Life In 19x19

This thread should be locked.

I don't see why it should be locked. It didn't break forum rules. Javaness made a popular non-admin request to stop rehashing the game 4 issue which has been heeded. People might want to discuss the Ali vs Eric game, though it seems not such a popular topic.

The next game is this Sunday: Eric Lui vs Pavol Lisy. They recently played at the IMSA event and Pavol won, though I didn't see a game record.

TheCannyOnion wrote:This thread should be locked.

IMO people's freedom should only be restricted with very good reasons. Even with very good reasons, it probably shouldn't.

Is the next game today? I can't see the confirmation on facebook.

TRANSATLANTIC PROFESSIONAL GO TEAM CHAMPIONSHIP
Game #6
TODAY, Sunday, June 9, 20:00 CEST (14:00 EDT)
Pavol Lisy 2p (EGF) vs. Eric Lui 1p (AGA)

Watch live on twitch: twitch.tv/usgoweb

https://transatlanticgo.org/

<edit>
BTW, could anybody point me to some post where the “L19 titles” are explained? Like “Dies in Gote”, or, what I now have: “Oza” … (I’m sure I’m actually “Loses won Game” or something )
</edit>

@Bonobo: viewtopic.php?p=226469#p226469

The game on OGS: https://online-go.com/game/18205527

P.S. Eric Lui won by 1.5 point.

This game (like B 217 & 219) is a good reminder not to take the game too seriously.

Even LZ goes weird reviewing them, elf seems more consistent.

Thanks for both, @jlt, I totally forgot to add the game link.

The game, lots of ups and downs, but Eric (white) won because he made the penultimate big mistake.



The story according to LZ 226 on 3300 playouts.

Around the mentioned moves B - with an overwhelming lead - connects several scattered stones and offers them up in bulk in an avoidable ko. This drops his chances from 97% to below 30% (which LZ interestingly only sees when all moves are played out, in its search it evaluates individual moves >90% - even the worst move creating the ko, though oc it prefers the unconditional connection).

Interestingly LZ also confirms my suspicion: the whole sequence is worse than passing at 215 (which in itself was fine). Passing retains 35% from 97% (W can only take smaller scales).

I want to try analysis of the game with more playouts. I find in my own games that analysis on the order of 25-50k playouts finds very different sequences due to it's more accurate reading. For example, I've noticed it can calculate long capturing races that are missed with lower playout counts. I think it also substantially improved ladder reading in most instances, which can alter the variations significantly.

Edit: I ran the first 250 moves through GoReviewPartner at 25,600 playouts per move with LZ#226. Results are similar. I am attaching the rsgf file, which may be opened with the software to see the top variations that LeelaZero recorded.

The screenshot shows the value network winrate in grey with black's winrate delta in red. The red basically shows how high black's winrate would have been, had they chosen LZ's optimal move.

Artem Kachanovskyi vs. Eric Lui:

https://online-go.com/game/18494065
https://www.twitch.tv/usgoweb

Artem won by resign, a pretty close game throughout with some to and fros in the endgame but then Artem just lived in Eric's corner. So that's 4-3 to EGF, but Artem is busy next weekend so Ilya will play instead vs Ryan Li, so if Ilya wins Europe wins the event, if Ryan wins he plays Artem the week after.

All throughout the game I kept thinking "white is in trouble". Shows how much I know
If Ilya plays next week, then all 10 players played at least a game regardless of the final result.