Reviewing a 1971 Otake game with Kata
-
xela
- Lives in gote
- Posts: 652
- Joined: Sun Feb 09, 2014 4:46 am
- Rank: Australian 3 dan
- GD Posts: 200
- Location: Adelaide, South Australia
- Has thanked: 219 times
- Been thanked: 281 times
Re: Reviewing a 1971 Otake game with Kata
I just want to say thanks Knotwilg for continuing this excellent series. I haven't responded in detail to games 2 and 3 because I've been obsessed with something else recently. But be assured that I'm following this thread with great interest.
-
Knotwilg
- Oza
- Posts: 2432
- Joined: Fri Jan 14, 2011 6:53 am
- Rank: KGS 2d OGS 1d Fox 4d
- GD Posts: 0
- KGS: Artevelde
- OGS: Knotwilg
- Online playing schedule: UTC 18:00 - 22:00
- Location: Ghent, Belgium
- Has thanked: 360 times
- Been thanked: 1021 times
- Contact:
Re: Reviewing a 1971 Otake game with Kata
Game 4
Otake takes Black here, against Kudo Norio White.
While the opening evolves fairly evenly, both sides taking a small lead at certain points, in the middle game Black takes the advantage, peaking at move 85
At 103, Otake seems to misjudge the strength of his stones at the top and in the middle. He attacks for profit on the right side, but White counterattacks, involving Black's groups at the top, on the left and on the right.
Through 119, Black clings to his stones at the top, which become heavy, while White's stones are light and can be exchanged for profit on the left side.
At 133 Otake decides to live locally on the right side, giving White the momentum to strengthen the centre, so that Black's top or left side will suffer. From that point onwards, there's no chance to come back
While Otake is known to play solidly and perhaps slowly in the early parts of the game, usually the strength he builds pays off in the end. This game is unusual in that respect: at 103 he could have strengthened the top and maintain the advantage.
Otake takes Black here, against Kudo Norio White.
While the opening evolves fairly evenly, both sides taking a small lead at certain points, in the middle game Black takes the advantage, peaking at move 85
At 103, Otake seems to misjudge the strength of his stones at the top and in the middle. He attacks for profit on the right side, but White counterattacks, involving Black's groups at the top, on the left and on the right.
Through 119, Black clings to his stones at the top, which become heavy, while White's stones are light and can be exchanged for profit on the left side.
At 133 Otake decides to live locally on the right side, giving White the momentum to strengthen the centre, so that Black's top or left side will suffer. From that point onwards, there's no chance to come back
While Otake is known to play solidly and perhaps slowly in the early parts of the game, usually the strength he builds pays off in the end. This game is unusual in that respect: at 103 he could have strengthened the top and maintain the advantage.
-
Knotwilg
- Oza
- Posts: 2432
- Joined: Fri Jan 14, 2011 6:53 am
- Rank: KGS 2d OGS 1d Fox 4d
- GD Posts: 0
- KGS: Artevelde
- OGS: Knotwilg
- Online playing schedule: UTC 18:00 - 22:00
- Location: Ghent, Belgium
- Has thanked: 360 times
- Been thanked: 1021 times
- Contact:
Re: Reviewing a 1971 Otake game with Kata
And game 5, where Fujisawa Shuko takes control of the game fairly linearly until he blunders and resigns (while the game is still close).
-
xela
- Lives in gote
- Posts: 652
- Joined: Sun Feb 09, 2014 4:46 am
- Rank: Australian 3 dan
- GD Posts: 200
- Location: Adelaide, South Australia
- Has thanked: 219 times
- Been thanked: 281 times
Re: Reviewing a 1971 Otake game with Kata
Game 2: move 135
White 36 also looks incredibly weird to me. I would play the atari at N15 almost without thinking. It seems that so would KataGo.
Actually, KataGo (and LZ network 258) has to do quite a lot of reading to find that J4 is best: it takes about 6,000 playouts when I try it. If you click "show policy" in Lizzie, you'll see why. The bots think that first, pulling out of atari at E5 is the most instinctive move, so they have to read that for a while before trying other things. (It's actually quite neat to see how white's F6 stones can connect out to safety.) Then H4 and J3 also get moderate policy values (over 10%), and KataGo (but not LZ-258) thinks the game move also looks better than J4 according to the policy network.It's strange that Otake, who's the "man of shape" misses the move that Kata thinks is better and which is visually better shape too
White 36 also looks incredibly weird to me. I would play the atari at N15 almost without thinking. It seems that so would KataGo.
-
Knotwilg
- Oza
- Posts: 2432
- Joined: Fri Jan 14, 2011 6:53 am
- Rank: KGS 2d OGS 1d Fox 4d
- GD Posts: 0
- KGS: Artevelde
- OGS: Knotwilg
- Online playing schedule: UTC 18:00 - 22:00
- Location: Ghent, Belgium
- Has thanked: 360 times
- Been thanked: 1021 times
- Contact:
Re: Reviewing a 1971 Otake game with Kata
Patterns found in these 5 games
It's a small sample for patterns but it gives me a hypothesis to further confirm or disprove.
1. Ko fights
On two occasions, Otake and his opponents at the time did not take an opportunity for ko which Kata recommends. Kata seems generally more adventurous or willing to make the exchange.
game 1 - move 33
game 4 - move 82
2. being forced (to capture what's yours)
When the dust of a fight is about to settle, forcing sequences occur, i.e. forcing the opponent to take what's already his but then at the smallest possible scale. The idea that being forced this way is bad, is not novel, but Kata regularly points it out in these games
game 1 - move 89, 154
game 5 - move 105
3. forcing (to stabilize with territory)
On the other hand, we often see Kata boosting the probabilities of a side which has just been "forced" to live with territory, where the status was previously not clearly settled. It appears to be better not to settle the shape this way and leave the aji.
game 1 - move 91
game 3 - move 66
4. boshi reductions/attacks for influence
while shoulder hits are bots' favorites, boshi reductions or even attacks clearly aren't; exception: when the influence cooperates with influence at the other hand, creating potential territory
game 2 - move 31, 60
game 3 - move 57, 61
5. simplifying the centre vs simplifying the corner
when there's a big fight in which there's a choice to simplify by taking a big corner, leaving the centre insecure, or simplifying the centre, leaving the corner at stake, bots prefer central complexity more than pros do
game 3 - move 54, 69
6. tactical "blunders"
there are situations where one player has a clear lead but then makes a tactical mistake that loses the game
game 3 - 139
game 4 - 133
game 5 - 149
It's a small sample for patterns but it gives me a hypothesis to further confirm or disprove.
1. Ko fights
On two occasions, Otake and his opponents at the time did not take an opportunity for ko which Kata recommends. Kata seems generally more adventurous or willing to make the exchange.
game 1 - move 33
game 4 - move 82
2. being forced (to capture what's yours)
When the dust of a fight is about to settle, forcing sequences occur, i.e. forcing the opponent to take what's already his but then at the smallest possible scale. The idea that being forced this way is bad, is not novel, but Kata regularly points it out in these games
game 1 - move 89, 154
game 5 - move 105
3. forcing (to stabilize with territory)
On the other hand, we often see Kata boosting the probabilities of a side which has just been "forced" to live with territory, where the status was previously not clearly settled. It appears to be better not to settle the shape this way and leave the aji.
game 1 - move 91
game 3 - move 66
4. boshi reductions/attacks for influence
while shoulder hits are bots' favorites, boshi reductions or even attacks clearly aren't; exception: when the influence cooperates with influence at the other hand, creating potential territory
game 2 - move 31, 60
game 3 - move 57, 61
5. simplifying the centre vs simplifying the corner
when there's a big fight in which there's a choice to simplify by taking a big corner, leaving the centre insecure, or simplifying the centre, leaving the corner at stake, bots prefer central complexity more than pros do
game 3 - move 54, 69
6. tactical "blunders"
there are situations where one player has a clear lead but then makes a tactical mistake that loses the game
game 3 - 139
game 4 - 133
game 5 - 149
-
xela
- Lives in gote
- Posts: 652
- Joined: Sun Feb 09, 2014 4:46 am
- Rank: Australian 3 dan
- GD Posts: 200
- Location: Adelaide, South Australia
- Has thanked: 219 times
- Been thanked: 281 times
Re: Reviewing a 1971 Otake game with Kata
Back to game 3 (sorry, I'm not quite keeping up with the pace here) -- a couple of interesting variations:
Position after move 35: Here, Otake played at a. I wondered if this was too slow and if white could play in the lower right instead. "Obviously" if white tenukis, black can cut at b. But I had trouble reading out a clear continuation. Here's one possibility, with a little help from LZ-258.
Next white a, and black b or c in either order. It looks as though black loses the capturing race but can squeeze on the outside, so on the whole black is ahead (around 60% winrate).
Later, after move 84: It looks to me like white needs one more move to make a living shape. But black can't afford to try and kill, because the circled stones are fragile.
If black does try a blunt attack now: Black can follow up at a and rip off three stones, leaving the rest of the corner as seki -- but black's marked stones probably die.
Position after move 35: Here, Otake played at a. I wondered if this was too slow and if white could play in the lower right instead. "Obviously" if white tenukis, black can cut at b. But I had trouble reading out a clear continuation. Here's one possibility, with a little help from LZ-258.
Next white a, and black b or c in either order. It looks as though black loses the capturing race but can squeeze on the outside, so on the whole black is ahead (around 60% winrate).
Later, after move 84: It looks to me like white needs one more move to make a living shape. But black can't afford to try and kill, because the circled stones are fragile.
If black does try a blunt attack now: Black can follow up at a and rip off three stones, leaving the rest of the corner as seki -- but black's marked stones probably die.
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: Reviewing a 1971 Otake game with Kata
My impulse is to play Otake's move at a, and Elf agrees.xela wrote:Back to game 3 (sorry, I'm not quite keeping up with the pace here) -- a couple of interesting variations:
Position after move 35: Here, Otake played at a. I wondered if this was too slow and if white could play in the lower right instead.
Maybe more later. Without doing any reading myself, I suspect that the weakness of the squared stones is also a factor. In game Honda extended from them to make a base, and then Otake played at a.Later, after move 84: It looks to me like white needs one more move to make a living shape. But black can't afford to try and kill, because the circled stones are fragile.
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
-
xela
- Lives in gote
- Posts: 652
- Joined: Sun Feb 09, 2014 4:46 am
- Rank: Australian 3 dan
- GD Posts: 200
- Location: Adelaide, South Australia
- Has thanked: 219 times
- Been thanked: 281 times
Re: Reviewing a 1971 Otake game with Kata
Up to game 4 now :-) I also did a double-take when I saw the iron pillar at move 19. But what's just as strange (to my eyes) is that LZ-258 wants white to reply not at O7 as in the game, but with the attachment at P7. This leads to a nice tesuji which I was actually able to spot because I've just been looking at the relevant chapter in the Segoe dictionary!
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: Reviewing a 1971 Otake game with Kata
Speaking of game 4, a brief note about White 126.
KataGo prefers the block at F-09, yielding a 67% winrate for White with 56k rollouts. Elf, in its commentary, also likes F-09, devoting almost 143k rollouts to it and giving it an 79.8% winrate. However, even though Kudo's play, the atari at R-10, is almost entirely off Elf's radar, getting only 4 rollouts, Elf gives it a 77.7% winrate, only 2.1% less than the block. That is well within Elf's margin of error. (With only 4 rollouts, Elf does not calculate a winrate directly for R-10, but inherits its winrate from Black 127. Elf does that for any play with fewer than 500 rollouts. )
KataGo prefers the block at F-09, yielding a 67% winrate for White with 56k rollouts. Elf, in its commentary, also likes F-09, devoting almost 143k rollouts to it and giving it an 79.8% winrate. However, even though Kudo's play, the atari at R-10, is almost entirely off Elf's radar, getting only 4 rollouts, Elf gives it a 77.7% winrate, only 2.1% less than the block. That is well within Elf's margin of error.
(With only 4 rollouts, Elf does not calculate a winrate directly for R-10, but inherits its winrate from Black 127. Elf does that for any play with fewer than 500 rollouts. )The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
-
xela
- Lives in gote
- Posts: 652
- Joined: Sun Feb 09, 2014 4:46 am
- Rank: Australian 3 dan
- GD Posts: 200
- Location: Adelaide, South Australia
- Has thanked: 219 times
- Been thanked: 281 times
Re: Reviewing a 1971 Otake game with Kata
This is an interesting one: the various bots disagree here. LZ-258 doesn't find the game move on its own (on 10k playouts), but awards it +1% when shown. Of course LZ-258 thinks komi is 7.5 not 5.5. Trying with KataGo 1.2 and 1.3 (out since you first posted this!): for me, KataGo 1.2 has the game move at -4% (different random number seed?) whereas 1.3 says -7%, setting komi at 5.5. The numbers move around a little if you change komi to 5.0 or to 7.5, but S15 often looks like a possible small loss not a clear mistake. (KataGo disagrees with LZ-258 even on 7.5 komi.) Interestingly, if you set komi to zero, KataGo searches a bit more broadly.Knotwilg wrote:At 103, Otake seems to misjudge the strength of his stones at the top and in the middle. He attacks for profit on the right side, but White counterattacks, involving Black's groups at the top, on the left and on the right... in the SGF comments: This move loses 8%
This is also a nice test case for KataGo 1.3's new score estimation (median instead of mean!) Changing the komi between 5.5 and 7.5 makes the score estimate move by about 1.7 or 1.8 points, whereas the same change for KataGo 1.2 results in an almost 4 point change!
-
Knotwilg
- Oza
- Posts: 2432
- Joined: Fri Jan 14, 2011 6:53 am
- Rank: KGS 2d OGS 1d Fox 4d
- GD Posts: 0
- KGS: Artevelde
- OGS: Knotwilg
- Online playing schedule: UTC 18:00 - 22:00
- Location: Ghent, Belgium
- Has thanked: 360 times
- Been thanked: 1021 times
- Contact:
Re: Reviewing a 1971 Otake game with Kata
After the interlude of "Otake's perfect game", I wanted to publish the 7th game in the series, against Kajiwara. It's by far the most complex analysis I've conducted and I'm not at all at the end of it.
Problem: after analyzing with LZ, the file has become too large.
The interesting thing about this game is that - for John's sake! - the pros seem to consider all the complex variations Kata sees. They play the important moves, just in a different order.
The exception may be move 94, a very natural extension from atari, maintaining pressure, which Kata suddenly evaluates as a 23% loss.
Instead Kata offers a variation which can lead to a furikawari, but it ramifies into a complexity I can't contain. From the game, I have no hint Otake or Kajiwara considered this line.
Otherwise though, the game makes a case for John's point that it's the human errors/blunders that make the difference, not the ingenuity of the bot, or its better developed intuition. The "blunder" in question is Kajiwara's 223, a first line hane which seems like it was meant to be a ko threat, keeping sente, but it wasn't.
Problem: after analyzing with LZ, the file has become too large.
The interesting thing about this game is that - for John's sake! - the pros seem to consider all the complex variations Kata sees. They play the important moves, just in a different order.
The exception may be move 94, a very natural extension from atari, maintaining pressure, which Kata suddenly evaluates as a 23% loss.
Instead Kata offers a variation which can lead to a furikawari, but it ramifies into a complexity I can't contain. From the game, I have no hint Otake or Kajiwara considered this line.
Otherwise though, the game makes a case for John's point that it's the human errors/blunders that make the difference, not the ingenuity of the bot, or its better developed intuition. The "blunder" in question is Kajiwara's 223, a first line hane which seems like it was meant to be a ko threat, keeping sente, but it wasn't.
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: Reviewing a 1971 Otake game with Kata
Well, if by blunder we mean a large reading error, there are at least three sources of error, 1) incorrect calculation of variations; 2) poor choice of candidate moves; 3) incorrect evaluation of results. As for 1) bots are perfect and human pros, nearly so, (Edit: OC, the bots can make a much more extensive search in the same amount of time than humans can do consciously; as for human unconscious search, quien sabe?) 2) is related to the bot's policy network, and 3) is related to the bot's evaluation network. It is the advances in both the choice of candidate plays and the evaluation of the results that enabled the bots to leap to superhuman play.Knotwilg wrote:Otherwise though, the game makes a case for John's point that it's the human errors/blunders that make the difference, not the ingenuity of the bot, or its better developed intuition. The "blunder" in question is Kajiwara's 223, a first line hane which seems like it was meant to be a ko threat, keeping sente, but it wasn't.
Last edited by Bill Spight on Sat Jan 25, 2020 11:26 am, edited 1 time in total.
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
-
lightvector
- Lives in sente
- Posts: 759
- Joined: Sat Jun 19, 2010 10:11 pm
- Rank: maybe 2d
- GD Posts: 0
- Has thanked: 114 times
- Been thanked: 916 times
Re: Reviewing a 1971 Otake game with Kata
Although, there is some room for better ingenuity too? One or two interesting points came up in a game review Michael Redmond streamed here recently about the right side battle in one of his own games. It doesn't seem quite right to call human pros missing the moves merely blunders, it feels more like, the bots have developed some understanding of some kinds of probes and flexible shapes that humans haven't quite internalized yet.Otherwise though, the game makes a case for John's point that it's the human errors/blunders that make the difference, not the ingenuity of the bot, or its better developed intuition.
https://www.twitch.tv/videos/540764912?t=2h40m20s
-
xela
- Lives in gote
- Posts: 652
- Joined: Sun Feb 09, 2014 4:46 am
- Rank: Australian 3 dan
- GD Posts: 200
- Location: Adelaide, South Australia
- Has thanked: 219 times
- Been thanked: 281 times
Re: Reviewing a 1971 Otake game with Kata
If you've analysed and saved the file using Lizzie, it will have added a lot of tags of the form:Knotwilg wrote:Problem: after analyzing with LZ, the file has become too large.
Code: Select all
LZ[0.7 45.6 254 move R17 visits 202 winrate 5537 scoreMean 0.582301 pv R17 Q16 Q18 P18 O18 R18 P19 S17 C16 D3 info move C16 visits 20 winrate 5140 scoreMean -0.0940582 pv C16 P16 O17 Q18 D3 C15 D15 E16 info move Q16 visits 14 winrate 5001 scoreMean -0.18847 pv Q16 P16 R16 R17 S17 S16 R15 S18 info move Q18 visits 8 winrate 5081 scoreMean -0.167722 pv Q18 R17 C16 D3 P16 info move Q4 visits 5 winrate 5046 scoreMean -0.224007 pv Q4 Q3 R5 P4 info move C3 visits 4 winrate 4957 scoreMean -0.350214 pv C3 C16 R17 info move D3 visits 1 winrate 4875 scoreMean -0.543076 pv D3 info ];
Those tags allow Lizzie to remember the evaluations and hover-over variations next time you open the file. But they make it a lot bigger. If you have a text editor that can do search and replace using regular expressions, then you can probably get rid of all the LZ[...] tags by replacing "LZ\[.*?\n.*\];" with ";". (I say probably because regex isn't perfectly standardised and can be tricky.) If this doesn't work for you, I'm happy to discuss other options either here or by PM. I'm keen to see your analysis :-)
Last edited by xela on Sun Jan 26, 2020 5:33 am, edited 1 time in total.
-
Gomoto
- Gosei
- Posts: 1733
- Joined: Sun Nov 06, 2016 6:56 am
- GD Posts: 0
- Location: Earth
- Has thanked: 621 times
- Been thanked: 310 times
Re: Reviewing a 1971 Otake game with Kata
Analyze with Katago and annotate in Sabaki a clean second copy is another option.
(Sabaki should probably add an option to get rid of the tags in the clean sgf dialog.)
(Sabaki should probably add an option to get rid of the tags in the clean sgf dialog.)