It is currently Thu Mar 28, 2024 3:13 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 44 posts ]  Go to page 1, 2, 3  Next
Author Message
Offline
 Post subject: Follow the wise old men - and solve problems faster
Post #1 Posted: Sun Aug 09, 2020 3:44 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
OC, reading to the end makes use of depth first calculation, and trying other moves emphasizes exploration. The idea is not just to solve a problem, but to understand it thoroughly.


I just read this in another tsumego thread by Bill in which he looks at the use of iterative deepening in solving problems. My vague reaction was somewhat along the lines of "Yeah, good point" and then I quickly read on. I think most people read forum stuff that slapdash way. We understand all the words, the idea expressed is in line with our own knowledge/prejudices or whatever - let's get down to the interesting stuff.

For me the real interest was in the mention of iterative deepening. That was because I clearly remember reading, very many years ago, an article about Newell and Simon's discovery that this technique worked wonders in chess programming. As I recall, there were two main benefits. One was that each iteration allowed an ordering of the moves so far, which helped enormously with pruning the alpha-beta tree efficiently. This was very counter-intuitive at the time. How could the huge amount of time to do a complete search at a low ply number with the hardware available in that era possibly be justified?

The other benefit, probably rendered nugatory now by modern hardware, was that if the computer was interrupted by the chess time control before it could search very deeply, it had at least looked at every move at a low ply and so could make a half-decent move without overlooking obvious (to us) little traps.

Anyway, that was enough for me to make a mental note to follow that thread.

Then, purely by chance, I turned next to my computer files and started doing a little bit of housekeeping - putting several files in a new directory. In the course of that I saw a file I had marked "Igo Tsumekata". It meant nothing to me until I opened it, and what I saw was the image below. And that got me thinking.

Attachment:
Capture.GIF
Capture.GIF [ 77.96 KiB | Viewed 12163 times ]


The first thing I did was to recall Bill's post, but the bit that re-surfaced was not the glittering bauble of iterative deepening. It was mainly the humdrum word "understanding."

This image is the beginning of a book by Kobayashi Tetsujiro, which I had taken a mild interest in long ago because Kobayashi was a rival of Honinbo Shuei. He was somebody Shuei did not treat very well. I therefore always had a soft spot for K.

This illustrates what was for a very, very long time the usual way to present tsumego collections. It's what appears in ancient Chinese classics such as Xuanxuan Qijing and Guanzipu and goes back over a millennium. It was also retained in Japan in the early 19th century, an era when go publishing bloomed and many presentational innovations were tried. Tsumego essentially resisted these innovations until well into the 20th century. There were some apparent nods towards novelty, such as the diagram on the right, which shows the notation at use in the book, but even that was more common-sense than innovation, since there were other notations in use at the time (e.g. the iroha notation).

I had always assumed this was because publishing was then an expensive undertaking and the idea was to save on paper. The fact that several problems appear on the same whole-board diagram supports that. The fact that the solutions are scribbled in the margins of the same page also seemed to support that. That could arguably be seen as an innovation, as a typical approach in older books was to put the answers on the same board as the problems. But the ancients did have recourse to the margins for text notes on variations. That required the use of a notation, and the same idea in the image above can bee seen even in the ancient Carefree & Innocent Pastime - our oldest diagrammed manual.

Yet I have always had a niggling doubt that survival of this practice really was just a matter of publishers' parsimony. Thinking about Bill's post is, however, the first time I have let that doubt have some wiggle room.

While I was marshalling my thoughts, the problem in the lower right caught my eye (Black to play). I think the reason was some unconscious thought along the lines of "why is such an easy problem in this book?"

Needless to say there was more than a touch of hubris there. I could do the problem, and I could even identify a couple of themes, so I could even claim I "understood" it. More hubris.

But I didn't really know the answer as I still had to check my so-called understanding. (I only knew how to get the answer.) I therefore had to read the text at the top. That was my Damascene revelation.

If the problem been presented in a modern book, I would have just turned the page, looked at an answer diagram, and said, "Yup, I've got it." And moved on.

But in this case I had to do two things I would not normally do. One was to convert the notation to a version I am more familiar with (e.g. 3-2 to S3) and the other was to visualise the moves for myself on the board. It was in the painfully slow (but therefore valuable) course of doing that that I realised my understanding was defective. I don't mean wrong. In fact what I understood was 100% right. It was defective because I had seen the themes as separate items in a list. Lists are usually a bad way of thinking. This was another example. I had to learn to see the elements as part of an organic whole. Or, as Bill says, "understand it thoroughly."

Still, I think the most valuable lesson for me was how powerful it is to be led move by move so as to visualise the solution on the bare board. if you think about it, this is 100% efficient training because you are not just practising visualisation - you are also following the route of the correct solution, thus reinforcing your intuition in the optimum way.

This was of special interest to me because it's essentially the way I have been presenting my most recent books in Go Wisdom format. There I have been concerned with complete games rather than problems, but I have likewise eschewed variation diagrams in favour of describing the variations within the text, either in coordinate form as here or by giving a detailed clue as to the outcome (e.g. it ends as a capturing race that Black wins by one move). I am forcing the reader to actually do some study work. In other words, I am being cruel to be kind.

I am inclining now towards believing that the old method of presenting tsumego problems was actually the best. I am even minded, therefore, to maybe publish Kobayashi's little book in precisely that format - answers in coordinates - so that other people can try out the technique for themselves.

If you want to try it at once, the solution to the problem above is: S2, R3, S3, Q4, S6, S7, S5.

You will see that this does not solve the problem down to the bitter end. But that was the aspect that led me to realise I had a poor (inorganic) understanding. It was thinking about why the author stopped his solution there that led me to framing the problem - or, rather, both problem and solution - in a more organic way. Although my thoughts are continuing and may change, at present I believe this "framing" is actually a human version of iterative deepening - it is ordering my thoughts in a way that makes the deeper search much, much more efficient.


This post by John Fairbairn was liked by 4 people: Bill Spight, gowan, Harleqin, SoDesuNe
Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #2 Posted: Sun Aug 09, 2020 8:57 am 
Gosei

Posts: 1590
Liked others: 886
Was liked: 527
Rank: AGA 3k Fox 3d
GD Posts: 61
KGS: dfan
John Fairbairn wrote:
For me the real interest was in the mention of iterative deepening. That was because I clearly remember reading, very many years ago, an article about Newell and Simon's discovery that this technique worked wonders in chess programming. As I recall, there were two main benefits. One was that each iteration allowed an ordering of the moves so far, which helped enormously with pruning the alpha-beta tree efficiently. This was very counter-intuitive at the time. How could the huge amount of time to do a complete search at a low ply number with the hardware available in that era possibly be justified?

The other benefit, probably rendered nugatory now by modern hardware, was that if the computer was interrupted by the chess time control before it could search very deeply, it had at least looked at every move at a low ply and so could make a half-decent move without overlooking obvious (to us) little traps.

These are both good points. A third is that (perhaps paradoxically) since the work required to search increases exponentially by number of ply, it means that if you're going to search to n ply, you haven't really wasted much time if you search to n-1 ply first!

Quote:
But in this case I had to do two things I would not normally do. One was to convert the notation to a version I am more familiar with (e.g. 3-2 to S3) and the other was to visualise the moves for myself on the board. It was in the painfully slow (but therefore valuable) course of doing that that I realised my understanding was defective. I don't mean wrong. In fact what I understood was 100% right. It was defective because I had seen the themes as separate items in a list. Lists are usually a bad way of thinking. This was another example. I had to learn to see the elements as part of an organic whole. Or, as Bill says, "understand it thoroughly."

Still, I think the most valuable lesson for me was how powerful it is to be led move by move so as to visualise the solution on the bare board. if you think about it, this is 100% efficient training because you are not just practising visualisation - you are also following the route of the correct solution, thus reinforcing your intuition in the optimum way.

This was of special interest to me because it's essentially the way I have been presenting my most recent books in Go Wisdom format. There I have been concerned with complete games rather than problems, but I have likewise eschewed variation diagrams in favour of describing the variations within the text, either in coordinate form as here or by giving a detailed clue as to the outcome (e.g. it ends as a capturing race that Black wins by one move). I am forcing the reader to actually do some study work. In other words, I am being cruel to be kind.

I am inclining now towards believing that the old method of presenting tsumego problems was actually the best. I am even minded, therefore, to maybe publish Kobayashi's little book in precisely that format - answers in coordinates - so that other people can try out the technique for themselves.

I've said this previously on this site, but having come to go to chess, I have always thought that it was a little unfortunate that go variations are presented by means of a diagram with the final position on it, with the moves taken to reach that position labeled with numbers. I feel that it makes it to easy to use the diagram as a crutch and avoid the hard work of visualizing a variation move by move, which is what you have to do when you're actually sitting at the board playing. In contrast, chess variations are presented purely textually, and it's your job as a reader to actually apply them in your head to the current position and imagine how the board state evolved. To me this is a better pedagogical way of presenting variations.

Of course every good chess player has internalized the coordinates and characteristics of every square on the board, so reading a textual variation has a lot of automatic semantic content (e.g., if I read "1.Nxf7 Rxf7 2.Bxe6", I immediately know what sort of position led to that sequence and what White's usual followup ideas are), while go coordinates are a lot more abstract for multiple reasons, such as the large board size and presence of symmetries (people don't really think about the characteristics of the S3 point in isolation), so go variations presented as lists of coordinates are necessarily harder to understand fluidly. I do like it when sequences can be described with terminology ("attach, hane, extend, block") but that doesn't always have sufficient precision.


This post by dfan was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #3 Posted: Sun Aug 09, 2020 10:24 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
In contrast, chess variations are presented purely textually, and it's your job as a reader to actually apply them in your head to the current position and imagine how the board state evolved. To me this is a better pedagogical way of presenting variations.


Thank you for reminding me of something I, as a quondam chess player, ought to have remembered. But remembering it now convinces me even more I'm on the right lines!

Quote:
Of course every good chess player has internalized the coordinates and characteristics of every square on the board, so reading a textual variation has a lot of automatic semantic content.


But this is even better. It's something that instantly makes perfect sense to me but I'm not at all sure I would have thought to consider it unprompted. However, considering it now, I think it's what is behind what I said about "framing" a problem (question plus answer) as an organic whole. There was a famous experiment (by de Groot?) in which pros and amas were asked to reconstruct from memory various chess positions shown briefly. The pros did very well and the amas poorly on positions taken from real games. With randomly generated positions both groups were like fish out of water.

I am assuming that what you mean by automatic semantic content is that when you say "h7" to a good player, he doesn't conjure up am image of an empty board with a diode flashing on that square. Rather, he simply locates it in the northeast corner of the board but bells and whistles start going off in his brain if the enemy king is castle on that side, if the h7 pawn has been pushed, if you have a queen with a ticket to ride, etc etc. I suspect there's much more work involved in creating these little biospheric bubbles mentally for go problems but, as you say, the ability to describe the flow of play by using specific terms. I added quite a few specific to tsumego (i.e. bells and whistles signalling special shapes) in the original version to Gateway To All Marvels, and I added a few more in the enhanced paper edition I am now working on. It's ready for proof reading, but I think now it might be a good idea to do a bit of re-writing as well to take account of the important point you raised.


This post by John Fairbairn was liked by 2 people: Bill Spight, gowan
Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #4 Posted: Sun Aug 09, 2020 11:52 am 
Gosei

Posts: 1590
Liked others: 886
Was liked: 527
Rank: AGA 3k Fox 3d
GD Posts: 61
KGS: dfan
John Fairbairn wrote:
Quote:
Of course every good chess player has internalized the coordinates and characteristics of every square on the board, so reading a textual variation has a lot of automatic semantic content.

But this is even better. It's something that instantly makes perfect sense to me but I'm not at all sure I would have thought to consider it unprompted. However, considering it now, I think it's what is behind what I said about "framing" a problem (question plus answer) as an organic whole. There was a famous experiment (by de Groot?) in which pros and amas were asked to reconstruct from memory various chess positions shown briefly. The pros did very well and the amas poorly on positions taken from real games. With randomly generated positions both groups were like fish out of water.

I feel like that is a related concept ("chunking" a lot of information into a small number of big pieces) but not quite the same thing. After all, you can do the same thing with go positions where an experienced player can glance at a board and identify without thinking not only what has already happened in various areas but what the likely continuations are.

Quote:
I am assuming that what you mean by automatic semantic content is that when you say "h7" to a good player, he doesn't conjure up am image of an empty board with a diode flashing on that square. Rather, he simply locates it in the northeast corner of the board but bells and whistles start going off in his brain if the enemy king is castle on that side, if the h7 pawn has been pushed, if you have a queen with a ticket to ride, etc etc.

Right, exactly. When I think about h7 I think about a White bishop on d3 pointing at it after Black has castled kingside, and maybe about a White knight on g5 and/or a White queen on h5 if things are really rolling. Of course, Black probably has a knight on f6 defending it in addition to his king on g8. And so forth. And I have stories like this about the other 63 squares as well.

There are stories of very strong players being completely at sea when they are asked to consider a standard opening position that has been either been flipped left-to-right or had the colors of the pieces reversed.

Quote:
I suspect there's much more work involved in creating these little biospheric bubbles mentally for go problems but, as you say, the ability to describe the flow of play by using specific terms. I added quite a few specific to tsumego (i.e. bells and whistles signalling special shapes) in the original version to Gateway To All Marvels, and I added a few more in the enhanced paper edition I am now working on. It's ready for proof reading, but I think now it might be a good idea to do a bit of re-writing as well to take account of the important point you raised.

I'm glad that my comments triggered some ideas for you!

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #5 Posted: Tue Dec 15, 2020 6:05 am 
Beginner

Posts: 1
Liked others: 0
Was liked: 0
Rank: 11k
Hi John,
I greatly appreciated your post and I would like to offer a book reference for chess which may be useful for your (coming) book.

As mentioned, the presentation of a tsumego with a notation (eg S3,T4,...) stopping a step before the complete solution helps visualizing stones and understanding the problem.

This is exactly what is described in Jonathan Tisdall's Improve Your Chess Now ! (1997) in the chapter "Blindfold Chess and Stepping-Stone Diagrams" ("stepping stones" should really be a Go-term).

Tisdall gives a small talk about calculating variations (chapter 1 about the infamous Kotov tree) and mentally seeing the board (chapter 2). To him, seeing is paramount to better calculation. Does it ring a bell ?

Tisdall offers as a training method the reading (the score sheet) of miniatures (or variations) and ask about the reasons of the defeat. You are supposed to get the position fixed in your mind eye at every move. It seems to be very similar to the tsumego you give where the moves are given up to a (supposedly) familiar shape.

For the curious chess player, here is Tisdall exercise about the following first moves :
1.e4 Nf6 2.Nc3 d5 3.e5 d4 4.Ne2 Ne4 5.c3 dxc3
Would you play Qa4+ to catch the knight? (Hint : black has two knights after all)


P.S. : when will the tsumego book be published ?

Cheers

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #6 Posted: Tue Dec 15, 2020 11:49 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Thank you for the Jon Tisdall reference. I used to work with Jon on MSO. Normally he worked from Norway and I was in London, but we met up a few times in the press room at various chess events. He (or maybe it was David Levy) gave me a couple of his books, and it turns out that "Improve Your Chess Now" was one of them. It appears from the bookmark that I never got past the first ten pages, so what you say doesn't ring any bells with me. But I now have a good excuse to try harder, and will.

While looking for it, the first book I encountered was "How to Calculate Chess Tactics" by Valeri Beim, and that sent a chill down my spine. I also still have the mental scars from trying to read the Kotov book many decades ago. But Jon is a journalist, so I expect his book to be readable. The reason I have never got much into it up to would almost certainly be simply because I don't play chess or mix with chess players any more.

Quote:
P.S. : when will the tsumego book be published ?


The book was written some time ago, and I got halfway through proof-reading before being distracted by something else. FWIW the combination of that proof-reading and my books that incorporate the same technique via Go Wisdom (these books being what distracted me) has convinced even more that the "effortful practice" this approach demands is the most reliable way to make progress.

I need to find a way to get distracted back. I've put the draft back on my desk. It's in a red file - a strong enough colour to burn into my conscience, perhaps? But I'll want to read Jon's book first.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #7 Posted: Tue Dec 15, 2020 1:19 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
hxd wrote:
Tisdall gives a small talk about calculating variations (chapter 1 about the infamous Kotov tree) and mentally seeing the board (chapter 2). To him, seeing is paramount to better calculation. Does it ring a bell ?


There's somebody besides me! :cool: :D

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

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #8 Posted: Tue Dec 15, 2020 2:06 pm 
Gosei
User avatar

Posts: 1810
Liked others: 490
Was liked: 365
Rank: KGS 1-dan
dfan wrote:
I've said this previously on this site, but having come to go to chess, I have always thought that it was a little unfortunate that go variations are presented by means of a diagram with the final position on it, with the moves taken to reach that position labeled with numbers. I feel that it makes it to easy to use the diagram as a crutch and avoid the hard work of visualizing a variation move by move, which is what you have to do when you're actually sitting at the board playing. In contrast, chess variations are presented purely textually, and it's your job as a reader to actually apply them in your head to the current position and imagine how the board state evolved. To me this is a better pedagogical way of presenting variations.


I woudn't imply that the means of presentation for chess is necessarily pedagogical better since chess cannot adopt the way go is presented. So it's rather difficult to compare those two. I concur that the way chess is presented needs more work to decipher. But more work is neither necessarily better learning nor pedagogical better. Reminds me of the old argument that comic books stultified the imagination of kids.

Going back to John Fairbairn's example above: I didn't understand the bottom right example and with the a self-imposed time-limit off around a douzen seconds I coudn't solve it (the slapdash way). To check the solution and to be sure where S2 actually starts I needed to count all the lines and remember that some letter is omited for some reason. Then I was going back and forth between the coordinates and the diagram (getting confused with Q4 along the way) and saw the end of the line. Visualisation was never a problem but compared to the common way a solution to a tsuemgo is shown I needed a lot more time.

With more practice I would certainly remember the coordinates of the go board and improve the time needed to check an answer. But isn't that all I've learnt then? Go board coordinates?

Now granted I'm - to a certain degree - able to visualise go positions already. A beginner who tried to solve a problem and then check the answer might have a harder time. Let's assume they can decipher coordinates easily: They would have to read through several moves to try on their own, maybe failing to do so and then having to visualise the book's solution to check. Double practice, twice the learning effect?

Going back a bit again: We already established that experienced chess players can't remember a random position better than amateuer players. So although they saw on which coordinates all the piece were placed, they coudn't visualise the position after the pieces were removed. Therefor I guess coordinates and visualisation don't really have a connection when you remember a position. I assume shapes and formations make you remember a position - as it does in go. Coordinates just do the talking comparable to go players saying "hane, crosscut might have been an overplay".

So our beginner checking the coordinates enriched solution, what does they actually do? Knowing chess players they will have stored the answer already in the form of S2, R3, S3, Q4 and so on. Seeing the answer should not require any more visualisation then. If the solution deviates from their attempt they will have to do some reading again to figure out why the solutions works. Not much change to the way I solve tsumegos right now, the difference being I didn't have to learn coordinates beforehand.

So my take: Chess annotation is what it is, not because it's pedagogical sound but because there is no better way so far. Learning the annotation is primarily a means of communciation and has in my opinion no impact on your chess performance. Therefor: causing friction between attempting a problem and being able to decipher the solution has no benefit when it comes to reading skill. It surely is more hard work but only results in more time spent. But I do see the point when it comes going over games. If you have a game discussion with around ten moves per diagram you can indeed be lulled into a "painting by numbers" kind of mentality (I would argue against it though but I guess I derailed enough : )). Go with all moves in one diagram then, a lot to learn regarding direction of play!

Since variation diagrams in a game discussion were another point raised in this thread: Here I'm all for using eg letters in the main diagram or just hinting at it in the text. Though the main point here - for me - is not offering effortful practice (certainly a benefit though) but to not distract from the game. Game discussions with a lot of variation diagrams are equally aweful to the way chess handles variations:

(A "simple" four-mover)
Move Move
Move (Move Move (... Move, Move! Move, Move#), Move Move (... Move!, Move Move!! (... Move??, Move#)), Move Move, Move Move+- (... Move, Move? Move!)) Move#

_________________
My "guide" to become stronger in Go

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #9 Posted: Tue Dec 15, 2020 3:36 pm 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
Visualisation was never a problem but compared to the common way a solution to a tsuemgo is shown I needed a lot more time.


This sentence leads me to infer that you are using visualisation in a different way from me.

Let's imagine you are walking to a new destination - a church on yon side of a brae. But you can see the spires of two churches. You have to choose and you don't want to get lost.

You can spend a lot of time memorising every bush and static landmark so that you can feel confident that you have covered every eventuality. You could argue that you have visualised the terrain. You then flip a coin, set out and accept that if Church A is the wrong one you can go B afterwards.

Alternatively, you can look not at the terrain but at the dynamic features: there's a path to Church A but not to Church B. Being a human, you go with the probabilities, don't waste time getting acquainted with bushes (unless Nature calls), and set off on the path. You have visualised the journey. That's my way.

For me, the first priority is solving a problem is therefore to spot the likely paths. Maybe we can call that perception or recognition rather than visualisation. Once we have a path to follow the second stage is to walk along it, in our minds. That is the part I am visualising. If I'm successful fine, and coordinates don't come into it. I may of course be unsure or unsuccessful, but the same principles apply. There is a path. There are steps on that path. The solution is sequential and path-like. If I follow those coordinates I will learn what that new path looks like and which steps I need to take. I can then forget all about the coordinates and just play over the solution a few times in the right order so that that new pathway is embedded in my brain. If another problem has a similar path, I have a head start. I have pruned the solution tree even before I have seen the problem.

Obviously, with harder problems there may be more than one likely path and some may be littered with quicksand, locked gates and mad bulls. So I might have to go back. But I can still go back to just another path. I still don't have to get acquainted with every bush.

I wish to stress: the coordinates themselves are not an issue. They are just a symbolic way of representing a sequence of steps in time. Variation diagrams represent sequences in space, which is confusing in itself but is also heinous if their apparent convenience tempts us into not spending the time to learn the steps. How many of us can learn to dance by looking at pictures of dance steps. You have to tread the light fantastic.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #10 Posted: Tue Dec 15, 2020 4:43 pm 
Lives in sente

Posts: 757
Liked others: 114
Was liked: 916
Rank: maybe 2d
dfan wrote:
John Fairbairn wrote:
For me the real interest was in the mention of iterative deepening. That was because I clearly remember reading, very many years ago, an article about Newell and Simon's discovery that this technique worked wonders in chess programming. As I recall, there were two main benefits. One was that each iteration allowed an ordering of the moves so far, which helped enormously with pruning the alpha-beta tree efficiently. This was very counter-intuitive at the time. How could the huge amount of time to do a complete search at a low ply number with the hardware available in that era possibly be justified?

The other benefit, probably rendered nugatory now by modern hardware, was that if the computer was interrupted by the chess time control before it could search very deeply, it had at least looked at every move at a low ply and so could make a half-decent move without overlooking obvious (to us) little traps.

These are both good points. A third is that (perhaps paradoxically) since the work required to search increases exponentially by number of ply, it means that if you're going to search to n ply, you haven't really wasted much time if you search to n-1 ply first!


A small aside about iterative deepening: for most people, it's not really an effective way to read, even if it may be effective for computers. Why? Well, relative to computers (even the modern neural-net MCTS ones) humans relatively speaking rely less heavily on conscious reading and far more heavily on subconscious pruning and intuition. The "n-1 doesn't waste much relative to n" argument becomes pretty flimsy if due to very powerful subconscious pruning, the subtree that you consciously read out ends up having an average branching factor, of, say, 1.25, instead of like 2, or 3 or more. Then, the multiplicative overhead of repeated work in searching n-1, n-2, n-3, becomes intolerable.

The other thing is that for humans, context switching back and forth between variations has a massive overhead compared to a computer - probably measured in many hundredths or few tenths of a second at minimum, plus more overhead for mentally re-playing the moves to reach the position again. So it pays to stick to each variation for longer and deeper before switching, instead of reading each one a little, then each one a little more, and so on. And if you do revisit a variation, to stick it for much more than just one extra move. Imagine for example if in modern-day neural net + MCTS, each time you ran a playout, instead of having all the prior evaluations cached and being able to blaze in an instant all the way down to the frontier of where you last left off, you were forced to uselessly and slowly re-feed every position along the way to the neural net to evaluate again. And again. And again. You can absolutely bet that programmers would have each playout do WAAAY more work going deeper before returning back to try another variation.


This post by lightvector was liked by: dfan
Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #11 Posted: Tue Dec 15, 2020 7:40 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
There is no question that for conscious search humans are built for depth first search. There is also little question that for amateurs to search go positions even moderately deeply, there will be mistakes in the search path. That's a problem.

Breadth first search is not the answer. The human "stack" is too limited.

Best first search seems promising. But again, if we knew the best move, go would be easy. ;) In a recent review I did, Black missed a net on the first play in a certain position. It is not unusual for SDKs to fail to find the correct play because they never even consider it. And they don't find it by depth first search, or even best first search, unless they they are lucky with the backtracking. They can be shown, however. :)

How many times on Sensei's Library would a player ask what was best play in a certain position from one of their games, when the real answer was, Don't get into that position. ;) I.e., the mistake was not what they played in that position, but an earlier play that led to that position. To me this suggests that the player's search strategy was not broad enough.

Iterative deepening is a depth first search strategy that does not miss the best play. OC, it is not likely to be a good strategy in a real game, but it may be a good exercise to help the player understand a position. :)

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

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #12 Posted: Wed Dec 16, 2020 3:50 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
I read the relevant chapters of Jon Tisdall's book last night. May main reaction was, "At last!" A chess grandmaster finally demolished Kotov's approach, and in clinical Joran Peterson fashion using common sense in place of dogma. I had always felt uncomfortable with Kotov. At first I thought it was me. Then I thought it was a combination him and me - he was a numbers man, I'm not. Then I eventually came to the view was wrong that his status came from piggy-backing on the Soviet School of Chess.

There are elements to Kotov that do make sense, and Tisdall stresses his good advice to make a list of candidate moves, for example. But essentially, even if Kotov's method worked for a Soviet tractor it doesn't work for humans.

Making allowances for the differences between go and chess, I'd say that the methods Tisdall and I propose are almost in synch. There are differences - he proposes pausing every few moves to burn a visualised image into the brain; I focus more on the time element. But I suspect these are superficial things and probably not real differences.

There is, however, one discrepancy. Jon says he came to his method as the result of noticing that his own powers of analysis (= reading) were waning with age. His method was therefore recuperative, and worked. But he doesn't appear to say what was different about his (even more natural?) method when he was younger.

I suspect I know wherein lies the answer, though. For that we can turn to Chomsky. One of his less cited many achievements was the observation that children are not actually taught their mother tongue. They are put in an environment where they can imitate and get feedback, but they don't have lessons. More importantly, and even less cited, is his observation that children don't actually learn their language. They grow into it, in the same way they grow in being ale to walk or in reaching puberty.

From this it should follow that ab initio children cannot be taught, in the sense we normally use for language learning. So why do we teach languages by making kids learn lists of vocabulary, paradigms such as amo, amas, amat? Why do grammarians teach Russian to English speakers by telling them that there are two conjugations and lots of irregular verbs, yet Russian native speakers look baffled and say, "There are at least 40 conjugations, and they don't really matter"?

All these unnatural ways of teaching a language are like teaching pure reading in go. They don't work because they are not natural. In contrast, look at the huge explosion in people all round the world learning (mainly) English, fairly well and quickly and certainly up to the Level 3 standard of the US Diplomatic Service. This has come about because people have been swamped with resources via the internet. Because so much is available, there is something for everyone. By choosing something that interests you (e.g. Hikaru no Go if you want to learn Japanese) you end up spending the requisite time on the language. You don't really learn it. You grow into it.

I think the same explanation underpins what is happening in go. When I started to play go, there was not only no internet, but (in England) no books or teachers. DVDs and videos had not even been invented. As a result amateur dan players were a great rarity. Now people have a host of ways of "growing into" the game and dan players are two a penny. Indeed, I would further claim that they are also much stronger than equivalent dan players of yore.

The time element is crucial. The US Diplomatic Service guidelines (which tally with other diplomatic services) measure language-learning requirements in hours. So, for an English speaker you need about 600 hours to learn French to Level 3. It's about 1100 hours for Russian and "hard" languages such as Japanese need 2200 hours, as I recall. Presumably, to reach native standard to need the fabled 10,000 hours plus.

What seems to happen (and it certainly applied to my own language "learning") is that if you spend the necessary hours, your brain takes care of all the conjugations and plurals and so on, without troubling your conscious you in the slightest. It does this in a probabilistic way. It sorts out what is most common, useful and natural; it strengthens the commonest associations. After the necessary X hours you can hold a decent conversation, read a newspaper and watch an episode of Hikaru no Go. But what you can't do (and should never ever need to do) is write down the conjugation of the past historic tense in French for Class II verbs.

So why is go not taught by the immersive method in the west? Why does the hairshirt approach prevail? I think the reason is the prevalence of mathematicians here, and it was interesting to me to watch a recent Michael Redmond video where he made the same point, though more diplomatically than me (yet more tellingly, because he has a mathematical background himself). If you have a strong interest in rules or in go topology or the like, you can immerse yourself in those aspects of go, and of course you will then be inclined to spend the necessary hours in doing the grunt work of studying josekis and tsumego problems. Having fun while learning. But telling people who do not share your fascination with numbers to improve at go by doing the same grunt work as you is not going to work. They are not immersed. They are not having fun.

My own Go Wisdom format is an attempt to provide an immersive experience for other types of players. I therefore usually offer lots of background history and biographical detail, a rich variety of comments from many different pros, and (if possible) material that is new in the west. The GW appendix is meant to be an additional and very amply stocked treasure chest where you are sure to find something that will interest you. The banishment of variation diagrams in favour of inline commentary is my way of encouraging you to do the grunt work in the most effective way (i.e. learning suji), but all that is set within an immersive experience.

I don't know whether I've reached wise old man status yet. I've certainly reached two of those stages, so maybe there's still some hope.


This post by John Fairbairn was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #13 Posted: Wed Dec 16, 2020 4:35 am 
Gosei
User avatar

Posts: 1753
Liked others: 177
Was liked: 491
The immersive method to learn a language works well if you frequently communicate with people who speak the language well. If you only interact within a community of people who make mistakes, you will reproduce these mistakes and teach them to other people.

In the same way, the immersive method can work in go if you frequently play against stronger players. This is not possible for everyone.


This post by jlt was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #14 Posted: Wed Dec 16, 2020 5:21 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
The immersive method to learn a language works well if you frequently communicate with people who speak the language well. If you only interact within a community of people who make mistakes, you will reproduce these mistakes and teach them to other people.


That describes one of many problems with the traditional method of language teaching. So, yes.

But no to the idea that that is the only way to be immersive. Listening to movies, reading, playing games, even going to ethnic restaurants or whatever else interests and thus motivates you are other ways. And may even be much better. I always found sitting in groups and telling each other "My name is X" and "Today it is raining" totally stultifying and more like an AA rehab session than the fun of absorbing a new culture.

And immersion among natives is no guarantee of proficiency anyway. T Mark Hall spent a couple of years at the British Embassy in Tokyo. He was not in the stream that was given language tuition, which meant going to Kamakura for (I think it was) 40 hours a week tuition for two years. As a result Mark knew virtually no Japanese. He wasn't interested in languages and never even went to Kyoto (one of his big regrets of later life). But he was interested in go and immersed himself thoroughly in that there. He became good enough to become British Open champion.

The paradigm is simple: interest >> motivation. Or for some people, ambition >> motivation.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #15 Posted: Wed Dec 16, 2020 5:43 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
jlt wrote:
The immersive method to learn a language works well if you frequently communicate with people who speak the language well. If you only interact within a community of people who make mistakes, you will reproduce these mistakes and teach them to other people.

In the same way, the immersive method can work in go if you frequently play against stronger players. This is not possible for everyone.


My favorite method of go study has always been playing over professional games, a method recommended by Capablanca in chess and Shusai in go. That's a form of immersion.

As for interacting with people who make mistakes, I have warned against weak players playing even games if they wish to advance. That's a good way to pick up and reinforce bad habits. Better to play against stronger players, and then, when they get better, to return the favor and play some against weaker players. OC, if your aim is to have fun, then do as you will. :)

These days you can play against much stronger players in the form of AI. There are many ways to do this. You don't just have to play them even and get crushed by move 23. ;)

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

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #16 Posted: Wed Dec 16, 2020 7:16 am 
Oza

Posts: 2356
Location: Ireland
Liked others: 662
Was liked: 442
Universal go server handle: Boidhre
jlt wrote:
The immersive method to learn a language works well if you frequently communicate with people who speak the language well. If you only interact within a community of people who make mistakes, you will reproduce these mistakes and teach them to other people.

In the same way, the immersive method can work in go if you frequently play against stronger players. This is not possible for everyone.


This comes up a lot with minority languages since they usually have relatively little native level material available, or what is available is narrowly specific. E.g. Irish has this issue but there is a very good free native radio station available to listen to, the issue is you're going to want to be interested in Irish rural matters, weather, farming, Irish politics etc. You're going to have to shape your interests to what's available. If you are going to delve into the literature, again you'll have to live with what kinds of genres, time periods and stories are available. You don't get to go in with a mindset of "I only read Science Fiction" unless you're happy with only a handful of books to work with.

It feels you're going to hit the same wall with go. I think things have massively improved since pro and extremely strong amateur players started to stream games. It's a "light" way to feed your "go ear" a lot of high quality moves. In general though, we have access to this material, it perhaps just isn't in the form we'd most enjoy. If you want to attain native level Irish spending your time talking to and reading works from non-native speakers of middle-upper middle level is worse than a dead end. In go maybe there's something similar, if most of you time is spent watching/playing other middling amateurs, why should we expect to play above that level? Similar to Irish, once you're happy to just speak it and don't care about accuracy etc it doesn't matter. If you want to play better, or similar to with a language not necessarily attain native level but be able to appreciate it, then yes, the answer seems obvious to me now in retrospect.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #17 Posted: Wed Dec 16, 2020 7:17 am 
Oza

Posts: 2356
Location: Ireland
Liked others: 662
Was liked: 442
Universal go server handle: Boidhre
Bill Spight wrote:
As for interacting with people who make mistakes, I have warned against weak players playing even games if they wish to advance. That's a good way to pick up and reinforce bad habits. Better to play against stronger players, and then, when they get better, to return the favor and play some against weaker players. OC, if your aim is to have fun, then do as you will. :)


Yes, much later did I realise that I really, really should have listened to you when you said that to me. Fortunately long breaks from the game offer wonderful opportunities to forget bad habits as well as good ones.


This post by Boidhre was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #18 Posted: Wed Dec 16, 2020 8:05 am 
Lives in gote

Posts: 646
Liked others: 62
Was liked: 116
Boidhre wrote:
It feels you're going to hit the same wall with go. I think things have massively improved since pro and extremely strong amateur players started to stream games.


I don't agree with you. I think for 99% of all Go players, we could still benefit massively from pro's even of 100 years ago. I don't think it hurts to try and learn from guys like Shusaku, Dosaku, Shuei... Professional players and high amateur players, sure, they'll have to keep up with the most modern trends, but I 100% believe you can become a very strong player even if you ignore everything AI. I don't think it's entirely comparable with a language that changes. I'd compare it with just more vocabulary being added to a language.

Boidhre wrote:
You don't get to go in with a mindset of "I only read Science Fiction" unless you're happy with only a handful of books to work with.

I read a lot. And I almost only read Sci-Fi and Fantasy. I can tell you, you can fill an entire lifetime with just those two genres ;) But I see what you mean :lol:

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #19 Posted: Wed Dec 16, 2020 8:41 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
There is a tendency, it seems to me, for players to recommend what they did to reach their level, what they believe worked for them. OC, that is only anecdotal evidence, and everybody is different. Moi, I try to make suggestions based upon research. But there is no question that from an early age I have been strongly influenced by Znosko-Borovsky's advice in How Not to Play Chess. Namely,

1) Analyze the position.
2) Make a plan.

Also: Do not get lost in "a maze of calculations."

John Fairbairn wrote:
I read the relevant chapters of Jon Tisdall's book last night. May main reaction was, "At last!" A chess grandmaster finally demolished Kotov's approach, and in clinical Joran Peterson fashion using common sense in place of dogma. I had always felt uncomfortable with Kotov. At first I thought it was me. Then I thought it was a combination him and me - he was a numbers man, I'm not. Then I eventually came to the view was wrong that his status came from piggy-backing on the Soviet School of Chess.

There are elements to Kotov that do make sense, and Tisdall stresses his good advice to make a list of candidate moves, for example. But essentially, even if Kotov's method worked for a Soviet tractor it doesn't work for humans.


AFAIK, the only part of Kotov's method that has been debunked by research is not revisiting previously explored nodes of the game tree. Everybody else does it. :) Moi, I found reading each branch only once to be good discipline. If you know you're not coming back, you do your best the first time. As Tommy Armour said, A cow could do it right the second time.

As I recall, Kotov begins talking about his method of calculation in Think Like a Grandmaster with an example he used in talks before an audience. He illustrates how a hypothetical player might decide upon a play in a particular position. He explores one move, but after searching a few moves deep, he decides he doesn't like it, and tries another move. After a search he doesn't like that play either, and returns to the first move. After further search he still doesn't like it, and returns to the other play. He goes back and forth for a while, and then sees another play, which he makes without further ado. The audience laughs and applauds, recognizing that they have done the same thing. (I have, too. :lol: ) This is what Terence Reese calls dithering. Don't dither. :)

John Fairbairn wrote:
Making allowances for the differences between go and chess, I'd say that the methods Tisdall and I propose are almost in synch. There are differences - he proposes pausing every few moves to burn a visualised image into the brain; I focus more on the time element. But I suspect these are superficial things and probably not real differences.


I have also found it helpful to spend some time just looking at the board, but without any effort to visualize it. As far as problems go, however, it is important to know where the key stones are. Often, when reviewing a problem, we may recognize it based upon some trivial details. Then, when a similar position arises in a game, we may play the problem solution, but it does not work because the game position is slightly different. I have observed this in real life. ;)

John Fairbairn wrote:
There is, however, one discrepancy. Jon says he came to his method as the result of noticing that his own powers of analysis (= reading) were waning with age. His method was therefore recuperative, and worked. But he doesn't appear to say what was different about his (even more natural?) method when he was younger.


Kotov's method was also remedial. At some point he realized that he was relying too much on intuition and general principles, I have the same problem, which is why I appreciate his way of overcoming it. :)

John Fairbairn wrote:
So why is go not taught by the immersive method in the west?


Until the advent of online go it was hardly possible. And the lack of FTF contact is also an obstacle.

John Fairbairn wrote:
Why does the hairshirt approach prevail? I think the reason is the prevalence of mathematicians here, and it was interesting to me to watch a recent Michael Redmond video where he made the same point, though more diplomatically than me (yet more tellingly, because he has a mathematical background himself).


I strongly doubt it. The mathematical approach is to make things easier. The woodshedding necessary to accomplish that may be hard work, but that is not go.

AFAICT, it comes from the practice of strong Eastern players, who recommend what worked for them.

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

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Follow the wise old men - and solve problems faster
Post #20 Posted: Wed Dec 16, 2020 10:02 am 
Oza

Posts: 2356
Location: Ireland
Liked others: 662
Was liked: 442
Universal go server handle: Boidhre
Ian Butler wrote:
Boidhre wrote:
It feels you're going to hit the same wall with go. I think things have massively improved since pro and extremely strong amateur players started to stream games.


I don't agree with you. I think for 99% of all Go players, we could still benefit massively from pro's even of 100 years ago. I don't think it hurts to try and learn from guys like Shusaku, Dosaku, Shuei... Professional players and high amateur players, sure, they'll have to keep up with the most modern trends, but I 100% believe you can become a very strong player even if you ignore everything AI. I don't think it's entirely comparable with a language that changes. I'd compare it with just more vocabulary being added to a language.


I was not for a second suggesting it wasn't worthwhile studying older games or that only go post-AlphaGo is worth having! What I was getting at is that this is a *very* digestible format for the modern age and being able to hear a pro comment on the game as they play it live is something new and very welcome. It also gives people an excellent opportunity to play "where would I play here?" and get instant feedback on it. Which I think is a very good way of actively engaging in watching stronger players play.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 44 posts ]  Go to page 1, 2, 3  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