainstream. We have discussed thickness in relation to bots but I don't think we've focused on thinness. It seems thin is fun. Thin is the new black (and white)? Thin is where the somersaults are. The sector lines are our beams.


This post by John Fairbairn was liked by 3 people: Marcel Grünauer, Pangolino, Uberdude
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #9 Posted: Tue Sep 04, 2018 6:20 am 
Lives in gote

Posts: 546
Location: Vienna, Austria
Liked others: 242
Was liked: 282
John Fairbairn wrote:
Kimu's reason for rejecting this move was that it is too thin. What I found interesting was that, hitherto, LZ had basically been showing about a single set of traffic lights as it pondered each move, but now it suddenly lit up like a funfair at night, with well over a dozen moves clamouring for attention. That suggests to me some sort of criterion for computer thinking. The human pro generally sees "thinness = bad," the computer sees "thinness = fun!" Both are seeing the same manifestation but interpreting it differently.


I tried to come up with a reasonable sequence of moves that leads to the original position and found that before White's invasion at h3 Black's winrate is already less than 10%. The invasion is apparently so bad as to bring it back to around 45%, but LZ suggests that Black should now descend to f2. Black's answer at h4 returns Black's winrate to below 10%.

For winrates below 10% sometimes LZ's candidate moves are all over the board. But I only see fewer than 10 candidates for all the discussed moves and LZ (with ELFv1 and less than 4000 playouts) prefers, for the "enclosing" move, the diagonal attachment at q15, then Kimu's "thin" jump and much later the keima. But all three moves are only about 3-4% apart; White's position is already that good.

That is, LZ would win with White. It assumes near-perfect play, but if White makes one or two slack moves, the winrate has heavy swings. So what is a good and safe move for LZ might be a risky move for humans.


This post by Marcel Grünauer was liked by: mitsun
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #10 Posted: Tue Sep 04, 2018 7:01 am 
Judan

Posts: 5590
Location: Cambridge, UK
Liked others: 313
Was liked: 2951
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
The examples I've picked so far have been invasion/reduction ones, probably because as a "Geronimo!" leap into the enemy area they are more dramatic and attract attention. But the bots also make nice surrounding moves, for example in that last example instead of white's dodgy peep Shuko suggested a surrounding move near tengen, whilst LZ had a similar idea but a bit more conservative (different judgement on current position / komi? or saw it was too easy to sabaki inside with Shuko's larger move?). With the black move I suppose it's aiming at some thinness of white's g15 knight move so can go a bit closer there, and the pro in the game did indeed attack/defend from j14 direction soon.

About sector lines, I recall reading/hearing somewhere that whilst a used to be a classic sector line breaking reduction idea, some pros had come up with :w1 as being better because they looked at a load of plausible continuations after black's surround at b and they always find a good (but complicated!) way for white to manage the situation. And just now I did a pattern search and it's an older idea than I imagined, Rin Kaiho did it against Takemiya in 1977 and Fujisawa several times in the 90s!

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

Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #11 Posted: Tue Sep 04, 2018 8:10 am 
Lives in gote

Posts: 396
Liked others: 165
Was liked: 179
Uberdude wrote:
The examples I've picked so far have been invasion/reduction ones, probably because as a "Geronimo!" leap into the enemy area they are more dramatic and attract attention. But the bots also make nice surrounding moves, for example in that last example instead of white's dodgy peep Shuko suggested a surrounding move near tengen, whilst LZ had a similar idea but a bit more conservative (different judgement on current position / komi? or saw it was too easy to sabaki inside with Shuko's larger move?). With the black move I suppose it's aiming at some thinness of white's g15 knight move so can go a bit closer there, and the pro in the game did indeed attack/defend from j14 direction soon.

About sector lines, I recall reading/hearing somewhere that whilst a used to be a classic sector line breaking reduction idea, some pros had come up with :w1 as being better because they looked at a load of plausible continuations after black's surround at b and they always find a good (but complicated!) way for white to manage the situation. And just now I did a pattern search and it's an older idea than I imagined, Rin Kaiho did it against Takemiya in 1977 and Fujisawa several times in the 90s!

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



Maybe :w1: has better followup contact plays than 'a'? It would be interesting to see some concrete sequences. How do you think of it?

Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #12 Posted: Tue Sep 04, 2018 8:58 am 
Honinbo

Posts: 8061
Liked others: 2280
Was liked: 2829
John Fairbairn wrote:
The position below is not a problem but part of a discussion from a book on how to "surround".

{snip}

Kimu then finishes off his explanation by giving the White move below.



LZ looks at five moves in the this area (as well as A) but once again they are all within a percentage point. So we can say the human pro has done acceptably well so far, and we can presumably follow his advice.

The tricky part comes next. Kimu discusses the White alternative below.



If I understand you correctly from what follows, this jump is also one of the plays that LZ liked.

Quote:
Kimu's reason for rejecting this move was that it is too thin. What I found interesting was that, hitherto, LZ had basically been showing about a single set of traffic lights as it pondered each move, but now it suddenly lit up like a funfair at night, with well over a dozen moves clamouring for attention.


These moves being Black's replies? IOW, Black had a number of good replies to the jump?

Quote:
That suggests to me some sort of criterion for computer thinking. The human pro generally sees "thinness = bad," the computer sees "thinness = fun!" Both are seeing the same manifestation but interpreting it differently.


Someone who understand LZ can correct me on this, but I don't think that the number of competing replies matters much, if at all, to its evaluation of a play.

Quote:
The specific problem with the thinness for Kimu is that it invites the good erasing sequence Black A, White B, Black C, and it is good for Black because even after it White still remains thin. And, for the umpteenth time, LZ accepts the human move as one of an elite bunch of moves each almost inseparable from the other, so Kimu is not really open to criticism.


Does LZ reply to Black A with White B?

BTW, I am attracted to this jump by White, not so much because it looks like a bot move, but because I have been playing over Dosaku's games. :)

Quote:
Now we come to the real reason I am posting this example. The erasing move, like the wacky centre moves already discussed and many other similar moves I have seen, involves "sector lines." This concept does not seem to get a mention much nowadays but I am going to suggest it may be a key to understanding AI play. The concept was not discovered by Wilcox (I have a much earlier book by Miyamoto with it in) but Wilcox did discuss it much more and I think we can safely go with his terminology.


I think that Wilcox discovered it independently. :)

Quote:
However, his understanding of the concept may possibly be questioned. His basic premise was that you play on a sector line to erase. Go any deeper and you get surrounded and cut off.


No, Wilcox said that if you go any deeper you are invading. Invasions do not necessarily get surrounded and cut off.

Quote:
Go any shallower and you are making things easy for the opponent.


I think that is Wicox's view, except that playing just above a sector line could be OK. I have seen cases where Go Seigen reduced without playing on or just above a sector line, but higher. :)

Quote:
I wonder whether this is yet another of the traps rationalists fall into - it's logical so it must be true.


As I understand it, Wilcox's rules were actually heuristics that he wrote into his go playing program. They are not logical, they are rules that made his program play better.

Quote:
The real truth is that logic all depends where you start from. In the case of sector lines, the bots seem to start from a different point.


It may well be that LZ and other bots play as though they were cognizant of sector lines, at least as far as reduction goes. In Uberdude's example above, Black's reducing play is one point within a sector line. An invasion, according to Wilcox. :)

Quote:
Here, although once more the LZ scores for all the possibilities cluster closely, it marginally prefers D or E, and it seems to me that sector line theory would advise against both.


D is less that one point above the sector line, and would not be disqualified as being too shallow. E would be considered an invasion and perhaps promoted as a peep. Whether Wilcox would vote against it is an open question.

Quote:
In the previous examples (and others I have seen) LZ likewise seems to go against our sector line theory and most typically goes one or maybe two steps deeper. What I am arguing for is that (in black box terms) LZ does appear, like us, to follow some sort of sector line theory in that it "sees" sector lines but handles them differently. Sector lines by definition are thin. We worry about that. For a bot that's fun.


Today's top bots' view of thickness and thinness is appreciably different from that of humans, but not out of reach, I think. But it will take us a while to refine the concepts.

Quote:
Balance actually has a lot in similar with thinness. The master of balance in go was Takagawa. Bill will remember this well. Mr T was highly successful with that style for a long time, but eventually Sakata, the master of disruption, toppled him. I have a feeling that amateurs tend to think of balance in terms of things settling down, or stability, but the real meaning should be achieving dynamic poise while still on the tightrope.

I believe that was what Kimu was striving for, but even he, as a pro, is not as nimble as LZ on the tightrope. Although we have allowed ourselves to be dazzled by AI which can do fancy tricks on the rope whereas we have so far only been able to achieve a slightly undignified arms-out wobble, I do recall that we felt exactly the same sort of wonder and astonishment when Olga Korbut came along and showed us somersaults and prodigious leaps on the high beam when before we had basically given high scores for just standing up straight with dignity on the beam. And then Nadia Comaneci came along and outdid even Olga with a perfect 10/10!

That seems to me an analogy for what we are seeing today. We humans may think we have achieved some sort of balance in our go. The bots are confounding that. For us to regain poise we need perhaps to make wackiness more mainstream. We have discussed thickness in relation to bots but I don't think we've focused on thinness. It seems thin is fun. Thin is the new black (and white)? Thin is where the somersaults are. The sector lines are our beams.


Except for saying flexible rather than thin, I pretty much agree. And we have had nimble players, such as Go Seigen and Dosaku.

_________________
There is one human race.
----------------------------------------------------

The Adkins Principle:

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

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #13 Posted: Tue Sep 04, 2018 9:17 am 
Honinbo

Posts: 8061
Liked others: 2280
Was liked: 2829
Uberdude wrote:
The examples I've picked so far have been invasion/reduction ones, probably because as a "Geronimo!" leap into the enemy area they are more dramatic and attract attention.


I think it is a good area of research, because I think that the bots have better judgement than we do about such positions, and their choices are not too alien.

Quote:
About sector lines, I recall reading/hearing somewhere that whilst a used to be a classic sector line breaking reduction idea,


Really? I missed that, unless someone was taking Wilcox too literally.

Quote:
some pros had come up with :w1 as being better because they looked at a load of plausible continuations after black's surround at b and they always find a good (but complicated!) way for white to manage the situation. And just now I did a pattern search and it's an older idea than I imagined, Rin Kaiho did it against Takemiya in 1977 and Fujisawa several times in the 90s!

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


Didn't Kajiwara make this invasion/reduction/approach?

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

_________________
There is one human race.
----------------------------------------------------

The Adkins Principle:

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

— Winona Adkins

Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #14 Posted: Tue Sep 04, 2018 2:15 pm 
Lives in sente

Posts: 1310
Liked others: 605
Was liked: 398
Rank: AGA 4k KGS 2k
GD Posts: 61
KGS: dfan
Bill Spight wrote:
Quote:
That suggests to me some sort of criterion for computer thinking. The human pro generally sees "thinness = bad," the computer sees "thinness = fun!" Both are seeing the same manifestation but interpreting it differently.

Someone who understand LZ can correct me on this, but I don't think that the number of competing replies matters much, if at all, to its evaluation of a play.

I think of the number of plausible replies as being somewhat (inversely) akin to temperature, although clearly it is not quite the same thing. If there are lots of replies with approximately the same win rate, then the position is less "sharp" and there is more room for personal taste, whereas if there's just one good reply then that local situation needs to be attended to right away, and moreover there's apparently only one good way to do so.

Of course, John said "fun" rather than "good", so I'm not sure any of us are actually at odds.


This post by dfan was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #15 Posted: Tue Sep 04, 2018 6:00 pm 
Oza
User avatar

Posts: 2096
Location: Tokyo, Japan
Liked others: 1918
Was liked: 1188
Rank: Jp 6 dan
KGS: ez4u
Uberdude wrote:
The examples I've picked so far have been invasion/reduction ones, probably because as a "Geronimo!" leap into the enemy area they are more dramatic and attract attention. But the bots also make nice surrounding moves, for example in that last example instead of white's dodgy peep Shuko suggested a surrounding move near tengen, whilst LZ had a similar idea but a bit more conservative (different judgement on current position / komi? or saw it was too easy to sabaki inside with Shuko's larger move?). With the black move I suppose it's aiming at some thinness of white's g15 knight move so can go a bit closer there, and the pro in the game did indeed attack/defend from j14 direction soon.

About sector lines, I recall reading/hearing somewhere that whilst a used to be a classic sector line breaking reduction idea, some pros had come up with :w1 as being better because they looked at a load of plausible continuations after black's surround at b and they always find a good (but complicated!) way for white to manage the situation. And just now I did a pattern search and it's an older idea than I imagined, Rin Kaiho did it against Takemiya in 1977 and Fujisawa several times in the 90s!

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


The basic theme of the large knight's play between two opponent stones on star points goes back at least to the Shin Fuseki era. Perhaps the first game really within the sector lines was this fun one from the 1947 Oteai. Fujisawa played mane-go up to :w24: and then invaded when Kitani played on tengen. Black ignored it to turn first to the left side. White won this game.
Click Here To Show Diagram Code
[go]$$Bcm23 Fujisawa Kuranosuke (white) - Kitani Minoru 1947-06-18
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O X O . . . . . . . . . . . X . . |
$$ | . . . O . . . . . 2 . . . . X X O . . |
$$ | . . . O . . . . . . . . . . . . X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . 5 . . . . . . . . . . . . . |
$$ | . . . O . 6 . . . 3 . . . . . X . . . |
$$ | . . . . . . . . . . . . 4 . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . X . . . |
$$ | . . X O O . . . . 1 . . . . . X . . . |
$$ | . . O . . . . . . . . . . . X O X . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


The other Fujisawa, Hideyuki's, use of the invasion as White against Takemiya did not work out so well (1-3 over 4 games). In addition, he earlier beat Sakata twice playing the Black side against Sakata's invasion into similar positions, the first time in game 2 of the 16th Oza final below.

Click Here To Show Diagram Code
[go]$$Wcm18 Sakata Eio (white) - Fujisawa Hideyuki 1968-10-24
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O X . . . . . . . . . . O . . . . |
$$ | . . X , . . . 4 . X . . . . . , O . . |
$$ | . . . X . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . 3 . 1 . 2 . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X , X . . 5 . X . . . . O , O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . 8 . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . 7 . . . . . . O . . . |
$$ | . . X , . . . . . O . . . . . , . . . |
$$ | . . . . X . . 6 . . . . . . . O . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/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: Wacky centre move training from bots.
Post #16 Posted: Wed Oct 31, 2018 12:00 am 
Judan

Posts: 5590
Location: Cambridge, UK
Liked others: 313
Was liked: 2951
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Another example from following along the live Ahan-Tongshan cup final between Gu Zihao and Fan Tingyu (yike english commentary at http://portal.yikeweiqi.com/online/goli ... om=8175754).

Click Here To Show Diagram Code
[go]$$Bc White (Gu Zihao) to play vs Fan Tingyu
$$ +---------------------------------------+
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . X . X X O . . . . . . . . . . . . . |
$$ | . X X O . O . . . . . . . X . . . . . |
$$ | . O X O . . . . . , . . . . . X . . . |
$$ | . O O O . . . . . . . . . . . . . . . |
$$ | . X . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . X . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . . O . X O . . . . . . . . . . X . . |
$$ | . . . O O O . O . . . . . . . . . . . |
$$ | . . . O X X . O X , . . . . . X . . . |
$$ | . . O X . X . O . O . . X . . . . . . |
$$ | . . O X X . X . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


Elf quickly homes in on what it thinks is the vital point of the board in this middlegame postion. Interestingly FineArt giving commentary on Fox has very similar opinion (one point to the right) which Elf gave a few playouts to, but when played agrees is <0.5% better.
Attachment:
ahan elf centre2.PNG
ahan elf centre2.PNG [ 512.2 KiB | Viewed 1262 times ]


In the game white shoulder-hit top right, Elf still says key point for black now. FineArt agrees exactly.
Click Here To Show Diagram Code
[go]$$Wc Game to 3, Elf/FineArt's 4.
$$ +---------------------------------------+
$$ | . . X . . . . . . . . . . . . . . . . |
$$ | . X . X X O . . . . . . . . . . . . . |
$$ | . X X O a O . . . . . . 2 X . . . . . |
$$ | . O X O c b . . . , 3 . 1 . . X . . . |
$$ | . O O O d e . . . . . . . . . . . . . |
$$ | . X . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O . . 4 . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . X . . . . . . . . . . . |
$$ | . . . . . X . . . . . . . . . . . . . |
$$ | . . O . X O . . . . . . . . . . X . . |
$$ | . . . O O O . O . . . . . . . . . . . |
$$ | . . . O X X . O X , . . . . . X . . . |
$$ | . . O X . X . O . O . . X . . . . . . |
$$ | . . O X X . X . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

As well as growing/reducing the white moyo on the left and making/breaing shape for black's (now) reduction stones, this point ends up in a perfect spot to counter the severity of black's pushing and cutting with a-e, which is what happened in the game. This is so important Elf thinks black should have ignored the shoulder hit to push there before white has a chance to defend (and white will ignore), and that white should have played f11 when he had the chance instead of l16 jump.


This post by Uberdude was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #17 Posted: Wed Oct 31, 2018 2:56 am 
Lives in sente

Posts: 1086
Location: Earth
Liked others: 344
Was liked: 180
When I review games were a "wacky centre move" was the best option, I encounter this problem:

Many times I am aware of the correct fuzzy target area, but not of the exact best play in this target area.

When I review the variations that show the move suggested by AI is indeed the best option in the local area, these tactical variations are most of the time above my reading depth.

I find it difficult to learn from these complicated variations that are simply above my abilities :)

I start to perceive some tactical combinations, that I was not aware of in the past. But in my games it is most of the time sheer luck if the combinations I try are adequate to the position.

Back to tsume go, were I know that I dont know.


This post by Gomoto was liked by: ez4u
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #18 Posted: Wed Oct 31, 2018 4:21 am 
Judan

Posts: 5590
Location: Cambridge, UK
Liked others: 313
Was liked: 2951
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Yes, I don't claim studying these examples is the most efficient way to improve, but I find them more interesting than drilling tsumego :) . Just staring at the position with the suggested move should train my neural network a bit and broaden my horizons. Also this one is not so wacky, on reflection it's quite a traditional "about middle of a sector line from around f17 to h8".


This post by Uberdude was liked by: Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #19 Posted: Wed Oct 31, 2018 8:24 am 
Honinbo

Posts: 8061
Liked others: 2280
Was liked: 2829
Gomoto wrote:
When I review games were a "wacky centre move" was the best option, I encounter this problem:

Many times I am aware of the correct fuzzy target area, but not of the exact best play in this target area.


As a rule, when things are fuzzy, playing close to the best point loses little.

Quote:
When I review the variations that show the move suggested by AI is indeed the best option in the local area, these tactical variations are most of the time above my reading depth.

I find it difficult to learn from these complicated variations that are simply above my abilities :)


Poco a poco. :)

Quote:
I start to perceive some tactical combinations, that I was not aware of in the past. But in my games it is most of the time sheer luck if the combinations I try are adequate to the position.


As against that, when your opponent is at the same level as you, that holds for him as well. :blackeye:

Edit: In this pro game White missed the target area altogether, and Black missed the best play. :)

_________________
There is one human race.
----------------------------------------------------

The Adkins Principle:

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

— Winona Adkins


This post by Bill Spight was liked by: Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: Wacky centre move training from bots.
Post #20 Posted: Wed Oct 31, 2018 8:57 am 
Honinbo

Posts: 8061
Liked others: 2280
Was liked: 2829
Uberdude wrote:
Another example from following along the live Ahan-Tongshan cup final between Gu Zihao and Fan Tingyu (yike english commentary at http://portal.yikeweiqi.com/online/goli ... om=8175754).


Thanks. :) A droll commentary. ;)

_________________
There is one human race.
----------------------------------------------------

The Adkins Principle:

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

— Winona Adkins

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 21 posts ]  Go to page 1, 2  Next

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