Life In 19x19

First round:

Lu Minquan vs. Yang Zixuan
Yu Zhiying vs. Oh Yoojin
Zhou Hongyu vs. Ueno Asami
Wang Chengxing vs. Cho Seungah
Li He vs. Hei Jiajia
Fujisawa Rina vs. Stephanie Yin
Ariane Ougier vs. Choi Jeong
Rui Naiwei vs. Xie Yimin

Rui Naiwei missed last year's tournament, so interested to see how she does.

First round results:

Lu Minquan vs. Yang Zixuan
Yu Zhiying vs. Oh Yoojin
Zhou Hongyu vs. Ueno Asami
Wang Chengxing vs. Cho Seungah
Li He vs. Hei Jiajia
Fujisawa Rina vs. Stephanie Yin
Ariane Ougier vs. Choi Jeong
Rui Naiwei vs. Xie Yimin

All 6 Chinese players advanced.

Second Round / Quarterfinal:

Fujisawa Rina vs. Lu Minquan
Choi Jeong vs. Yu Zhiying
Zhou Hongyu vs. Wang Chengxing
Li He vs. Rui Naiwei

Results:

Fujisawa Rina vs. Lu Minquan
Choi Jeong vs. Yu Zhiying
Zhou Hongyu vs. Wang Chengxing
Li He vs. Rui Naiwei

Good game between Choi and Yu. Choi's end game was stronger again to pull it out. Yu now has lost 5 in a row against Choi this year.

Semifinals:

Choi Jeong vs. Rui Naiwei
Fujisawa Rina vs. Zhou Hongyu

Looks like Choi's better endgame will clinch it vs Rui too... (f15 big mistake)

Here is the game. LeelaZero (15-blocks trained on 40 blocks) doesn't see move 83: Black's winrate is 41% before the move, and 68% after.

jlt wrote:Here is the game. LeelaZero (15-blocks trained on 40 blocks) doesn't see move 83: Black's winrate is 41% before the move, and 68% after.

Katago don't see fast, but if you let it run 50k playouts, it start to visit it, and after KG has visited this move 4 times (basically, reading the R13, S14 P13 Q13 sequence), it's immediatly becoming the blue move

The reason Katago can see it is probably that LeelaZero doesn't see that G6 is a ladder breaker, while Katago has been trained to read ladders.

Interesting problem. Perhaps more suited for Computer Go than here, but I'll leave it for continuity. (mods, have at it!)

I loaded the game into the current LZ 40 block weight. It tried R13 13 times in 50k layouts, but only had a 41.1 win percentage with it. I let it run to 125k, and it never looked again. I buy jlt's ladder theory. It's actually amazing that LZ is as strong as it is.

On my first run, Katago found R13 in 2000 playouts and chose it (light blue in Lizzie) in less than 3000. Lest you think I am overly proud of my KataGo, on it its second try it only checked R13 twice in its first 50k playouts, and was obsessed with black P13 instead. It finally found R13 in 86k playouts and chose it at 99k. Then I ran it a third time and it found and chose it within 250 playouts. Bots are weird.

I guess currently, the best way to review a game is to use both LZ and KataGo, the latter for ladder influences and endgame, the former for everything else.

Anyway, cool game! Sad for Rui Naiwei, who will always be one of my heroes.

For the 3rd year in a row, Choi won the tournament.