It is currently Thu Mar 28, 2024 4:57 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 113 posts ]  Go to page Previous  1, 2, 3, 4, 5, 6  Next
Author Message
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #41 Posted: Wed Dec 04, 2019 1:16 pm 
Gosei

Posts: 1733
Location: Earth
Liked others: 621
Was liked: 310
“All that glisters is not gold; often have you heard that told.”

“AI that glisters is not gold; often have you heard that told.”

I don't toss away my go books. But I don't rewrite them either.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #42 Posted: Wed Dec 04, 2019 1:19 pm 
Gosei

Posts: 1733
Location: Earth
Liked others: 621
Was liked: 310
3-3 is better than approach / approach is better than 3-3

All this urge to teach.
Attachment:
approach.JPG
approach.JPG [ 166.11 KiB | Viewed 6240 times ]

Attachment:
threethree.JPG
threethree.JPG [ 163.9 KiB | Viewed 6240 times ]

At least AI told us already, we are talkin about a huge difference, if we do not follow the one or the other advice.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #43 Posted: Wed Dec 04, 2019 3:03 pm 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Kirby wrote:
You've indicated that the slides don't result in significant loss.


No no no! After posting, I realized my post could be misleading.

Kirby wrote:

Then isn't it possible that this sequence, which optimally results in slides on the second line, is inferior to white's 3-3 invasion, because the 3-3 invasion here is better than this sequence which results in white's two plays on the second line?


Yes yes yes! That's actually what I implied, the jump is worse than the 3-3 invasion, probably because it needs a base and that base can only be found on the second line.

BUT, I still don't rule out the fault only partly lies with the fact that one needs to play on the second line, but also partly with the loss of the corner and perhaps even, on a higher level, with early commitment.

AND, I could still envision a pattern that includes a play on the second line, but is still good. I remember an astonishing sequence of successive 2nd line plays, which I would never think of because I would be "crawling on the line of defeat" but LZ valued the goal of that crawling, connecting two weak groups.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #44 Posted: Thu Dec 05, 2019 7:58 am 
Honinbo

Posts: 9545
Liked others: 1600
Was liked: 1711
KGS: Kirby
Tygem: 커비라고해
Knotwilg wrote:
That's actually what I implied, the jump is worse than the 3-3 invasion, probably because it needs a base and that base can only be found on the second line.


Ah, ok. Maybe I misunderstood what you meant. The question of "why" a move is good or bad is fascinating. In some ways, for both humans and computers, I suppose it comes down to, "because my model of how go works tells me so". Proverbs, like the one in the SL page that's referenced here, are examples of such models, though, they may often be oversimplified - or maybe in some cases, simply wrong (play the last big point, hane at the head of two stones, don't do early 3-3 invasion).

I believe that there are often cases where proverbs like these have some degree of truth, but are oversimplifying things. Thinking in terms of statistics, I'd say that many of these models are prone to being "underfit" - they don't sufficiently describe reality.

LZ is more flexible, because you can take an arbitrary board position, and see what the AI thinks is good. You have a lot more detail here, and more nuances are taken into account.

From that perspective, just using LZ to make new proverbs may have value. That being said, I think we should be careful of overfitting (https://en.wikipedia.org/wiki/Overfitting) our internal models of how go works from looking at LZ analyses. If we look at, say, 10 examples to understand the nuances of a position, there's a good chance that some of the inferences we draw from LZ include bias toward exceptional scenarios that may not be true in the general sense. To reduce that effect, I guess the best thing to do is to analyze more and more games with LZ in order to reduce the impact of outliers.

It's from this idea that I'm wary of dropping some of the proverbs that are out there. Yes, LZ and other bots may show some deficiencies - the models given by proverbs may underfit reality; you can find several exceptions to the rule stated by the proverb. But the same danger is still present with LZ analysis, if your intent is to create a general rule or heuristic.

That being said, maybe the types of models and heuristics we're learning from bots are already more useful than some existing human proverbs. But it's hard for me to really measure that.
¯\_(ツ)_/¯

_________________
be immersed


This post by Kirby was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #45 Posted: Thu Dec 05, 2019 9:42 am 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
I have performed a master edit and a rename of the SL page leading to this discussion. Please check if you find this rendition satisfactory.

https://senseis.xmp.net/?TedomariExercise1


This post by Knotwilg was liked by: Kirby
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #46 Posted: Thu Dec 05, 2019 10:55 am 
Honinbo

Posts: 9545
Liked others: 1600
Was liked: 1711
KGS: Kirby
Tygem: 커비라고해
Knotwilg wrote:
I have performed a master edit and a rename of the SL page leading to this discussion. Please check if you find this rendition satisfactory.

https://senseis.xmp.net/?TedomariExercise1


Seems like a moderate and balanced analysis to me. I like it :tmbup:

_________________
be immersed

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #47 Posted: Thu Dec 05, 2019 11:39 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
I think the result is disastrous. I can see that a big effort has been made to apply caution and fairness, but you can't make a silk purse out of a sow's ear. And when you try you end up destroying the sow's ear anyway. Plus, two wrongs don't make a right.

The real problem lies with the nature of SL. The theme of the awful original page is supposed to be "how to get the last big point." "Awful" because I can't see anywhere that tells you how to do that. It's just an example with no theory. I would expect to see mention of things like creating miai, supported by sente sequences, plus some discussion of how we evaluate each stage of the process (as in "you may win the battle but lose the war" - is there a risk of "you may get the last big point but lose the game"?) Etc. etc. None of that is the fault of the current editor. It is just a glaring example of the typical faults of SL.

But where I think the current editor has gone astray is that, while the LZ move may well be better, the LZ move does NOT answer the question posed as the theme of the page: How to get the last big point. LZ is just answering the question: what is the best move in this position?

As I said, the human version does not answer its own question, but at least it stays on topic, and the example may well inspire others cleverer than me to understand the "how." If it does, that would be a valuable and transferable skill and the fact that the example move is not coincidentally the best move in this (made-up?) case is neither here-nor-there.

Giving LZ half the page and not answering that question - and not explaining that it doesn't - is calamitous for the original page. I don't think it is sufficient to say the page is now "coloured" by LZ analysis. It is swamped. If half your house was daubed in paint by a renowned graffiti artist, I don't think you would be pleased, and you wouldn't describe it as "coloured."

There is plenty of scope for valuable work on teasing out what AI can teach us, but this should be in stand-alone pages. In due course, these pages and human-based pages can simply be cross-referenced to each other.


This post by John Fairbairn was liked by: Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #48 Posted: Thu Dec 05, 2019 12:20 pm 
Honinbo

Posts: 9545
Liked others: 1600
Was liked: 1711
KGS: Kirby
Tygem: 커비라고해
John Fairbairn wrote:
But where I think the current editor has gone astray is that, while the LZ move may well be better, the LZ move does NOT answer the question posed as the theme of the page: How to get the last big point. LZ is just answering the question: what is the best move in this position?


I saw this more of a correction to an example that was intended to be a good sequence. The original example is maintained, and the LZ analysis points out that this example is not necessarily ideal. Maybe that could be more explicitly stated, but this example is not on the main page for getting the last point, either.


Quote:
The theme of the awful original page is supposed to be "how to get the last big point." "Awful" because I can't see anywhere that tells you how to do that.

I agree with this. The page doesn't do a great job of explaining *how* to begin with. If we really want to revamp the page, maybe it'd be good to make an effort to explain the *how* on the main page. Then, give the examples. Then maybe a separate section pointing out that the examples aren't necessarily perfect, showing how several bots don't necessarily think that the examples are good.

---

But my feeling was that Knotwilg's intention was to point out that the existing examples may not necessarily be good in those board positions.

I suppose there may be another school of thought which says, "Even if an example isn't optimal, it's good if it's showing a principle we want to learn". In that case, I suppose any LZ corrections should be omitted if we think we are detracting from the theme of the page..

_________________
be immersed

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #49 Posted: Thu Dec 05, 2019 3:06 pm 
Gosei

Posts: 1733
Location: Earth
Liked others: 621
Was liked: 310
Josekipedia is a nice go related wiki. When I take a look at Josekipedia I do not expect absolute truth. I am more interested what others are thinking about certain sequences, and why a joseki is played or was played in the past. Sometimes Josekipedia helps to identify blind spots of a bot too.

If I want the opinion of a bot, I look at the bot and not at Josekipedia.

(It is fine if somebody puts some AI josekis in Josekipedia or comments older variants with bot help.)

SL is unwieldy if you are searching for a bot move or the strongest move in a position (I just use the bot in this case). SL is a nice place if you are interested in the opinions of other people.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #50 Posted: Thu Dec 05, 2019 4:00 pm 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
So the original page was awful and the new page is disastrous. That's not really encouraging. I'll just do my thing with LZ then, not share the insights on SL and let it be as awful as it was.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #51 Posted: Thu Dec 05, 2019 4:43 pm 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Kirby wrote:
I suppose there may be another school of thought which says, "Even if an example isn't optimal, it's good if it's showing a principle we want to learn". In that case, I suppose any LZ corrections should be omitted if we think we are detracting from the theme of the page..


I don't dig that school of thought. SL has plenty of pages which are still under the assumption an early 3-3 invasion is bad and which feature the hane-connect in the resulting pattern. Any player who started Go since the AI revolution will have seen bots and pros play early 3-3 invasions and omit the hane connect. The Korean pro go teacher whose name I forgot, talks about these things on her youtube channel. But we should refrain from updating any of the old material because it would detract from the theme of the page?

I know my sample page was less clearly defined than that, but the mechanics underneath are the same. There is some wisdom passed on to us mortals, we notice the bots deviate significantly ... and then we should just be quiet about it and let that wisdom be.

Like I said, if the results of coloring conventional wisdom with bot insights are desastrous, then I'll keep the insights to myself but that's just not the nature of this beast.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #52 Posted: Thu Dec 05, 2019 6:51 pm 
Gosei

Posts: 1590
Liked others: 886
Was liked: 527
Rank: AGA 3k Fox 3d
GD Posts: 61
KGS: dfan
Knotwilg wrote:
So the original page was awful and the new page is disastrous. That's not really encouraging. I'll just do my thing with LZ then, not share the insights on SL and let it be as awful as it was.

As someone who is dan-level in their skill at SL and L19, you should have acquired the ability not to let John get under your skin by now (or maybe one needs to be an L19 pro for that!). I thought the update improved the page, for what it's worth.


This post by dfan was liked by: Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #53 Posted: Thu Dec 05, 2019 8:40 pm 
Lives in sente

Posts: 757
Liked others: 114
Was liked: 916
Rank: maybe 2d
Some thoughts:

* I am sympathetic towards the idea that if you're going to teach an example application of a principle, you should make reasonable effort to ensure that the example is close to correct, as best as you can tell. If we can literally find no example that both illustrates the principle and is correct - i.e. if the principle is one that never gives a move you believe is correct - maybe we shouldn't be teaching that principle in the first place.

* I don't hold this view absolutely - there's also a place for teaching moves that you believe are suboptimal, but are "simpler" or "educational". For example, this happens occasionally on some of the beginner lectures on Guo Juan's teaching site. But even there, as a dan player I see a lot of care to still try to make many examples correct, not merely "correct if you are weaker than some dan/high-dan/pro, and incorrect if you are that level or stronger".

* If a Sensei's Library page was teaching something that we now truthfully believe to be quite wrong (and not just in some slight way or some arguable detail), it is useful information to indicate that this is the case. So I like what Knotwilg has been experimenting with.

* I do think that one wants to be unobtrusive if possible. Ideally one would just find a better example entirely, but barring that, it would be sufficient to just note that there might be something incorrect or deeper to think about with a little bit of analysis. Analysis that tries to present some digestable ideas is of course better than analysis that just gives sequences or moves.

Actually - I don't see why the bots actually change all that much here. It seems to me this discussion wouldn't be much different if we were discussing a page that, say a kyu player put up to try to teach beginners, such that stronger dan players all agreed the examples of the page had major flaws. Would it? Surely this has happened at least once before, before strong bots existed? And if so, what I imagine happened in such cases is that one of those dan players mentioned it and left some useful analysis if the example was sufficiently wrong given its audience (but usually not if it was only slightly inaccurate or in only in a noncentral way), and either the example would remain - still usefully teaching its original idea, but with the deeper analysis also there to help balance it - or the example would be retracted or replaced, depending on the situation. Seems fine to me.


This post by lightvector was liked by 2 people: ez4u, Gomoto
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #54 Posted: Fri Dec 06, 2019 1:47 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
lightvector wrote:
Some thoughts:

* I am sympathetic towards the idea that if you're going to teach an example application of a principle, you should make reasonable effort to ensure that the example is close to correct, as best as you can tell. If we can literally find no example that both illustrates the principle and is correct - i.e. if the principle is one that never gives a move you believe is correct - maybe we shouldn't be teaching that principle in the first place.


The principle of getting the last oba (big play of the opening), at first glance, is plausible. But noone has ever proved it. In fact, it took Berlekamp and Wolfe in 1994 to prove that the principle of getting last play of the game is correct, but only if that play gains 1 pt. or more. We now understand that the value of getting the last play, at any stage of the game, depends upon the subsequent drop in the temperature of the board. I say We, but I doubt if most pros understand that.

Since the middle game features fights that can reach high temperatures, the idea that the last oba precedes a temperature drop is highly questionable. Again, nobody has proved anything one way or the other. Or even made a plausible argument that getting the last big point of the opening is worthwhile. And there is the counter-argument "Urgent plays before big plays" (Oba yori kyuba). As bridge players might say, you should take big points, except when you shouldn't. ;) The bots in particular think that one class of big plays, those on the side, are not only generally worth less than other big plays, such as occupying an empty corner, they are worth less than plays that have not previously struck humans as urgent. (And the last big play is very likely to be one of the second tier big plays on the side.) There is at least one book here, and books will be written, but it will take at least a few years before we see one, I think.

Edit: It seems to me that LZ's choice of plays on the SL page that sparked this discussion is a good example of urgent plays before big plays. :)

Quote:
* I don't hold this view absolutely - there's also a place for teaching moves that you believe are suboptimal, but are "simpler" or "educational". For example, this happens occasionally on some of the beginner lectures on Guo Juan's teaching site. But even there, as a dan player I see a lot of care to still try to make many examples correct, not merely "correct if you are weaker than some dan/high-dan/pro, and incorrect if you are that level or stronger".


This is a bit of a conundrum when teaching. If a student thinks that 2 + 2 = 6 and revises his opinion to 2 + 2 = 5, psychology tells us to praise the improvement. But you still teach the class that 2 + 2 = 4. Go has a lot of heuristics, and it is best to own up to that fact. For instance, the largest play is best in the vast majority of cases, but not always. A lot of beginners are not taught the qualification. IMHO, that's a mistake. Not that you should teach them exceptions that are over their head, but let them know that that rule is not gospel. SL is advanced enough to handle the exceptions, and the bots are strong enough to show us exceptions to human thinking, and perhaps to dispel mistaken ideas.

Edited for correctness.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.


Last edited by Bill Spight on Fri Dec 06, 2019 5:50 am, edited 1 time in total.

This post by Bill Spight was liked by 2 people: Gomoto, Knotwilg
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #55 Posted: Fri Dec 06, 2019 2:28 am 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
Thanks Bill. If we zoom in on the discussion "Is taking the last big point in the opening a correct principle", which is the concrete case of the meta-discussion "should we color SL advice with LZ analysis, even if the advice is given by pros", then I want to come back on the definition of "big point" and "last (in the opening)". I still haven't seen a precise definition of either, so a fortiori the advice is not very helpful, maybe wrong.

A precise definition could be:
"the biggest point (in the opening) is one that is played in the largest open space between two existing stones" (which is one of dwyrin's heuristics)
and
"if you can't play a stone anymore so that next it can make a two space extension (or establish itself in the corner), then there are no more big points in the opening".

So the last oba would then be the one before this state.

I know this description is not generally accepted, if only because of 3-3 invasions (hence the brackets in the second definition). But at least it would make the heuristic clear and we would have something to disprove.

Edit: I'll make a separate thread for "take the last big point"

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #56 Posted: Fri Dec 06, 2019 4:12 am 
Lives with ko

Posts: 131
Liked others: 1
Was liked: 18
Rank: AGA 4 dan
Universal go server handle: telegraphgo
Bill Spight wrote:

The principle of getting the last oba (big play of the opening), at first glance, is plausible. But noone has ever proved it. In fact, it took Berlekamp and Wolfe in 1994 to prove that the principle of getting last play of the game is correct, but only if that play gains 1 pt. or more. We now understand that the value of getting the last play, at any stage of the game, depends upon the subsequent drop in the temperature of the board. I say We, but I doubt if most pros understand that.


If the greatest AI in the world somehow spoke its first word and said that the key to its dominance was super high-level oba, I guess I wouldn't be completely flat-footed. But I would still be very tempted to argue that the principle is pointless. I think humans can always get a better answer if they apply another, more consistent principle. Think of it in cases:

If every area is equal value, the principle doesn't apply.

If there's one obviously extra valuable area of the board, we don't need to tell ourselves that the temperature is about to decrease to convince ourselves that it's the most efficient area to play. It's the place to play by definition.

If there's two obviously valuable areas of the board, then we don't need to say anything about decreasing temperature to try to get the most out of them before moving on. The principle tells us we should not be bothered to lose sente for the best result here, because the next play, the biggest tenuki, is smaller. That, to me, is needlessly complicated. Surely everyone who would consider this in a game can estimate the value of tenuki themselves? The value of taking sente, gote, and biggest tenuki are what dictate a player's decision between sente and gote variations - no need to build the size of biggest tenuki being smaller into the principle.

If there's more than 2 particularly valuable areas on the board, then playing with an eye towards all of them resolving is beyond my thought process, any thought process I can project onto AI, and any thought process I've ever heard. It's not (prove me wrong if you can) humanly possible to perfectly project that far, because your opponent is practically guaranteed an alternate move somewhere. Perhaps friends who have played each other many times before can think like this - but probes are a much better way for someone who knows exactly how their opponent will play to take advantage.

So why bother with the principle? I can't think of a single game I've seen where the only way to convince myself of a move was related to this at all. It's another way to think, but not a very powerful one, as far as I can tell. So why bother?

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #57 Posted: Fri Dec 06, 2019 4:25 am 
Lives in gote

Posts: 580
Location: Adelaide, South Australia
Liked others: 207
Was liked: 264
Rank: Australian 2 dan
GD Posts: 200
Knotwilg wrote:
I have performed a master edit and a rename of the SL page leading to this discussion. Please check if you find this rendition satisfactory.

https://senseis.xmp.net/?TedomariExercise1

John Fairbairn wrote:
I think the result is disastrous...the LZ move does NOT answer the question posed as the theme of the page: How to get the last big point. LZ is just answering the question: what is the best move in this position?

And that nicely sums up why renaming the page was a good idea!

Knotwilg is navigating some treacherous waters here. I had no idea quite how treacherous when I innocently posted "let's try to improve this instead of removing it". We haven't arrived at an answer that's going to please everyone, but I think the edits and conversations so far represent a constructive first few steps.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #58 Posted: Fri Dec 06, 2019 5:25 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
A precise definition could be:


I think we create a minefield for ourselves as soon as we use the word 'precise."

In general, I have a strong aversion to arguments based on logic, not because logic is bad but because the destination reached by logical steps can vary enormously depending on where you start. The starting point has to be agreed before you can apply the logic sensibly. So I can easily understand why one wants to start with a 'precise' definition.

But the problem here is whether that is a pursuable goal. It is further complicated by the fact that we want here to link it with tedomari and that may also need a precise definition - and that extra goal may or may not be pursuable.

I don't know whether either goal of a precise definition is achievable, but what I do know is that we are starting from a very murky position. I know you don't like me talking about Japanese terms, but that is where 'big point' and 'last big move' come from, and so - if you stick with those terms - you have to have some feel of what raw material consists of.

(Parenthetically, I think we should start afresh with completely new Japanese-free terms based on what AI shows us, but that is probably too far away to sustain our interest at present.)

Oba is a go technical term. It is not at all a common word in ordinary Japanese. Morohashi (a 13-volume dictionary) doesn't even list it. Kojien gives some exotic uses but starts off by giving the sense as "wide place" (okii tokoro) (NB no mention of 'big'). Some Japanese (especially in go) see the adjectival sense as 'important' (NB not 'big'). Some Japanese are not even sure how to read it and say odokoro. Some Japanese go players claim there is a difference between oba and odokoro, the latter having a wider (sic) range.

So we start off with a situation where it vaguely means wide or important to a Japanese but big to English speakers, with different connotations in each case.

And that's only the first word in 'big point.' There's a major problem with 'point', too, since for many that will suggest a precise intersection. It should be area or place.

Since oba is a technical term, though, we might expect a definition. Yet even the great Kido editor Hayashi Yutaka ducked that, and instead quoted Segoe Kensaku (who wrote in the days when pros wrote their own stuff).

This is what Segoe said (I use the term 'big point' because we are stuck with it, not because I think it's accurate):

1. Big points occur in the early opening and are then often the vantage points on the four sides.
2. Big points often mean extending to the limit – that is, five spaces – along the side.
3. In cases where friendly and enemy corner positions face each other, the central points between them on the side are often big points.
4. In cases where there are friendly spheres of influence in two adjacent corners, the central points between them are often big points.
5. An extension, even if it is a narrow one, which prevents expansion by the opponent and/or creates a weak, floating group may be superior to a big point.
6. When invading the opponent’s sphere of influence, try to choose moves that allow two-space extensions to left and right, or above and below.

Note that he uses locutions such as "often" and there is a general woolliness about the whole list. Other writers add to this woolliness by claiming that an oba has to be on the 3rd or 4th lines. There are others who like to use the term yoten (key point or vantage point), which is often preferred when a precise point for both players applies. Yet others like to throw in the historical phrase Tennozan, which adds quite a few connotations for Japanese people. Furthermore, for some Japanese writers these are all terms for use in the fuseki. Other writers freely use them late in the game (e.g. in Shuei's Games the commented occurrences occur on move 30, 36, 60, 86, 60, 72, 9, 17, 25, 25, 20, 99, 46, 112, 11, 38, 7, 107, 63, 19, 8, 13 and 16).

And all that is before we even get onto tedomari, which is equally woolly and (in my opinion) rather more common in English in Japanese because of much greater interest in counting and the endgame than you see in Japanese. An extra dimension we need to assess there, therefore, is whether that extra interest is actually justified. Bill's comments seem to suggest it really is questionable, which I imagine will come as a shock to many :)

For those who do wish to pursue a definition of tedomari, however, I will mention that (a) it can have the meaning of simply 'the final move or moves in the endgame' and (b) one common definition, as in a Nihon Ki-in dictionary, in Japanese is "a good point [ten] or odokoro which is left over." There is no mention of oba directly (and that Nihon Ki-in dictionary is one of the sources that distinguishes odokoro as having a wider sense than oba). Again, all the associations and connotations change depending on where you start from.

Some people enjoy tinkering and will want to try to sort that mess out. I personally don't agree with that approach when the situation is as messy as it is here. Others (and I'm one) will believe the best approach is to rip out all the old rotten woodwork and start again.

As a thought experiment, let us start with a new term for oba. Let me propose 'major area'. That sounds awful but (a) it captures the main meanings and (b) as a slightly unusual phrase it is not yet bound up with existing associations - and in particular it gets rid of the 'size' connotation.

If we then look at the SL pages in that 'corrected' light, the theme of the page for the tedomari example would become something like "how to get the last major area." It still wouldn't be a good page because it still lacks the "how to" element, but then Dieter's LZ counter-example would suddenly makes sense (the 3-3 invasion is in a 'major area'). It doesn't actually help the example page itself, of course. It would just show it is plain wrong and so should be deleted (or perhaps left as an example of human pro fallibility).

But that would entrain going back into higher-level pages in SL and re-writing them (perhaps too problematical on Japanese-based pages).

A better approach, in my view, is to start afresh with brand-new English-based terms and to derive definitions based on AI examples. It may be that that is not practical yet, but I think Gomoto's technique could be promising. AS far as I can see, most people here (including me) have concentrated on looking at whole-board, whole-game positions but he seems be the only one that has reported looking at some common local shapes, and to have already found some significant human/AI differences. That may not yet lead us to a new definition to replace oba or tedomari, but it may be the way to get a handle on the way to proceed.


This post by John Fairbairn was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #59 Posted: Fri Dec 06, 2019 6:09 am 
Oza
User avatar

Posts: 2408
Location: Ghent, Belgium
Liked others: 359
Was liked: 1019
Rank: KGS 2d OGS 1d Fox 4d
KGS: Artevelde
OGS: Knotwilg
Online playing schedule: UTC 18:00 - 22:00
John Fairbairn wrote:

And that's only the first word in 'big point.' There's a major problem with 'point', too, since for many that will suggest a precise intersection. It should be area or place.


Agreed.

John Fairbairn wrote:

This is what Segoe said (I use the term 'big point' because we are stuck with it, not because I think it's accurate):

1. Big points occur in the early opening and are then often the vantage points on the four sides.
2. Big points often mean extending to the limit – that is, five spaces – along the side.
3. In cases where friendly and enemy corner positions face each other, the central points between them on the side are often big points.
4. In cases where there are friendly spheres of influence in two adjacent corners, the central points between them are often big points.
5. An extension, even if it is a narrow one, which prevents expansion by the opponent and/or creates a weak, floating group may be superior to a big point.
6. When invading the opponent’s sphere of influence, try to choose moves that allow two-space extensions to left and right, or above and below.



Having Segoe's "definition" of big point definitely helps.

John Fairbairn wrote:
Note that he uses locutions such as "often" and there is a general woolliness about the whole list.


It's the wooliness that puts me off. It occurs to me that a lot of the existing advice is afraid to be what I call precise, because one can't be precise in defining the next best move, because the game isn't solved. The solution is not to have imprecise definitions.

The way I'd like to solve this is to have precise defintions, like one for "big point", and then make statements that "playing the big point in the opening is often good". How often? Not sure. But it's a heuristic. When you wonder what to do in the opening, then think of playing a big point. That requires "big point" to be precise, otherwise the heuristic becomes "when ..., play a good move".

John Fairbairn wrote:
If we then look at the SL pages in that 'corrected' light, the theme of the page for the tedomari example would become something like "how to get the last major area." It still wouldn't be a good page because it still lacks the "how to" element, but then Dieter's LZ counter-example would suddenly makes sense (the 3-3 invasion is in a 'major area'). It doesn't actually help the example page itself, of course. It would just show it is plain wrong and so should be deleted (or perhaps left as an example of human pro fallibility).


OK, that helps me in the sense that you think the effort was well intended but didn't help (you) and you prefer deletion or marking rather than tentative and failed improvement.

John Fairbairn wrote:
A better approach, in my view, is to start afresh with brand-new English-based terms and to derive definitions based on AI examples.


That's an exciting proposal. As you know from previous discussions, I'd rather talk about English terms because the Japanese terms is laden with linguistics. So I'm not happy that I renamed "how to take the last big point" to "tedomari example 1" and I will try a better course, but it's one that requires a lot of energy.

Top
 Profile  
 
Offline
 Post subject: Re: reviewing SL articles using LZ and criticism
Post #60 Posted: Fri Dec 06, 2019 6:27 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Knotwilg wrote:
Thanks Bill. If we zoom in on the discussion "Is taking the last big point in the opening a correct principle", which is the concrete case of the meta-discussion "should we color SL advice with LZ analysis, even if the advice is given by pros", then I want to come back on the definition of "big point" and "last (in the opening)". I still haven't seen a precise definition of either, so a fortiori the advice is not very helpful, maybe wrong.


I think that last has a clear English meaning. As John Fairbairn points out, it is a mistake to look for a precise definition of oba. IMHO, that is in general true for traditional go terms, because they have been around so long that they have acquired the fuzziness of informal natural language, even if they started out with a limited technical meaning.

Quote:
A precise definition could be:
"the biggest point (in the opening) is one that is played in the largest open space between two existing stones"


I know that's pretty much what I said, based upon my limited observation of usage, but, except for what JF gives below, I think that the SL page gives the best English definition of big point in the opening. After our exchange I took a look at the SL page, and discovered that Bob Myers apparently based its contents on the Japanese Wikipedia page on oba.

Edit: Having looked again at JF's post, which quotes Segoe, it appears that the Japanese meaning of oba has changed since then, to include more places. That may be in large part because of Ishida Yoshio "The computer", who is referenced on the Japanese Wikipedia page.

Quote:
"if you can't play a stone anymore so that next it can make a two space extension (or establish itself in the corner), then there are no more big points in the opening".

So the last oba would then be the one before this state.


That's a good heuristic. Sometimes an open corner remains unoccupied until a good bit of middle game fighting has occurred. The corner was the last big place to play in the opening, but in such cases playing there was apparently not as important as the urgent plays in the fight.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.


Last edited by Bill Spight on Fri Dec 06, 2019 8:04 am, edited 2 times in total.
Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 113 posts ]  Go to page Previous  1, 2, 3, 4, 5, 6  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group