I have been enjoying the 44th Meijin title games between Cho U and Shibano Toramaru. Unfortunately, because these games sometimes span days of play and are broadcast “live” in real-time on YouTube, I miss parts of them, or I never find out who won, etc.
Up until today, I have had a really difficult time finding game records and results, but I finally found all the games here:
I have saved the SGF files from the first 3 games to my SGF library on OGS, and you are welcome to view them here if you are interested in replaying them. I have enabled a full AI review so you can see where each game turned for B or W, etc.
Game 1: Aug 27-28
Game 2: Sept 10-11
Game 3: Sept 17-18
I’ll post the results of Game 4 as soon as it is available.
Ok - the 4th game from the 44th Meijin is now up. Shibano Toramaru is on a roll!
Side note: if you look at the AI analysis of games 1-3, those games didn’t tip towards one player over the other until later in the game.
Whereas, if the AI analysis is to be believed, Toramaru took the lead in Game 4 somewhere around move 51, and managed to hold on to that lead until the end. Impressive!
Thanks a lot for both sharing the games and the analysis.
I usually don’t look for pro games because I hardly understand a small part of them, but I‘m very curious about this fight between Cho U and Shibano Toramaru.
OK - here is the SGF of Game 5 of the 44th Meijin. Shibano Toramaru wins as White by resignation - in high drama with a long set of ko battles that ends with a big capture for white. Exciting!
I don’t know how these Meijin games work - if one person has won 4 out of 5 games, do they bother playing the last two?
Thanks again for posting these as usual I’ve bumped it up to a full strength review!
I think it’s the best out of 7 games.
Since Shibano won 4 games, he’s the title winner.
Also history for previous editions seems to confirm that:
@BHydden, talking about AI smoking something, have you noticed this?
Move 61 in the last game linked (match played on 2019-10-08).
AI doesn’t even consider that move since move 50, nor for black, nor for white.
And then, after white F17 (-19% BTW) suddenly appears the god’s move: +19%
I’ve never seen a positive number so big before, usually the best is +2/+3%
Despite that, AI would play B4 instead, that is a miserable + 0.8%.
In my (not so wide) experience, when AI smells a good move it insist to ask for that move to both players until it’s done. In my games that happens frequently: I overlook a crucial move and then AI stresses that move for the rest of the game until me or my opponent play it.
Here is another one: move 77.
Again +19%. Again AI would play E15 instead, that is 0.8%.
Isn’t that weird?
The weaker AI (15x192) has a less drastic graph (the one above):
but has more or less the same behavior.
Move 61 is +14%
Move 77 is +26% but at least AI wants to play it,
Move 66 H13 is a +6.6% for white, but lets black play H14 which is +45%!!! I’m astonished.
Move 83. AI would play 0.5% instead of 30%:
There are also other examples from the ko fight, but in that situation I’m more inclined to accept weird scores.
I had the same concerns @lysnew
@mark5000 told me it comes about when the pros play suboptimal moves and so it branches into a tree that the engine hadn’t explored (since they’re programmed to concentrate most on the branches they think are best)
The big numbers mean the move is better than originally estimated given the new context, but still not better than the original line the bot preferred. (I hope I wrote all that right)
I just like the image of the Leela AI hitting a big doobie and going
“Duuuude, if you played here, your winrate would increase by like … 45%…”
To me it’s more like “well, you f***ed up… But, I guess this is ok too…”
And to me is more like: “oh, I was so focused on my moves that I didn’t notice yours!”