Following Iyama Yuta (no world ranking discussions)

[johnsmith]

For all the strong players, I analyzed the game with LZ #208 with 3200 visits and fast timemanage. So, below you can find:

- software that I used - Go review partner 0.10. https://uploadfiles.io/8z4rq I emptied LZ folder for smaller zip.
- output from the software above -> game in rsgf format, can be viewed with the same software. I mixed that output with dozen of very precise Zen7 territorial estimates on key moments of the game (special technique used, not weak score estimate button). https://uploadfiles.io/z3zm8

I didn't put my comments in there, just LZ and Zen, so that every player can draw conclusions on their own. We can see that the game was mosly close all the time. White lost almost 10 points with move 22.

[sorin]

Here's my analysis with LZ 207 about why block other side for 199 was better. Conclusion, it is: having the stone at g13 comes in very useful 40 moves later when black wants to counterattack against white's middle group in a semeai he can then win by 1 liberty, or creating the shortage of liberties in the e10 bamboo combo-ing with e11. There are some lines where white does kill the left group but that involves ignoring a move in the middle so black captures a bunch of stones there in sente and wins anyway. The intuition of these bots to see block above better in just hundreds of playouts (<1 second) when it depends on ~40 move sequences of delicate use of aji is impressive. I expect Iyama could do it too if he had enough time and concentration though.

That sequence you found is absolutely amazing!

I wonder how much of the game did they play in byo-yomi. Yamashita was already in byo-yomi at move 196, when he is obviously wasting a ko-threat just before peeping at 198.

As for me, I feel that the most important learning from this game is the first line tesuji that Yamashita played at move 114: R19, making miai O18 and S16, just when I was wondering what is wrong with him to commit suicide with that invasion
I guess I would never have found R19 myself (unless if this was presented to me as a tsume-go).

[Bill Spight]

Here's my analysis with LZ 207 about why block other side for 199 was better. Conclusion, it is: having the stone at g13 comes in very useful 40 moves later when black wants to counterattack against white's middle group in a semeai he can then win by 1 liberty, or creating the shortage of liberties in the e10 bamboo combo-ing with e11. There are some lines where white does kill the left group but that involves ignoring a move in the middle so black captures a bunch of stones there in sente and wins anyway. The intuition of these bots to see block above better in just hundreds of playouts (<1 second) when it depends on ~40 move sequences of delicate use of aji is impressive. I expect Iyama could do it too if he had enough time and concentration though.

Am I right that, although you include variations, you have only one main line for the upper block and none for the lower block (having only the pro vs. pro play for comparison)? In the interest of go science, how about generating at least another main line for the upper block? My guess is that it will be different. (Even the much weaker Leela 11 prefers the upper block, and its preference does not rely upon accurate reading 40 moves deep. )

Oh, yes. By main line I mean that each move is the bot's first choice.

[Uberdude]

Here's some more variations with the questions I am getting Lizzie to answer made explicit in comments.

[Bill Spight]

Out of curiosity I entered 3 main line variations for Leela 11 (Deep Leela) for each of the blocks, and one variation for the atari at H-10, when it got promoted. It looks like Deep Leela learns from entering variations, as the upper block increased its lead in win rate to 12% over the lower block. None of these variations, however, are like the variations of the stronger bots after the Black extension to E-18. With the upper block after White plays the nose attachment, Black replies with the clamp instead of going for the approach ko (if it even sees it). With the lower block White does not even play the nose attachment. It is not clear that those White and Black plays are wrong.

For entertainment value, check out the roller coaster of the first Deep Leela variation with the upper block. (Variation 2 at moves 199 and 200, main line after that.) If Deep Leela only understood the double ko that arises towards the end.

[Uberdude]

Game 7 so far, Iyama is white. Yamashita fell behind with the lower-left sequence; at least the b5 atari wasn't wrong like last time, in fact now LZ and Elf agree Iyama's c3 atari was a mistake because it helps black to resolve the ko and instead he should just capture at g5 and leave black with the ko headache (white has lots of threats on right side). But then Iyama misplayed the subsequent defence on the right side and Yamashita is getting back into the game. p12 natural shape pokey out being a mistake is interesting: it appears the bots would prefer to set up the ko life on the side and then answer in the top left corner. This is not a tenuki because by taking the eyespace there the surrounding black group is no longer alive and even if the right side group dies black can get indigestion: with p13 hane white cuts in sente and the upper right corner black group has to run for life, and with the lower left black group not alive either they see white with a large advantage (as a bonus if white gets p2 later black will likely have to spend a move at q10 eventually).

Also o6 being a mistake and better as the q6 nose tesuji is a good shape to know these days with that shape often arising from the approach then attachment inside big high shimari.


Game is to 96, Iyama has just played grappling hook tesuji at n14. LZ suggested variation for 96 is fun, check out tesuji of 120 to get sente seal, and 126 wedge milking every last drop of value out of the dead group. Nevertheless, with that variation black has a lot of solid points he can't mess up whereas white's lead is dependent on converting a lot of that open upper left moyo to cash and humans can easily mess that up.

[Bill Spight]

Game 7 so far, Iyama is white. Yamashita fell behind with the lower-left sequence; at least the b5 atari wasn't wrong like last time, in fact now LZ and Elf agree Iyama's c3 atari was a mistake because it helps black to resolve the ko and instead he should just capture at g5 and leave black with the ko headache (white has lots of threats on right side).

I suspect that the point is more the value of the ponnuki than the ko. If White has the advantage in ko threats it makes sense to increase the value of the ko moves, which is what the atari does. White does not want to win the ko by filling it.

Edit: Looking two moves ahead I see that Black won the ko. Than being the case it is questionable for White to increase the value of the ko moves, since then Black gains more by taking and winning it.

On a related note, humans typically use the difference between consecutive win rate estimates as a measure of the the value of a play. A more appropriate measure is the difference between the win rate estimates of alternative moves, as long as each move has enough playouts to yield a reliable estimate. (The ponnuki has 25k playouts while the atari has 3.4k playouts.) By that token the atari loses an estimated 15% instead of 10%, because the ponnuki is such a good play that it gains 5%. (Whether that makes sense is another question. )

[Bill Spight]

The proposed eye shape move, S-11, for is reminiscent of a play by Sakata, who was known for his skill at shinogi.

[Uberdude]

It looks like Iyama won by 6.5 in the end, thus defending his Kisei title. Yamashita managed to hang onto all of his cut bits surrounding the dead group and keep it dead, but with that yoke around his neck Iyama turned enough of the rest of the board into cash.

[Ember]

That was SUCH an incredible game, I'm still spent just from watching. I can't imagine how exhausting it must've been for the players and I have the highest of respect for Iyama who played most of the 2nd half in byoyomi and still was able to turn the tables.

And since I'm already writing... MANY THANKS to you, Uberdude, for your continued coverage on Iyama's games! I really do appreciate and enjyo it a lot (even if I usually only read without logging in).

[Uberdude]

[Uberdude]

Meanwhile, in the Meijin league to decide Cho U's challenger Iyama is not doing well: he's still 1-2 (beat Mutsuura Yuta, lost to Shibano and Suzuki Shinji) whilst Kono Rin is storming ahead 4-0 (Shibano, Murakawa, Shinji, M Yuta) and Shibano is 3-1 (I Yuta, Murukawa, Son Makoto, lost to Kono). If it's not going to be Iyama I hope Shibano wins as Kono Rin vs Cho U is pretty retro. Iyama also faces Yamashita again in the 3rd round of the (knockout) Gosei tournament to decide who gets to challenge Kyo Kagen.

[Uberdude]

Judan game 2 against Murukawa today. Fighting seemed fairly even until Iyama (black) allowed Murukawa to cut off the left group with 62 following obvious resistance to 59 attach, can this really be good? LZ says 'no way jose', -35% (Elf v2 -50%). Game is to 122 with Iyama recovering somewhat, I put in lots of variations exploring why p15 atari is a bad exchange: I guess Iyama thought it was kikashi before defending with push on the 2 stones because later white might not connect and it sets up some liberty problems if he gets o16 hane later, but LZ sees it as slight minus, but only becomes apparent with some very interesting variations later of black trying to live in the top side after o17 jump and black filling in his own liberties changes the senteness of white threating to go into the top right corner, which in turn changes the life and death problem on the top! Also the tesuji combo of j17 g15 before that is very nice, I expect Iyama will find it if those lines become relevant. Take 2nd branch at 118, 2nd branch at 126, 2nd branch at 143, and compare to 2nd branch at 112 (LZ recommended order).

Update: Iyama resigned at 148. Seemed a bit generous to me as he's won games from worse positions but I suppose there's not so much scope for Murukawa to mess up. Still, I wouldn't resign until the white right group is clearly settled and both the lower left and middle left are aji-free white territory. (e.g. my plan to give black an opportunity to lose would be try to make ko on left side, lose that ko and get 2 moves on right and try to kill / bully and profit as last hurrah).



Edit: fixed colours

[Shenoute]

Thanks Uberdude!
I think you got the colours reversed in your post though

[hyperpape]

Meanwhile, in the Meijin league to decide Cho U's challenger Iyama is not doing well: he's still 1-2 (beat Mutsuura Yuta, lost to Shibano and Suzuki Shinji) whilst Kono Rin is storming ahead 4-0 (Shibano, Murakawa, Shinji, M Yuta) and Shibano is 3-1 (I Yuta, Murukawa, Son Makoto, lost to Kono). If it's not going to be Iyama I hope Shibano wins as Kono Rin vs Cho U is pretty retro. Iyama also faces Yamashita again in the 3rd round of the (knockout) Gosei tournament to decide who gets to challenge Kyo Kagen.

Also retro: Hane Naoki leads with 5-1 in the Honinbo league. Shibano is at 4-2 and higher in the league, so I believe he can still challenge if he beats Ko Iso and Hane loses to Yamashita Keigo.