My mistake.
Elf seems to think that this 2 is a brilliant (and difficult to spot) move that punishes 1 in all cases.
Seems reasonable - black gets the corner without all that pushing and the white wall (or if white takes the corner, it's small and the centre stones start looking weak)
The only reason Elf agreed with master in the teaching tool position is because Elf failed to spot the move (2 above) on that occasion. Show Elf the move and you get the same huge swing.
Does this mean that Master missed the move? Or does Master have some incredible sequence to refute the refutation?
MikeKyle Analyses Komoku, high approach, low 1 space pincer
-
Uberdude
- Judan
- Posts: 6727
- Joined: Thu Nov 24, 2011 11:35 am
- Rank: UK 4 dan
- GD Posts: 0
- KGS: Uberdude 4d
- OGS: Uberdude 7d
- Location: Cambridge, UK
- Has thanked: 436 times
- Been thanked: 3718 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
That move is known to humans (but not me!), and josekipedia (citing some joseki dictionary) says it's good for white to then live in the corner: http://josekipedia.com/#path:qdodmcnepf ... rcsdsbsfra. Is that judgement usually correct or usually wrong according to super-strong bots? Or maybe this whole board situation changes things...MikeKyle wrote: Elf seems to think that this 2 is a brilliant (and difficult to spot) move that punishes 1 in all cases.
<snip>
Does this mean that Master missed the move? Or does Master have some incredible sequence to refute the refutation?
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
"Show Elf the move and you get the same huge swing."MikeKyle wrote: Elf seems to think that this 2 is a brilliant (and difficult to spot) move that punishes 1 in all cases.
{snip}
The only reason Elf agreed with master in the teaching tool position is because Elf failed to spot the move (2 above) on that occasion. Show Elf the move and you get the same huge swing.
Does this mean that Master missed the move? Or does Master have some incredible sequence to refute the refutation?
This observation underscores the fact that all players are fallible, superhuman bots included.
We need to maintain a healthy skepticism about their evaluations, even while recognizing that they are better than ours. Errors in evaluations are greatest, as a rule, in the opening. From a scientific point of view, one problem with current bot evaluations is that they do not come with an error estimate. We would like to say that if play A is evaluated as 2% better than play B, that play B is a mistake, but we do not have enough evidence to say that. We are guessing.
Currently the focus in go bot research seems to be on producing the strongest bots, not on producing the most accurate evaluations, even though the two are related. As a human, I would find the latter goal more useful.
Even so, I would like evaluations to come with error estimates.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.
-
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: MikeKyle Analyses Komoku, high approach, low 1 space pin
Move a is the best if you calculate this deep.
Move b is the best if you calculate this deep.
Calculation depth is approaching infinity.
-> There is no absolute best move.
(If you have difficulties to imagine the infinite calculation depth, consider ko and an increasing number of unstable stone groups on the board all related to the status of the whole board and the game.)
Move b is the best if you calculate this deep.
Calculation depth is approaching infinity.
-> There is no absolute best move.
(If you have difficulties to imagine the infinite calculation depth, consider ko and an increasing number of unstable stone groups on the board all related to the status of the whole board and the game.)
-
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: MikeKyle Analyses Komoku, high approach, low 1 space pin
And for all practical purposes, which of the good moves is accessible to you? It may be a very strong move, but can you handle it? 
I love go, finest game discovered yet.
I love go, finest game discovered yet.
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
I'd be really interested to query Elf specifically on some of these sequences when I got the chance, but I gave it a fair amount of time for each move in that previous continuation sequence - it seems Elf likes to threaten to live but then thinks its a better idea to strengthen the floaty outside group (in sente as black needs to deal with the corner)Uberdude wrote:says it's good for white to then live in the corner
I agree absolutely.all players are fallible, superhuman bots included.
I'm also very guilty of forgetting this attitude at times though. Sometimes I treat anyone 2 grades stronger than me as an oracle of absolute truths!
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
Yes - Absolutely, this is problematic. Also though I wonder if any metric of confidence might end up being a bit misleading. Confidence is pretty hard to come by when faced with this kind of complexity. Variance of reading seems like a fuzzy metric.Bill Spight wrote: From a scientific point of view, one problem with current bot evaluations is that they do not come with an error estimate.
I guess our best proxy to confidence is the degree to which the win% jumps around. This would tell us that Elf is not confident in these branches
Me too. And your other post is a good reason to:Gomoto wrote: I love go, finest game discovered yet.
The depth and complexity is such that we can get the illusion of a great breakthrough, but there is never actually an answer.
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
Actually, the researchers could generate error estimates for evaluations if they wanted to. But the accuracy of evaluation is not their concern. Beating other bots is. So evaluations, which make up only part of the strength of bots, are not tested, except indirectly, through play against other bots or self play.MikeKyle wrote:Yes - Absolutely, this is problematic. Also though I wonder if any metric of confidence might end up being a bit misleading. Confidence is pretty hard to come by when faced with this kind of complexity. Variance of reading seems like a fuzzy metric.Bill Spight wrote: From a scientific point of view, one problem with current bot evaluations is that they do not come with an error estimate.
I guess our best proxy to confidence is the degree to which the win% jumps around. This would tell us that Elf is not confident in these branches
Edit: This is not by accident. It turns out that improving some aspects of a bot can make it perform worse overall. Strong bots are complex.
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.
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
So one of the sources of confusion from Elf might have been a ladder that pops up in some of Elf's continuations. The ladder weakness is a pretty big, pretty obvious reason to give these reads a pinch of salt. Variation with a ladder missread here, but also there's a fun tesuji. Elf abandons the tesuji after a little bit of further reading but I thought it looked interesting.
In summary I think to play the original kosumi you need to be prepared to give away a large corner to take the outside. It seems that you can be pretty sure of this result if you choose to block at the 3-2 point choice. If you choose the other 3-2 point then you must be prepared to take the corner and fight in the centre.
In summary I think to play the original kosumi you need to be prepared to give away a large corner to take the outside. It seems that you can be pretty sure of this result if you choose to block at the 3-2 point choice. If you choose the other 3-2 point then you must be prepared to take the corner and fight in the centre.
- Attachments
-
- Elf curious joseki reading 2018-05-18.sgf
- (335 Bytes) Downloaded 881 times
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
To look at the other major Elf patern a little..
What Elf likes next seems to be highly dependant on the whole board.
Here Elf favours tenuki. 'a' is the best local follow up, but approaching is bigger (about 2% difference.) I supose that there isn't a very pleasing extension across the bottom without leaving a strange spacing somewhere(?)
In this case Master and Elf agree again. 8 is just about a certainty. In this case, it seems to me that settling this group is too important. 'a' is Elf's second favourite again, but it's about 5% behind. White 'a' is followed by black 8, white 'b'.
What Elf likes next seems to be highly dependant on the whole board.
Here Elf favours tenuki. 'a' is the best local follow up, but approaching is bigger (about 2% difference.) I supose that there isn't a very pleasing extension across the bottom without leaving a strange spacing somewhere(?)
In this case Master and Elf agree again. 8 is just about a certainty. In this case, it seems to me that settling this group is too important. 'a' is Elf's second favourite again, but it's about 5% behind. White 'a' is followed by black 8, white 'b'.
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
I want to start looking at human joseki and different board positions and the different choices involved but it's been tough to find time. In the meantime, I have some more to add to my personal stream of consciousness about this Ai knight's move main line.
I'm at the 'Not the London open' tournament this weekend. Pavol Lisy 3p is here as the tournament pro and Uberdude is at the same venue for the UK Challengers'League.
In my second round game I got to try out my planned opening in an effort to induce this new pattern and see what people try.. White butted but then descended below 8, letting me harass the eye space and get a nice result (before making much more serious errors myself in the rest of the game.) But I was lucky enough to get the game reviewed by Pavol who suggested an interesting line..
In a separate conversation, Uberdude suggested that maybe Elf and AG are comfortable with this because of something like...
(this is my memory of the discussion, please attribute any kyu level errors to my tired kyu brain)
It seems black must be happy to get this weird thin looking formation if black is going to play the knight's move at the start.
I'm quite interested to ask the Elf when I'm back home.
I'm at the 'Not the London open' tournament this weekend. Pavol Lisy 3p is here as the tournament pro and Uberdude is at the same venue for the UK Challengers'League.
In my second round game I got to try out my planned opening in an effort to induce this new pattern and see what people try.. White butted but then descended below 8, letting me harass the eye space and get a nice result (before making much more serious errors myself in the rest of the game.) But I was lucky enough to get the game reviewed by Pavol who suggested an interesting line..
In a separate conversation, Uberdude suggested that maybe Elf and AG are comfortable with this because of something like...
(this is my memory of the discussion, please attribute any kyu level errors to my tired kyu brain)
It seems black must be happy to get this weird thin looking formation if black is going to play the knight's move at the start.
I'm quite interested to ask the Elf when I'm back home.
-
Tryss
- Lives in gote
- Posts: 502
- Joined: Tue May 24, 2011 1:07 pm
- Rank: KGS 2k
- GD Posts: 100
- KGS: Tryss
- Has thanked: 1 time
- Been thanked: 153 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
I think white will play like this instead :
And it's worse for black (there's a severe weakness at S6 )
And it's worse for black (there's a severe weakness at S6 )
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
Ah yes. Thanks. I thought there was something fishy around my 19 & 20.
Surely black can't try to steal sente and leave this hanami ko?
And if black patches up with 22 white can either break through the gap or even just play along the bottom and let black take gote if he wants some weird thin influence.
Interested to see if this upsets Elf again. I have more confidence in master, but we don't get to ask too many follow up questions.
Surely black can't try to steal sente and leave this hanami ko?
And if black patches up with 22 white can either break through the gap or even just play along the bottom and let black take gote if he wants some weird thin influence.
Interested to see if this upsets Elf again. I have more confidence in master, but we don't get to ask too many follow up questions.
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
Elf is on it.
1 seems to be kind of a trick play, and 2 is getting well and trully tricked. Black looses about 30% win rate with 2 for mainly the reasons we were looking at (or also white can block the other way, live by kicking the pincer stone and then a little bit later hastle the lower side group)
Elf likes:
Then a is the next local move for black, or b for white, but Elf sugests that neither is super urgent.
1 seems to be kind of a trick play, and 2 is getting well and trully tricked. Black looses about 30% win rate with 2 for mainly the reasons we were looking at (or also white can block the other way, live by kicking the pincer stone and then a little bit later hastle the lower side group)
Elf likes:
Then a is the next local move for black, or b for white, but Elf sugests that neither is super urgent.
-
MikeKyle
- Lives with ko
- Posts: 205
- Joined: Wed Jul 26, 2017 2:27 am
- Rank: EGF 2k
- GD Posts: 0
- KGS: MKyle
- Has thanked: 49 times
- Been thanked: 36 times
Re: MikeKyle Analyses Komoku, high approach, low 1 space pin
I'm confident that John was talking about this human joseki:John Fairbairn wrote:One characteristic of the commonest line in this joseki, which involves a lot of very close contact, is that it offers several ataris that amateurs love to play but pros avoid.
I've been having fun with go review partner and I used it to look at Elf's opinion of this sequence in 75 9p vs 9p games containing this patern (method briefly explained in the inroducing grp thread. I know that 9p vs 9p is not an exact 1-1 match to quality games but it seemed a reasonable way to get a good number of games)
It actually surprised me a little to see that Elf wanted to play this patern on at least one ocasion in my analysis. I think if Elf wrote a joseki book then this patern would be listed as a special case. Elf likes it for example in this game:
White ignores the pincer for a long time in this game, according to Elf, white should have played the 3-3 point sooner, but as soon as he does, the following play is exactly as Elf would have played it on both sides.
However in general, Elf likes other sequences..
First of all - the hane after the 3-3 (move 2 above.) Elf really likes that knights move! It prefers the knights move to the hane in around 50 of the 75 board positions in which the 9d's chose the hane (Elf's opinion on the remaining 25ish moves were split - not all for the hane.) It seems to most commonly be around a 4-5% mistake.
There's a little bit of noise at move 4 too. Elf mainly agrees with the popular human move, but on a couple of occasions looks at other moves - for example here:
I think Elf plays 1 as black here in order to take sente for the open corner. This patern appears in 'The 21st century dictionary of basic Joseki' on a diagram with the words "foiling the trick move" (1 is the trick move) except that according to the dictionary, black must play a and finish in gote because leaving white the ko is too good. Elf thinks that black can now tenuki and take the final corner and then (after making an outside exchange) white will turn in order to set up the ko for later. (the ko is white c, black d and then throw in at e to set up a 2 stage ko. If white started it now I think black is very happy to take 3 large points while white takes the stones.)