Life In 19x19

I agree the paper might have benefitted from conferring with expert Go players.

I also have a critique on the mathematical side of things. The author writes:

We can see that opening sequences to a depth of 50 moves tend to cluster into distinct groups characterized by the very first two moves ...

I looked at the example and based on my understanding of how the search and neural net interact I suspect you shouldn't read too much into this case, unless you can actually find some variations where failing to make the exchange earlier costs something? Sometimes you can actually find some ...

Thanks!

Do you know where one can obtain the game record? With some searching online I found some videos of earlier matches at https://www.youtube.com/@nakaneigo-doujyo/videos but seemingly not of the match you presented the opening of here.

@And - super cool posting these games, thanks.

It's very interesting to see how massively it improves with even 3 visits.

Pure shape intuition. Don't know and haven't tried to read it out.

Feels like black should win this unconditionally.

$$B
$$ ------------------
$$ . . X . . 2 1 . |
$$ . . X X X O . . |
$$ . . . . X O O . |
$$ . . . . X X O . |
$$ . . . . . O X 3 |
$$ . . . . . O X . |
$$ . . . . . O X ...

Was it in significant use prior to your own usage in research as well? Unfortunately, I would guess that "transposition" is still probably 10 to 100 times more common, if for no other reason that it is the term in chess and chess players both greatly outnumber Go players among English speakers, as ...

Each move of each reversion sequence abides by the used superko rule.

Yes, but the value of situation Q may depend on the sequence that it was reached from, for example if you reach situation Q via move sequence A, it might be that a *later* followup from situation Q might have an illegal move ...

If I guess correctly, hardware and software libraries are not suddenly 20 times as efficient on average but it is just a matter of how pointers are redirected over time and how to give more load to the CPU while the GPU remains bottlenecked. Am I right?

Mostly. Pointers aren't redirected, the ...

What is "local life and death region"? Are you setting up a whole-board position where the endgame is all finished but the only thing unfinished is a life and death problem? Or is it just a normal whole-board position where the life and death position is the key to swinging the game? Or is it using ...

Just to be clear on terminology, if the same position is reached by multiple sequences, but that position has not occurred before in the history, then it is normally called "transposition" in the modern tree search and algorithm literature, not "reversion". The word "reversion" shares the same stem ...

Usually, the load is roughly 16% CPU (8C/16T), 95% GPU, 5000 visits/s. If the next move has two reversion candidates with about the same percentages and scores, the load goes up to roughly 85% CPU, 85% GPU, 100,000 visits/s, maintains high load for several or dozens of seconds and then fluctuates ...

alphago introduced the concept of playing endgame moves that lost points because they were safer moves that assured a half-point win.

Since I am not a fan of the AlphaGo/DeepMind invented chewing gum, lobotomy surgery, Star Wars and the penny post school of thought that seems so popular now, I'd ...

If you're using LizzieYZY, there should be a setting where you can configure it to pause analysis if the total number of visits hits a certain number. If you use a value that will be large enough for your purpose but that won't exhaust all your RAM, then it should be fine for you to leave it ...

Yep I think your experience is transferable, but probably to midgame positions with extreme and deep tactics, some of which end in huge winner-take-all swings. For opening and for "common" game positions, the komi and/or current winrate doesn't make nearly as much of a difference, to the degree that ...