alphago self-analysis by fan hui

[Bill Spight]

Update: An important part of science is to state your hypothesis before testing and publish results regardless . This is new install of kombilo with the 50 Master self-play and I seem to have the next colour setting messed up, but with white (or black if colours are swapped) to play next in the big quadrant I get 15 games in total (4 with black 4-4, 11 with white, is this Master preferring 4-4s over 3-4s more when white?). Slide has 0 hits, as expected. But most popular is extend to hoshi (5), one point jump up (3), mini-Chinese extend (3), and attach is only 4th with 2.

Update 2: Doing the search on the 20 AG Zero vs AG Master games, there are 7 hits, now attach is #1 with 5, and 1 each of one point jump out and a stone appearing on top side as part of a group there. That's the result I was thinking of, though we only have 20 games so not a great sample (and the openings are all pretty much the same for a while).

Hypothesis testing is not the only kind of scientific research. One kind is to come up with hypotheses. As you point out, the Master games are too few to do much in the way of testing, but they are great for generating hypotheses.

[Uberdude]

I wonder how big is the difference between the traditional slide at a compared to AG's preferred tsuke.
It may be so small that it does not matter for practical purposes in human play.

How big by what metric? In AG "win %" it can often be about 5% in fairly common openings (e.g. choice to slide or make micro Chinese opening), which in my interpretation is a fairly significant mistake. Or if a player of Park Junghwan calibre makes the slide mistake and then plays the rest of the game at his consistent level against the same level opponent how many points does he lose by? 1 or 2 is my guess.

Obviously at my weak 4d amateur level it's not a big deal, and if I could play the whole game as well/badly as a sliding 9p did from the 1990/2000s I wouldn't be 4d for long, but even so I think ideas about the slide can improve my game. I don't play it so much now, and also will more readily ignore it if my opponent slides because I'm more aware of the difficulties he can have in attacking my group when I keep tenuki-ing. Making the early attachment instead though is not something I've really worked into my game yet. The argument John mentioned from Takagawa about the tewari inefficiency is one I was aware of (not sure if I worked it out myself or heard it) pre-AlphaGo.

I noticed that FineArt prefers the slide at a in fact: I watched many of its 4 handicap games on FoxGo server and it seemed to almost always start with approach, slide and 2 space jump (when black replies the slide at 3x3). Which by the way seems a very dull/slow way for white to start in a 4 handicap game, but FineArt somehow managed to still win many such games against Fox 9d amateurs.

Interesting; maybe because it doesn't settle the shape quite so much? I suppose the humans overplay and it outfights them later? Do you have an example?

[sorin]

I wonder how big is the difference between the traditional slide at a compared to AG's preferred tsuke.
It may be so small that it does not matter for practical purposes in human play.

How big by what metric? In AG "win %" it can often be about 5% in fairly common openings (e.g. choice to slide or make micro Chinese opening), which in my interpretation is a fairly significant mistake. Or if a player of Park Junghwan calibre makes the slide mistake and then plays the rest of the game at his consistent level against the same level opponent how many points does he lose by? 1 or 2 is my guess.

Obviously at my weak 4d amateur level it's not a big deal, and if I could play the whole game as well/badly as a sliding 9p did from the 1990/2000s I wouldn't be 4d for long, but even so I think ideas about the slide can improve my game. I don't play it so much now, and also will more readily ignore it if my opponent slides because I'm more aware of the difficulties he can have in attacking my group when I keep tenuki-ing. Making the early attachment instead though is not something I've really worked into my game yet. The argument John mentioned from Takagawa about the tewari inefficiency is one I was aware of (not sure if I worked it out myself or heard it) pre-AlphaGo.

The metric is tricky, since we cannot really measure it using human play - too few games, too hard to control for the experiment.
For either slide or attachment, a strong player will make it work in the end.
It is possible that an AI stronger than AG will conclude that after all the slide is better than the attachment, for them.

I noticed that FineArt prefers the slide at a in fact: I watched many of its 4 handicap games on FoxGo server and it seemed to almost always start with approach, slide and 2 space jump (when black replies the slide at 3x3). Which by the way seems a very dull/slow way for white to start in a 4 handicap game, but FineArt somehow managed to still win many such games against Fox 9d amateurs.

Interesting; maybe because it doesn't settle the shape quite so much? I suppose the humans overplay and it outfights them later? Do you have an example?

Of course I cannot easily find such games now that I am looking for them to give examples
While doing this search, I realize I may have misspoken: FineArt doesn't play predominantly slides, judging by the sample of games I just looked at.
Nevertheless, here are some examples:

FineArt responding to slide by playing 3x3:
http://foxwq.com/qipu/newlist/id/2018020173164837.html

FineArt playing the slide:
http://foxwq.com/qipu/newlist/id/2018020785331443.html
http://foxwq.com/qipu/newlist/id/2018020169143108.html

[dhu163]

just finished the self-analysis 1 video above.

[dhu163]

Number 3
up to 23 mins (edit: 50) (edit: finished all 108 mins)

particularly crazy ko fight in the opening.

[dhu163]

finished the set of 5

[Uberdude]

Thanks dhu, great work making these available to English-speaking audiences. For ease of reference here are all the games for download:

Game 1:
download/file.php?id=9629

Game 2:
download/file.php?id=8654

Game 3:
forum/download/file.php?id=9962

Game 4:
download/file.php?id=8663

Game 5:
download/file.php?id=8701

And in online eidogo players:

Game 1:


Game 2:


Game 3:


Game 4:


Game 5: