Japanese title match streaming

[gowan]

I just learned that the Meijin match is being streamed live by the Nihon Ki-in and it is available on Youtube : https://www.youtube.com/watch?v=2ehEr5COmwU

This link is eight hours plus long! You can see the real behavior the players, the game recorder, and the officials attending at the beginning of the game, at least. This is game one of the match between Cho U Meijin and Shibano Toramaru challenger.

edit: There is a running AI assessment of the position. By around move 30 the AI estimate was Black (Cho) 55%, White (Shibano) 44%.

[hyperpape]

I'm sure there is a better way to find them, but I found out about the streams because they're usually posted on online-go.com. There's often someone posting a relay, in addition to linking to the youtube stream.

[gowan]

At the end of day one the AI estimation was 76% for Shibano and 24% for Cho. Quite a swing during the game. I don't know which AI was used.

[hyperpape]

Depends on the AI, but that might be as little as a four point swing.

[Bill Spight]

Did they publish the number of playouts with those estimates? It matters.

[jlt]

See also this post:
viewtopic.php?p=248115#p248115

LeelaZero 157 says that White's next move should be G4, with 81% winrate (given 17k playouts).

Capture.PNG (685.06 KiB) Viewed 7439 times

FWIW, Leela 0.11 with 1 million simulations agrees that G4 is the best move, and says that White leads by 3 points (but leela 0.11 is probably too weak to evaluate a Meijin title match).

[Zenit]

See also this post:
viewtopic.php?p=248115#p248115

LeelaZero 157 says that White's next move should be G4, with 81% winrate (given 17k playouts).

Capture.PNG

FWIW, Leela 0.11 with 1 million simulations agrees that G4 is the best move, and says that White leads by 3 points (but leela 0.11 is probably too weak to evaluate a Meijin title match).

Meanwhile, the AI (FineArt, JueYi) on fox seem to think white should cut at K7, which looks like Leela's second choice. After a sequence where black makes some shape for his stones in the middle and white gets to capture L7 in sente, white sacrifices his H3 stone by attaching with H5. This sequence gives white a solid 91.6% lead. But I think FineArt's estimations tend to fluctuate more than other AI's. But it looks nice for Shibano for now in any case.

[Bill Spight]

See also this post:
viewtopic.php?p=248115#p248115

LeelaZero 157 says that White's next move should be G4, with 81% winrate (given 17k playouts).

Capture.PNG

FWIW, Leela 0.11 with 1 million simulations agrees that G4 is the best move, and says that White leads by 3 points (but leela 0.11 is probably too weak to evaluate a Meijin title match).

Meanwhile, the AI (FineArt, JueYi) on fox seem to think white should cut at K7, which looks like Leela's second choice. After a sequence where black makes some shape for his stones in the middle and white gets to capture L7 in sente, white sacrifices his H3 stone by attaching with H5. This sequence gives white a solid 91.6% lead. But I think FineArt's estimations tend to fluctuate more than other AI's. But it looks nice for Shibano for now in any case.

Well, LZ's second choice has only 5.4k playouts, and besides, its winrate estimate is only 1% lower than that for its first choice. How do FineArt's playouts and estimates compare?

As for FineArt's estimates fluctuating more than other AI's, perhaps that is an indication that it is stronger. After all, the fluctuations for a perfect player will be 100%.

[hyperpape]

As for FineArt's estimates fluctuating more than other AI's, perhaps that is an indication that it is stronger. After all, the fluctuations for a perfect player will be 100%.

I used to assume this, but isn't ELF consistently much more judgmental than Leela Zero, despite not being much (or is it any?) stronger?