Following Nakamura Sumire

[Shenoute]

Can someone explain what happened at the end of the first game?

The sgf says B won on time, with the last move being S18, but on the video it seems that white answered rather quickly at T18. No countdown is even heard so white"s move must have been made on time. Nakamura Sumire then plays one more move and then the game apparently stops immediately...

2:40:33 B plays S18
2:40:43 W plays T18
2:41:02 B takes at T1
the game ends at once

Video starting at 2:40:28

[John Fairbairn]

Hane forgot to press her clock. Sumire played her move anyway to be on the safe side but then signalled that she claimed the game on time. The referee Goto had to explain to startled Ayaka what happened then she accepted it.

Sumire did try to drop a hint by flashing her eyes to the clock.

[Shenoute]

Ah, thanks! I completely missed that. I had noticed Sumire looked a bit uneasy before playing T1 but couldn't understand why. Too bad for her opponent, must be quite painful to lose a game like that.

[John Fairbairn]

I might usefully have mentioned that Sumire had a little earlier beaten Ayaka in an exhibition game in the Osaka version of the Hankyu Cool of the Evening go festival. Two things about that struck me as specially interesting. One was that the exhibition attracted over 1,000 fans.

The other was that Sumire made what is still a new move in a very common AI "direct 3-3" joseki. It was introduced at the end of August in Korea and a few Chinese players have taken it up, but it's still rare yet Sumre knew it. Kobayashi Satoru asked her about it and she confirmed she had been studying it. (It's move 24 if you want to look up the game, and the move that won the game, 126, impressed Murakawa Daisuke.)

Just think about that for a moment. A brand-new pro, just 10 years old, is already studying rarities at the cutting edge of go theory. I'd guess most people here would struggle to give the value of a simple thing like a hanetsugi boundary play on the third line, and there's little Violet already going boldly on beyond where no AI has gone before.

For me the name Violet is indelibly associated with Violet Elizabeth Bott, who would try to get her own way by stamping her childish foot at William and saying, "I'll thcream and thcream till I'm thick." She sounds like she would have made a good go player, but in reality was quite obnoxious. The Japanese Violet, however, makes me thcream, but it's a thcream of admiration, if you can imagine that.

[Uberdude]

Here's the game John mentions. http://www.igo-kifu.com/kifu.php?id=3709. Some pics at https://twitter.com/mainichi_igo/status ... 8499900416

[SoDesuNe]

Here's the game John mentions. http://www.igo-kifu.com/kifu.php?id=3709. Some pics at https://twitter.com/mainichi_igo/status ... 8499900416

What impresses me lowly kyu-player are those beautiful shapes in the running fight. I don't particulary understand especially after locally answering since the cut ( ) looks to my lowly kyu-player eyes overplay-ish.

I also admire here patience with not trying to "attack" (for lack of a better, ie japanese, word ; )) Black's floating center group.

And saving after is also cool. But I'm taken easily with such sequences since my own reading is leading nowhere most of the times.

Time to go ask Lizzie.

[SoDesuNe]

Sorry for the double.

is indeed not a good move. The cut ( ) is apparently, but should be played sooner.

Win graph according to Lizzie (15k playouts)

Analysis and variations (albeit quite a few wonky ones for my taste ^^) of GoReviewPartner, latest LZ weighs, 30 seconds per move (I marked every 10%-wingraph change in the SGF):

[sorin]

Sumire won yesterday against Yamamoto Kentaro 5p (male) in Honinbo group C.
This was a long game (3h time limits), and she attacked a lot, I don't have the game record.

[Uberdude]

[SoDesuNe]

Unless otherwise stated all variations were at least looked at with 10k playouts.

With 50k playouts White's winrate after is 83% according to Lizzie (actually White had a 80% winrate at move 38 already). But my heart sunk when I saw this atari.

$$Bcm79
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . O . O O O X . . O X X . . . |
$$ | . . . O . O X O X , X . . O O X . . . |
$$ | . . . . . O X . X . X . X . . X . X . |
$$ | . O O . X X X . X . . . . . O . X O . |
$$ | . . X . . X O . O X X O O O . X . O . |
$$ | . . . . X X O . O O O X X O O . O . . |
$$ | . . X X O O . . . . . . . X O . . . . |
$$ | . . O O 1 . a . . , . . X . . , . . . |
$$ | . . . . 2 . . . . . . . X X O O O O . |
$$ | . . . . . . . . . . . . . O X . X X . |
$$ | . . . . . . . . . . . . . . X . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , X . . |
$$ | . . . . . . . . . . . . . X . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bcm79
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . O . O O O X . . O X X . . . |
$$ | . . . O . O X O X , X . . O O X . . . |
$$ | . . . . . O X . X . X . X . . X . X . |
$$ | . O O . X X X . X . . . . . O . X O . |
$$ | . . X . . X O . O X X O O O . X . O . |
$$ | . . . . X X O . O O O X X O O . O . . |
$$ | . . X X O O . . . . . . . X O . . . . |
$$ | . . O O 1 . a . . , . . X . . , . . . |
$$ | . . . . 2 . . . . . . . X X O O O O . |
$$ | . . . . . . . . . . . . . O X . X X . |
$$ | . . . . . . . . . . . . . . X . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , X . . |
$$ | . . . . . . . . . . . . . X . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

My choice - due to concerns about my center group - would have been 'a' and Lizzie agrees but nonetheless the atari just loses 4%. (Btw, Black's cut was Lizzie's top choice by far.)

Until move they followed the line Lizzie would have played too.

$$Bcm79
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . O . O O O X . . O X X . . . |
$$ | . . . O . O X O X , X . . O O X . . . |
$$ | . . . . . O X . X . X . X . . X . X . |
$$ | . O O . X X X . X . . . . . O . X O . |
$$ | . . X . . X O . O X X O O O . X . O . |
$$ | . . . . X X O . O O O X X O O . O . . |
$$ | . . X X O O 4 . . . . b . X O . . . . |
$$ | . . O O 1 3 . . . d . . X . . , . . . |
$$ | . . . . 2 5 7 . 9 c . . X X O O O O . |
$$ | . . . . 8 6 a . . . . . . O X . X X . |
$$ | . . . . . . . . . . . . . . X . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , X . . |
$$ | . . . . . . . . . . . . . X . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------

Click Here To Show Diagram Code

[go]$$Bcm79
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . O . O O O X . . O X X . . . |
$$ | . . . O . O X O X , X . . O O X . . . |
$$ | . . . . . O X . X . X . X . . X . X . |
$$ | . O O . X X X . X . . . . . O . X O . |
$$ | . . X . . X O . O X X O O O . X . O . |
$$ | . . . . X X O . O O O X X O O . O . . |
$$ | . . X X O O 4 . . . . b . X O . . . . |
$$ | . . O O 1 3 . . . d . . X . . , . . . |
$$ | . . . . 2 5 7 . 9 c . . X X O O O O . |
$$ | . . . . 8 6 a . . . . . . O X . X X . |
$$ | . . . . . . . . . . . . . . X . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . , X . . |
$$ | . . . . . . . . . . . . . X . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

loses a whooping 11%! Lizzie's choice would have been 'a'.

But I - as White - would have been very, very scared after ... Lizzie agrees with White's handling of the attack but would have begun with forcing at 'b' before 'c'. I guess I could spend the rest of the afternoon figuring out, why this attachment "saves" White's group...

After Black answering 'c' at 'd', Lizzie gives White a winrate over 90%, so good job Sumire for judging/reading, that the center group was in no danger : )

Some general comments on imo interesting parts:

Of course no split!

Lizzie overwhelmingly prefers to attach below the hoshi right away. Not in terms of winrate but in terms of playouts: 75k to 2k, which would be attaching left of the bottom right komuku.

I thought this is too early but Lizzie (albeit prefering approaching the bottom left - just approaching! - and then pincer at the top) thinks this is okay winrate-wise (~1%).

Lizzie doesn't really consider this move very much with just 14 playouts to 20k, which would be attaching left of the bottom right komuku - just probing mind you! ( loses ~2%)

Lizzie prefers the shoulder-hit at M14 but that's minor winrate-wise (~-1.5%).

Gets some smack, though! (~-6%) Lizzie prefers the not-anything-strengthen-kosumi at K14.

Playouts-wise Lizzie prefers the mentioned attachment left of the bottom right komuku again (by a lot, just gets 450 playouts. The attachment gets over 20k). Again just probing and then switching to the hane in the upper right at P18 (P18-Q18-N18-M18-P11). But winrate-wise it's not even ~1%.

Lizzie thinks this should be played in the opposite direction, that is extending at H15. With over 10k playouts Lizzie would also abandon this stone to play H15. (~-4.7%) But just extending first to H15 and then poking at N14 to show some kiai!

This seems too thin for Lizzie's taste? (~-8%) It prefers the attachment at G16.

also gets some heat from Lizzie (~-6%). Direction-wise it prefers the right side, either at the top (N18) or at P11. Although even after N18, Lizzie would play P11 right after that. I guess this covers the cut at L12?

But gives the advantage right back to White (-~8%). Fighting spirit was called for: L13 and cut!

...so from to Black lost around 20% winrate according to Lizzie. The main reason being (I guess), Black did not really fight back.

[Tryss]

Here is Katago winrate (green) & score (purple) graph (with low playouts) :

Sumire Katago.png (83.81 KiB) Viewed 22449 times

[dshin]

I guess I could spend the rest of the afternoon figuring out, why this attachment "saves" White's group...

Here is a review of the game that goes over that sequence: https://www.youtube.com/watch?v=hgkxYbL6YbQ

Video shortcuts:

8:52 black E10 cut
9:44 white K9 jump

It is in Korean, here are a few commentary highlights:

9:49 The jump works because of the aji of the white stone at O8
10:13 This sequence leads to a seki with great outside influence for white

I think the rest you can follow along visually without the audio.

[John Fairbairn]

Yamamoto's record last year was 4-10 and so far this year is a respectable 4-7, and of course he's 5-dan (and young - 39). I hope I hear the sound of onions sizzling.

[Bill Spight]

SoDesuNe, thank you for this excellent review!

I thought this is too early but Lizzie (albeit prefering approaching the bottom left - just approaching! - and then pincer at the top) thinks this is okay winrate-wise (~1%).

Lizzie doesn't really consider this move very much with just 14 playouts to 20k, which would be attaching left of the bottom right komuku - just probing mind you! ( loses ~2%)

14 playouts is not enough to get a reliable winrate estimate. Do what the Elf team did in their commentaries when a game play got fewer than 500 playouts. Inherit the winrate estimate from Lizzie's best reply to . That may also apply to .

Edit:

also gets some heat from Lizzie (~-6%). Direction-wise it prefers the right side, either at the top (N18) or at P11. Although even after N18, Lizzie would play P11 right after that. I guess this covers the cut at L12?

P-11 not only bolsters White's central group, it reduces Black's right side.

[Bill Spight]

Here is Katago winrate (green) & score (purple) graph (with low playouts) :

Sumire Katago.png

Very interesting!

For the first half of the game it seems like there is a close approximation between winrate percentages and points. What is the average winrate difference per point difference?

Just eyeballing it, it looks roughly like 1 pt. = 2½%. This would be in line with statistics on professional play in the 1970s, where retrospectively altering komi by 1 pt. would make a difference in wins of between 2% and 3%.