It is currently Thu Nov 23, 2017 12:11 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 153 posts ]  Go to page Previous  1 ... 3, 4, 5, 6, 7, 8  Next
Author Message
Offline
 Post subject: Re: Commonsense Go
Post #101 Posted: Tue Apr 25, 2017 9:04 pm 
Judan

Posts: 7588
Liked others: 1379
Was liked: 1162
KGS: Kirby
Tygem: 커비라고해
Javaness2 wrote:
Surprised that this concept of Commonsense Go has generated so many responses, you couldn't even get to the level of the 4 step beginners guide with this.


For me, this thread has entertainment value. I never know what will be posted about next! The long posts are the best - winding paths of different topics that leave me scratching my head, wondering what I've just read. It's a nice break from reality.

_________________
What have I learned from this game that I'll be able to use in my own game if the opportunity presents itself?


This post by Kirby was liked by: Waylon
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #102 Posted: Tue Apr 25, 2017 11:18 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
Attachment:
26042017058.jpg
26042017058.jpg [ 460.59 KiB | Viewed 1485 times ]
“Books! - 'Tis a dull and endless strife
Come, hear the woodland linnet,
How sweet his music! - On my life,
There's more of wisdom in it.”
− William Wordsworth (1798)
Attachment:
alphacat.png
alphacat.png [ 16.13 KiB | Viewed 1485 times ]


PS please join the campaign for a saner linux with a bit of commonsense
https://askubuntu.com/questions/910135/ ... -for-chmod


Last edited by djhbrown on Sat Apr 29, 2017 1:45 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Gomap
Post #103 Posted: Thu Apr 27, 2017 4:23 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
Swim vs Alphago: i am as big a fan of Alfredgonefishing as anyone; my nicknames for her are affectionate, like the nicknames English schoolchildren give each other. btw, if you still don't believe me about all this, have a look at my work history and books and videos; if that doesn't convince you i'm not barking mad, or am!..., nothing will.

Alphagoat et al perform state-space search which creates a projected move tree, but icGo's Swim really does 'think' and can tell you what she's thinking in plain English. Swim might even give Miss A a run for her money and make you rich beyond Croesus when Google buys your company! (i will probably be dead by then, because 10 years ago i was given 5 years to live by a lung specialist, so i'm already on borrowed time).
Swim's planning methods don't have blind spots like A does, despite her being top of the tree. It will take the pros a while, but if they get the chance to play with her a lot more, they may start to find out where they are (it will require them to play moves that are not overly risky and be willing to sacrifice for positional value, because A has no idea what that is).

Any bot can Monte-Carlo; A's dcnn is better than old Zen's large patterns, which were better than the rest's smaller ones, but Swim looks at the whole board.

We know that A can look ahead more than 50 moves (indeed, all Monte bots read up to 361 moves deep), giving her the edge over human pros in tactical combat - but any bot, including Swim, can do that too. What matters is not how much you read, but what you read, and that's where Swim could have the edge over A.

There's a lot of bulldust being sprayed around about A's inventive moves and new joseki and blah blah spin blah - but the reason those unorthodox yet tactically sound moves have worked so far is because she can look deeper and wider than even Ke Jie, The only sane voice on the subject i have heard so far is Michael Redmond's in his new Redmond's Reviews series - from his various comments and analyses, i think Michael can see what i can, which no-one else seems able or willing to say in public for fear of being ridiculed by the mob because A is top dog.

Alfa is a whole other story, and one that science is just starting to scratch the surface of (and one which intrigues me greatly, because like just about everyone, i never anticipated dcnn+monte could be so powerful at Go. Separately, they aren't all that much better than GnuGo; Zen did get to 6D, for example, but couldn't reach pro level even with parallelisation.

But together, dcnn and monte are astonishing. the DARPA fellow does a pretty good job of explaining dcnn, but that doesn't explain why the combination of dcnn and monte works so well.

To learn how Alfadog thinks, follow the following procedure:
Code:
1. pick a pic of someone or something you're curious about
2. right click, copy link
3. Google search for "Google image search"; click the top link
4. mouse to its search box. left click. paste.
5. Enter
and then you will know as much about them or it as Alfadog does about Go - honestly!

Maybe, that such a mindless technique as autonomic pattern reflexes + throwing dice works so well at Go, tells us more about the mindfulness of Go than it does about the mindfulness of dcnn. Maybe Go isn't so different from chess after all, despite what we used to think 40 years ago.

Be that as it may, Go is still a captivating pasttime and we can read into it all the poetry in the world.

Announcement

pnprog, author of GoReview Partner, has published a python widget called Gomap that performs stage I of icGo's 3-stage colour-map algorithm.

You give it an sgf of a game record without variations and it will iteratively draw the map on a board image, one step at a time, as you repeatedly press its 'Color' button.

You can step through your game with the 'prev' and 'next' buttons, and redraw the gomap at any move.

'Export' button produces a .ps file of the current frame. You can assemble a sequence of frames into an .mp4 movie and convert it to a .gif

Here is the gomap for the position after move 116 in the Nick vs Andrew game discussed in Mig34:
Attachment:
lastframegm116.png
lastframegm116.png [ 274.71 KiB | Viewed 1452 times ]

There is an animation of the iteration that produces the map, which i made with Simple Screen Recorder, at https://www.youtube.com/watch?v=BoKi2rw ... S&index=30

The blobs are "colour-controlled" points - they are neither territory nor influence - the map is just the first of several icGo stepping-stones towards identifying those.

But even this preliminary map might be helpful to beginners, as it shows how stones are connected or not. You may be surprised to see just how much difference one stone in the right place can make!

Gomap installation instructions:

1. Equip your machine with a python interpreter and the Ktinker package
2. Download Gomap.zip from GitHub.
3. unzip and run (pun intended) from within your gomap-master folder.

On Linux, i created the following gomap.sh script which i plug into an icon on a Xubuntu panel, so running gomap is a one-click operation:
Code:
cd /home/d/go/gomap/gomap-master
python gomap.py

To run it under Windows or other o/s, you'll have to ask pnprog


Last edited by djhbrown on Mon May 08, 2017 1:39 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Gomap
Post #104 Posted: Thu Apr 27, 2017 10:54 pm 
Judan

Posts: 7588
Liked others: 1379
Was liked: 1162
KGS: Kirby
Tygem: 커비라고해
djhbrown wrote:
(i will probably be dead by then, because 10 years ago i was given 5 years to live by a lung specialist, so i'm already on borrowed time).


Best wishes. :salute:

_________________
What have I learned from this game that I'll be able to use in my own game if the opportunity presents itself?

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #105 Posted: Fri Apr 28, 2017 8:22 am 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
thanks for your wishes, Kirby. i'm cool with it - i only mentioned it to encourage programmers to realise that if they did anything about icGo, it would be for their own benefit rather than mine, although of course it would be pleasing to see something happen.

pnprog is a good programmer with interesting ideas of his own; for example, Go Review Partner has the nice feature of flashing up Leela-generated sequences on the board as you hover over its move candidates, which i see as a big improvement over the cumbersome old-style variation navigation of GoGui and all its clones. of course, this means it can only handle one variation per stone, but if one added a button-press to a hover... trouble is, keyboards have the arrow buttons on the right-hand side of the keyboard, so you would end up like a piano player crossing hands!

but then again, there's all those function buttons at the top which i hardly ever use....
alphaville wrote:
Emphasis on: you chose a particular move, then you worked backwards to justify how Swim is great
WRONG! :) i didn't do that - you have just provided us with another example of "projection"
https://www.psychologytoday.com/blog/tr ... d-identity

you think you know what i did, because you think to yourself (usually subconsciously): "well, if i had said what he said, i would have said that because i did it in way X for reason Y"

it's what people do; projection arises out of our mirror neurons, which do more than just monkey see, monkey do, which is how children learn to imitate their parents; it's also the driver of our empathy circuits, by which we imagine we can feel what another person is feeling when we see them behaving in a certain way, such as crying their eyes out or guffawing.

it's one of the many flaws in our genome, one that evolution hasn't yet had time to iron out, because we are one of the most recently evolved and hence still wet behind the ears intelligent plastic machines on the planet. http://lcipm.blogspot.com.au/2013/11/li ... astic.html

it's a flaw, because projection gets in the way of real empathy. sometimes people cry because they're chopping onions - Marilyn Monroe's acting teacher was a great believer in not sticking onions into an actor's eyes, but sticking metaphorical pins into them instead so they really hurt and the cry was a genuine cry of distress. https://en.wikipedia.org/wiki/Lee_Strasberg

whether antisocial activities such as bullying, trolling, serial killing, bossing around, teasing, and trying to rule the world, or doing such in virtual reality computer games (and war games like Go!) are partly the result of malfunctioning or non-existent empathy circuits is something many psychologists have studied.

as it happens, i was a bit disappointed when Swim said : "let's consider J13" because, not being quite as strong as Lee Sedol (no, really, he actually is a bit stronger than me), i didn't see the point of J13 at first, but thought "oh well, let's carry on and see what happens".

the rest is in the paper.

the literary character called Jesus (named after Horus) is said to have said "Physick, heal thyself". Good advice. i need to take it. There are more than enough Doubting Thomases and thicker than two short planks rabid weasels (http://animalinyou.com/animals/weasel) out there who can't see the wood for the trees, or don't know a good thing when it's dangled right in front of their eyes because they're blind to reality for whatever reason that they don't see it for what it is but instead only see their own personal prejudiced delusion, as if they were watching a Noh theatre play. https://www.japan-zone.com/culture/noh.shtml

So i need to take Horus' advice and heal my paper so that it says things that even True Believers of a different faith can imagine might just be remotely possible.

You know, something is telling me that none of Alphaville's questions are actually questions, but rather motherly pointers to guide me towards doing what i should have done in the first place to avert trolls kneejerking out of their caves with things like "this is crap, it can't possibly work, who the hell d'you think you are?!, you've never written a program, i am better than you, stop posting your drivel, i will cast the first stone and burn your books like the National Socialist Party did to authors of whom they didn't approve because they sang a different song" - the same sort of thing that the self-proclaimed tolerant English did to Jeanne d'Arc.

so, thank you, my new friend :bow: - unless you're an old one in disguise, but i can't imagine that, as none of them play Go, except Tony, who kicked the bucket recently so you can't be him unless you're a ghost who can type and i've never seen one of those (i suppose you could be his typist...); but besides, he would have just pulled my leg even further, as he would do when switching colours and beating me again from a position i had resigned from. Hang on a minute, you are pulling my leg even further....oo-er, spooky... but it's nearly 40 years since we both didn't work in the same place! Parallel universes?

go and get some fresh air, and you still haven't got those pedals off... oh yes i have! there's something satisfying about manual labour and getting your hands dirty, and the pleasure of bringing something discarded back to life; i remember how impressed i was, in Ceylon and India, how people kept ancient Morris Minors and mashupped Harleys going, decades after the Pohms and Tanks had given up trying and bought new plastic ones from Japan; China is following fast in Japan's footsteps and it won't be long before China engineering (eg JueYi) is as horsepowerful as Porsche's Alfetta; just hope that Donald Duck doesn't blow up the world before the icebergs melt and drown Manhattan and all its nefarious Projects.

PPPPPPS
Major: "Why do we bother, Fawlty?!"
Basil: "I didn't know you did, Major"
:)

_________________
i ink, therefore i arm

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #106 Posted: Fri Apr 28, 2017 6:00 pm 
Dies with sente

Posts: 101
Liked others: 21
Was liked: 16
djhbrown wrote:
alphaville wrote:
Emphasis on: you chose a particular move, then you worked backwards to justify how Swim is great
WRONG! :) i didn't do that - you have just provided us with another example of "projection"
[...]
as it happens, i was a bit disappointed when Swim said : "let's consider J13" because, not being quite as strong as Lee Sedol (no, really, he actually is a bit stronger than me), i didn't see the point of J13 at first, but thought "oh well, let's carry on and see what happens".
the rest is in the paper.


It is very likely you are right with the projection part! Can you please explain in more detail then J13 - did it come up as the only candidate (I guess not); what other candidate moves where there? Can you also give examples of bad moves that were suggested by Swim, for balance?

djhbrown wrote:
You know, something is telling me that none of Alphaville's questions are actually questions, but rather motherly pointers to guide me towards doing what i should have done in the first place to avert trolls kneejerking out of their caves with things like "this is crap, it can't possibly work, who the hell d'you think you are?!, you've never written a program, i am better than you, stop posting your drivel, i will cast the first stone and burn your books like the National Socialist Party did to authors of whom they didn't approve because they sang a different song" - the same sort of thing that the self-proclaimed tolerant English did to Jeanne d'Arc.

so, thank you, my new friend :bow: - unless you're an old one in disguise, but i can't imagine that, as none of them play Go, except Tony, who kicked the bucket recently so you can't be him unless you're a ghost who can type and i've never seen one of those (i suppose you could be his typist...); but besides, he would have just pulled my leg even further, as he would do when switching colours and beating me again from a position i had resigned from. Hang on a minute, you are pulling my leg even further....oo-er, spooky... but it's nearly 40 years since we both didn't work in the same place! Parallel universes?


Looks like we are in a love-hate relationship :-) I don't really believe in what you do, but I respect you for believing in it. I also think you can do a better job explaining what you do and dealing with your critics.

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #107 Posted: Fri Apr 28, 2017 9:24 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
alphaville wrote:
Can you please explain in more detail then J13 - did it come up as the only candidate (I guess not); what other candidate moves where there?
the trail of perception and reasoning that led to J13 is in the paper. it was the first thing that came up. of course, there would be other candidates, because Swim's methods have lots of "ors" in them, and the idea is that the more goals a move achieves, the better it is. this is stated explicitly in the paper. i didn't explore further once L10 turned up :) - Nor will i; like i said before, you do it. just follow the algorithms; it isn't hard to do. just takes a bit of time.

alphaville wrote:
Can you also give examples of bad moves that were suggested by Swim, for balance?
How dare you suggest Swim would ever suggest anything bad! You will be punished severely for your impudence by being smashed to pieces by her once she comes out of the birth canal.

alphaville wrote:
I also think you can do a better job explaining what you do and dealing with your critics.
A writer's best friend is his critic. Trolls, however, can get stuffed. Norman Mailer agrees with me, although useless google cant find a citation where he says some literary critics should be punched out :)... no, wait, sorry, your Lordship G, here is one: http://www.bostonphoenix.com/archive/tv ... AILER.html


Last edited by djhbrown on Mon May 08, 2017 1:35 am, edited 2 times in total.
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #108 Posted: Fri Apr 28, 2017 11:47 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
djhbrown wrote:
you do it
don't do that, i've got a better idea:

choose any position in any game, then:

To find out Swim's move, do the following:

turn the pic into a .sgf
feed it into Gomap which will do the first bit of Swim's thinking for you
then identify dead clusters following the rule described on p18 of icGo's running draft https://sites.google.com/site/djhbrown2/cg.odt (it's the same as in the CG paper).
then redraw the colour map, add the shadow map, count the score, count the open space,
and start up the honte strategy (p37) to create Swim's first top-level goal
i bet you Swim decides that white should playsafe

then apply the goal to the metamethod (p35) and follow the subgoal-creating methods down to see what moves it comes up with

then, for each of these (or just one of them if you're feeling lazy), switch over to black, recompute colour and shadow etc, and see what black's counter-strategy will be, find his top-level goal, apply the metamethod, and resolve to a move or moves

then find white's next move.

that's enough searching.

by now it should be clear to you which move Swim prefers for white, at this stage, without looking deeper. it will either be the one that achieves most subgoals (preferred), or the one that achieves her highest-priority subgoal.

then post your answer to this thread

if you get stuck or lost, just ask here, showing me what you have done up to then (attachment is best so you can edit it easily before posting), and i will guide you to the right page in icGo draft to continue. be sure to check icGo documentation every so often, as i periodically change it so page numbers may change,

you do that while i get on my bike (still not road-ready yet; ive been clicking and typing too much - bad for the health).


Last edited by djhbrown on Mon May 08, 2017 1:38 am, edited 2 times in total.
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #109 Posted: Sat Apr 29, 2017 9:43 pm 
Dies with sente

Posts: 101
Liked others: 21
Was liked: 16
djhbrown wrote:
To find out Swim's move, do the following:

turn the pic into a .sgf
feed it into Gomap which will do the first bit of Swim's thinking for you
then identify dead clusters following the rule described on p18 of icGo's running draft https://sites.google.com/site/djhbrown2/cg.odt (it's the same as in the CG paper).
then redraw the colour map, add the shadow map, count the score, count the open space,
and start up the honte strategy (p37) to create Swim's first top-level goal
i bet you Swim decides that white should playsafe

then apply the goal to the metamethod (p35) and follow the subgoal-creating methods down to see what moves it comes up with

then, for each of these (or just one of them if you're feeling lazy), switch over to black, recompute colour and shadow etc, and see what black's counter-strategy will be, find his top-level goal, apply the metamethod, and resolve to a move or moves

then find white's next move.

that's enough searching.

by now it should be clear to you which move Swim prefers for white, at this stage, without looking deeper. it will either be the one that achieves most subgoals (preferred), or the one that achieves her highest-priority subgoal.

then post your answer to this thread and i will reply by giving you the url of the lecture so you can find out what Go Seigen did.


This is where we disagree: I have a very strong belief (prejudice?) that in most cases, moves suggested by any such rigid, top-down, rule-based heuristic are not very good. Of course, with a finely tuned heuristic (such as Swim's may be), it will identify some "hot areas" on the board, but I don't see how it would single-out "best moves".
I don't doubt that in some cases it will (maybe in this Go Seigen game it does, as you imply), but the probability of that is low, and in most cases it will not be able to find good moves.

I would like to try what you suggest, but I don't have time.

By the way, I think I understand (by reading more of your posts) why you claim that Swim is different than say GnuGo: you refer to GnuGo as "bottom-up" and to Swim as "top-down". Let me know if I got it right:
- bottom-up GnuGo: analyzes the postion starting from individual stones, builds some higher-level concepts (groups, influence); looks at all legal moves (maybe with some basic filtering); tries them out, and compare the results somehow
- top-down Swim: analyzes the position starting from individual stones, builds up some higher-level concepts (groups, influence); makes an assessment of strength/weakness; decides on some goals; tries to find moves to reach the goals (without trying-out moves, since that would be bottom-up?)

On a different note: shouldn't one be able to apply this in their games too? No need to be a computer to make use of Swim?

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #110 Posted: Sat Apr 29, 2017 10:29 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
http://www.alanturing.net/turing_archiv ... 0AI09.html

anyone can apply Swim's techniques to their own thinking, but only if they have the time to think.

however, i can't imagine anyone making colour maps in their head, beyond colouring the words of others to reinforce their own prejudices about colour.

as to whether GnuGo is top-down or bottom-up, i have previously explained my reasons for my categorisation of it on several occasions to the best of my ability. According to its documentation, GnuGo does not make plans; instead, it chooses moves based on stone patterns and low-level goals related to a highly idiosyncratic object called a "dragon" (links through ogeima). I have nothing further to add.

i had no hand in its development and have no more information on it than is in the public domain. If you choose to call it top-down, i won't argue with you; i refuse to argue with anyone. i don't have time for that.

i enjoy scientific debate and discussion, but i detest "I am right and You are Wrong" arguments as much as does Edward de Bono, the author of a book with that title.

If you wish to explore GnuGo further, you could try asking
http://lists.gnu.org/archive/html/gnugo-devel/


Last edited by djhbrown on Sun Apr 30, 2017 1:08 am, edited 2 times in total.
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #111 Posted: Sat Apr 29, 2017 11:12 pm 
Dies with sente

Posts: 101
Liked others: 21
Was liked: 16
djhbrown wrote:
http://www.alanturing.net/turing_archive/pages/reference%20articles/what_is_AI/What%20is%20AI09.html

anyone can apply Swim's techniques to their own thinking, but only if they have the time to think


Point well taken!

To do a top-down approach, one still needs to start bottom-up (from individual stones) in order to build the high-level concepts right?
I don't buy into (or maybe don't really understand yet) how the top-down vs bottom-up are so different, in particular in the Swim vs. GnuGo-like systems.

A cursory look at GnuGo's source code tells me that it may not be as different than Swim's heuristic: they do what looks like top-down analysis, looking for goals/reasons to play:
http://git.savannah.gnu.org/cgit/gnugo. ... h-3-8#n293

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #112 Posted: Sun Apr 30, 2017 3:15 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
ALL information processing functors are necessarily bottom-up because information doesn't come to you in high-order chunks, but as a stream or screen of bits. Top-down processors interpret inputs and determine ways to achieve prior objectives which are situation independent. For example, seeking cooperation and seeking competition are different objectives that, applied to the same situation, yield different results.


Last edited by djhbrown on Tue May 02, 2017 1:47 am, edited 2 times in total.
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #113 Posted: Sun Apr 30, 2017 7:43 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
How to use Gomap:

1. open, in Gomap, a game in which you had a nasty surprise and lost a group because it got cut off and its eyes were squeezed shut by your opp, or you lost a stupid semeai you should never have started
2. fast forward to the move when the truth finally dawned on you
3. go back to your previous move

4. hit the colour button, slowly, so you can see the map grow

now do you see your previous hallucination revealed in all its true glory?

if, not, go back one step further and recolour - now do you see??

ok, go back a few more steps and colour again

repeat if necessary


Last edited by djhbrown on Tue May 02, 2017 1:31 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #114 Posted: Sun Apr 30, 2017 8:00 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
update to icGo documentation Q&A:

you'd be making a bot similar to GnuGo
GnuGo and Alphago are "generate-and-test" methods, whereas Swim uses deduction.


Last edited by djhbrown on Tue May 02, 2017 1:33 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #115 Posted: Mon May 01, 2017 1:54 am 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
icGo preliminary perceptions:
Code:

colour; identifyclusters; lad; colour; shadow;
groupidentification; groupstatusidentification,
sector-lines; greenlines;
key-cutting-points;

# colour = pnprog.gomap
# cluster identification procedure:

   board = [1:19,1:19]point
   point = [int i,j; bool stone; colour colour, int clusternumber]
   colour = [oneof {white, black, neutral}]
   player = [colour, {clusters}],
   cluster = [{point}; number]
   clusters = [{cluster}, numberof]
   
   
   makenewcluster(point, colour)) =
       colour.clusters.numberof +:= 1;
       let newcluster = [{point}, clusters.numberof]
       paint(board.point, point.newcluster.number, point.colour(point)
       
# cluster number colours are black = blue, white  = yellow
# maybe numbers inside thin circles (or on stones)
# in my hand-drawn examples, i wrote cluster numbers on stones,
# but the first point found for each cluster is just as good
       
   identifyclusters =
       for colour in {black, white} do
           colour.clusters.numberof := 0;
           for point in {A1 to T19} do
              if all-links(point) are same-colour or neutral then
                 for each coloured-link(point) do
                     if member(otherend(link, point), existingcluster)
                     then add(point, existingcluster)
                     else makenewcluster(point)
                       
# cluster preliminary life and death computation:
# olad computes "obvious" life and death of clusters which can be identified without reading

olad (clusters) =
   foreach cluster in clusters do
      identify(cluster.eyes);
      if number(cluster.eyes) > 2
      or size(cluster.eyes) > 3
         and shape(cluster.eye) not in {square4, bulky5, flower5, rabbity6, wierd7}
      then cluster.lad := alive
      else if surrounded(cluster, enemies)
           and foreach enemy in enemies (enemy.lad = alive)
           then cluster.lad := dead
           
identify (cluster.eyes) =
   foreach point in cluster
     if colour-controlled(point) and not border(point) or stone(point)
     then append(point, cluster.eyes)
     
surrounded (cluster, enemies) =
   not(forany point in border(cluster)
          path(friend(point)
          or path(openspace, point))
         

bug finders and translators invited

zzzzzzzzzzzzzzzzzzzzzzzz

icGo function keys:
(note: lastmove is not indicated on board display,
but can be seen in variation tree pane, which autoscrolls to maximise nodes in view)

c = colour mode (keep colouring/shadowing/etc while pressed)
b = nocolour mode
>,< = single move step along branch (autcolour if in colour mode)
End = skip to end of game (ditto)
Home = skip to start of game (ditto)
uparrow = skip to start of previous variation (ditto)
dnarrow = skip to start of next variation (ditto)
spacebar = skip to end of next variation (cycles back to first from last; keep skipping while pressed)
d = open up skip delay slider (like a volume control; autocloses on mouse leave)
s = startup Swim to add new variation to current node
h = Huh? = [see description in icGo documentation]
r = startup grp on current variation
j = joseki matchup; adds matching joseki variations to tree

zzzzzzzzzzzzzzzzzzzzzzzzzzzzzz

and now for something completely different:

Top
 Profile  
 
Offline
 Post subject: 3 views
Post #116 Posted: Wed May 03, 2017 6:53 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
the picture below illustrates 3 different views of the same thing. Striking differences are:

1. GnuGo and Sabaki both give white the top left corner, but icGo's colour map (hand-drawn by me) thinks the black stone still has some aji (Swim's olad predicate (defined in my previous post) doesn't conclude that C17 is obviously dead, because white's colour map doesn't wrap right around it).

2. all three give black the top right, but Sabaki and icGo both think black has potential down the right side. in icGo's case, this is only because of the black stone at O17, which causes the map to percolate right around the corner from the top. I had to redraw the map twice, because at first i thought i had misread my own algorithm - but i hadn't. Had O17 been at O16 instead, icGo's view of the right side would have been very different - it wouldn't have given any of it to black.

3. GnuGo and Sabaki both give white's nikken tobi at the bottom some territory, but icGo doesn't.

4. GnuGo and Sabaki both give white's left side position some territory, but icGo doesn't.

5. Sabaki is very generous to black in the middle of the top side, but GnuGo and icGo aren't.

6. GnuGo and Sabaki both think N17 is black's territory - where GG gets such a crazy idea from is in its documentation (Sabaki's territory estimator is undocumented).

Personally, i am persuaded by icGo's view that black's keima shimari (in top right) is more valuable than i used to think - more valuable than his bottom right shimari - so i will be playing the hoshi-keima combination more often in my own games. Of course, it's only potential territory, not certain territory, and is open to a kakari, but even so,.. (kakaris can be counter-pincered and pressed down). I am reminded of a comment Haylee made in episode 2 of MiG, that these days, pros prefer the keima shimari over the ikken tobi (gesturing to the top, rather than the right, in this example). As for Alpha's favourite 4th line nikken tobi shimari, i do not deign to remark.

Hopefully, this example illustrates to unbelievers one of the many differences between GnuGo and Swim...
..I doubt it, for blind faith in one's own rightness is, by definition, blind, but as Martin Luther King almost said: "you have to dream, don't you!" :)
Attachment:
3views.png
3views.png [ 292.03 KiB | Viewed 1072 times ]
I won't ask you to vote on which is the best, as there is overwhelming evidence from both sides of the pond and both sides of the ditch that the vast majority of voters have no idea what they're voting for, nor why they vote the way they do, making pre-election polls, and elections and referenda themselves, a spin-doctor's wet dream; something that George Carlin pointed out over a decade ago. Vote not, lest ye be voted upon.

_________________
i ink, therefore i arm

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #117 Posted: Thu May 04, 2017 3:03 am 
Gosei

Posts: 1904
Liked others: 12
Was liked: 2817
I've only been following this thread in a very cursory way, so may be guilty of missing something crucial, but the above diagrams see to confirm my impression of the intended destination. My impression is that the bus is heading in the wrong direction.

I gathered that the basic idea was to produce a program that can explain moves to humans in a human-like way. Excellent goal. But it seems that the way chosen to achieve this is based on amateur ways of looking at go, and so if the goal is achieved, we will end up with a program that tells amateurs how to play amateur moves and how to think about them in an amateur way. I suspect most people would be rather keener on doing things the pro way.

I have noticed that even very strong amateurs, and pre-AI programmers above all, have often had an obsession with counting - territory, potential territory, thickness, aji. It gives us something to get hold of, I suppose, but it's a baby's teething ring when what we need is a butcher's knife.

When you read what pros say about the fuseki (which seems to be the main focus here), it just doesn't sound remotely like the way even strong amateurs talk. The best that the amateurs achieve is to sound like the commentaries written in modern magazines - but they were mostly written by strong amateurs.

If you look at what pros themselves write, they never never never never talk about talk about counting like this. Aji - beloved of strong amateurs - in the fuseki is also super rare. There's not even a sense that these things are in the background. They talk about things that appear rarely or never here, like safety, efficiency (the pro equivalent of counting), the centre, heaviness, kiai, battle lines.... And also in ways that may be unfamiliar. Some of the words will ring strong bells with amateurs, but the frequency with which pros use them, and the hierarchy, creates the sort of discord we get when we try to talk about, say, Indian music in terms of C major.

Until very, very recently modern amateurs have not been exposed to much pro talk, though a little more is now creeping in via videos made by pros themselves. In print, it's been almost all amateur talk.

However, if you go back to the 1920s and 1930s, almost all commentaries were done by pros themselves. By far the commonest mode was for both players to comment. This is how, for example, all the Oteai games were presented in Kido and its Oteai Bulletins (the precursor of the Kido Yearbooks). We thus don't get just to hear a pro view, but a pro conversation. For much of this period, New Fuseki was in vogue. This meant that they tended to concentrate on the fuseki and it makes the conversations even more valuable as they are describing also an ongoing creative process. But even without that innovation, the conversations sound barely like the ones amateurs have, and not at all like the ones programmers have.

In fact, Kido tried for a spell in the very early days to present pro commentaries as above but larded with extra "explanations" by the journalistic staff. Useful up to a point, maybe, but you really do see the language difference in its starkest form that way. For whatever reason, that trial didn't last long.

The war brought all this to an end, and after the war, in Japan, American notions of democracy helped shift the focus to go for the masses instead of elite amateurs. The result was, I suppose, what we would call dumbing down today - or the reason people like me play :).

The rise of tournaments for pros, which enabled/forced them to duck most of the commentary work, then made the discontinuity starker. If we want to get back in touch with how and what they explain, and chiefly how they see things in fuseki, I think that programmers and amateur theorists headed for Amaville need to stop the bus and try a different direction to reach Proville.


This post by John Fairbairn was liked by: Waylon
Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #118 Posted: Thu May 04, 2017 6:07 am 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
yum list *hfs*
blkid -c /dev/null

mount -t hfsplus -o force,rw /dev/sdc1 /media/external
chown -R myuser:myuser /media/external

let's see if i understand John's points (this is an open reply, not a personal one)

1. "the above diagrams confirm my impression that the bus is heading in the wrong direction"

am i correct to infer that the wrong direction is that of not following in the footsteps of Our Lord Alphago, for She is the Way, the Truth and the Light?

2. John gathers that "the basic idea was to produce a program that can explain moves to humans in a human-like way".

I do not accept any responsibility for whatever false impression anyone has who has not read my paper nor watched any of my videos.

To repeat in this thread for the umpteenth time, No, it wasn't the basic idea. It never was.

When it became clear to me, some way down the track, that Swim could explain itself, i saw that as a bonus, one which could make Swim a useful learning aide, and said so in my blog. http://lcipm.blogspot.com.au/

Perhaps i confused John by recalling in an earlier post that in 1972, i had started out by setting myself the task of seeing if i could teach a computer how to learn a language. But that was 45 years ago.

Swim is an experimental vehicle for exploring my theories of how an intelligent computational mechanism could deploy hierarchical reasoning, using the toy domain of Go as an experimental testbed. because it is free from the complication of noisy data that complicates matters for robots situated in the real world.

The basic concepts of Swim originate in my 1979 IFCAI paper, which draws upon much prior work in psychology, including and especially Bartlett's theory of Remembering and De Groot's essay on Thought and Choice in Chess. The 34 videos of MiG, one of my retirement hobbies, which i started on once i had discovered where the whole idea of God comes from, describe my current thinking.

3. As for counting, it is glaringly obvious that John has jumped to completely the wrong conclusion about the role it plays in Swim's computational processes. I infer that he has done this because he believes that what Demis says in his recent Royal Society presentation, parts of which feature in my video "Demis and Noam", is correct - namely, that it is impossible to estimate the balance of power in a Go game before its very end, which Demis also says is impossible to detect because it doesn't end like a simple checkmate.

The role of counting in Swim is spelled out as simply as i can spell it out, in my paper, so i will not attempt a simpler explanation here. But, to repeat once more, unlike Alphago and every other Go program ever written, Swim does not use counting to choose a move - instead, it only uses rough estimates of potential territory and influence to form its preliminary perceptions of the state of play to choose a strategy to look for goals to achieve.

4. On the subject of pro commentaries, MiG begins, in episode 2 of what are now about 30 episodes, by observing that pros do not know what they are thinking, because nobody, not even Christoph Koch, one of the world's leading experts on the science of consciousness, knows what they are going to say until they say it. Chritof features in another MiG episode although offhand i can't remeber which. The only difference between Christoph and the rest is that he knows he doesn't know. Antonio Damasio estimates that maybe 80% of our thought processes are subconscious, which by definition we are unaware of. If you are disinclined to google any names of famous scientists that are unfamiliar, i cannot elucidate further.

Pro commentaries are useful, and entertaining, for the audience can delude itself into thinking it understands what the pros are saying, even though the pros themselves don't know why they are saying what they say. Picasso was a rare exception to a general rule when he replied, when asked what his paintings meant, said: "my art speaks for itself". And to me, Michael Redmond is head and shoulders above the rest at attempting to drag himself down to my level.

5. i note that John disparages "amateur" opinion quite a few times in his historical precis. Evidently, in his eyes, unless one is paid by some benefactor or an entrepreneur who can see a way of making money out of one, one's opinion is not worth the paper it is written on, so i have nothing to say, beyond noting that in 1985 i enjoyed listening and watching pro talk every Sunday on NHK TV, and learned that major Japanese newspapers had been carrying pro Go commentaries since before World War I, although quite what world wars have to do with the history of Go theory and popularity, i simply cannot imagine.

To quote from my latest plagiarised/mashupped video, which Google tells me most Americans didn't watch more than 48 seconds of, presumably jumping to conclusions based on a single frame, well before all the good bits start, It is an ex-policy. It has ceased to be.

If only that were also true of policy makers, who seem hell-bent on starting world war 3. But as Black Adder has just started, i will stop here and post it as it is.
Attachment:
googleanalytics.png
googleanalytics.png [ 49.63 KiB | Viewed 989 times ]

_________________
i ink, therefore i arm

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #119 Posted: Thu May 04, 2017 3:30 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
My bus is going in the wrong direction, because all the passengers want to go the other way.

Of course, i am not one to jump to unwarranted conclusions based on a cursory scan of a couple of pictures - but since everyone around here agrees with me, i must be Right.
Attachment:
q&a.png
q&a.png [ 210.44 KiB | Viewed 935 times ]

If this is the best the best AI can do, then AI is not Going down the tubes, it has already Gone down them, for it is no better than before Dewey invented his decimal classification system, when text-based corpus retrieval by mindless robots (called monks in those days) was just as good then as it is now.

But that's not all, for if the Go playing equivalent of Google search is truly better than any living human at Go, which will be proven to be the case in a couple of weeks, then it follows unerringly and unquestionably that Go is a complete waste of time, for as an intellectual task, it requires nothing deeper than mechanical and superficial search by a couple of thousand bots kneejerking their way through a finite forest of improbable moves.

The Improbability Drive has been turned from fiction to fact, by the judicious application of nothing more profound than a few million dollars and some glue and string to tie together a basket of perceptrons and one-armed bandits, plus the application of several times that much to pay clickbaiters to muckspread the entire Internet to make the pro(le)s go "Ooh - isn't she lovely/scary/whatever".
http://hitchhikers.wikia.com/wiki/Infin ... lity_Drive

Top
 Profile  
 
Offline
 Post subject: Re: Commonsense Go
Post #120 Posted: Thu May 04, 2017 5:48 pm 
Lives with ko
User avatar

Posts: 247
Liked others: 12
Was liked: 30
Rank: NR
PS wondering who this John fellow is, and wondering in some puzzlement how he could be the same mild-mannered John that i once met in Tokyo in 1985, i discovered that he isn't, for that other John is John Power. But Google having a much better memory than me, mentioned that this John as well as that John has written some stuff published by Richard Bozulich, who rather took me under his wing during my brief sojourn of nine months in one of the most crowded yet cleanest cities i have ever visited, memories which come flooding back, which is what i guess happens to all old fogies with nowhere to Go and nothing to do, except put the finishing touches to my bike.

But Google told me by first pointing to http://www.anusha.com/fakeishi.htm which i read in open-mouthed astonishment and dismay, not least because one of Richard's ambitions way back then was to have me become his US agent for Ishi Press.

that kind of explains why Richard changed its name to Kiseido, or sold it to them, or whatever.

Hopefully it has all been sorted out by now, and if you're still in touch with John and Richard, John, please give them my regards, and to that young (now old) Dutch fellow who at the time was #1 in Europe but couldn't make it past Insei, whose girlfriend made my eyes pop when she showed me how wonderful the kotatsu is when used in the proper way.

Rob and i played once, me taking a six-stone handicap, and i remember him laughing out loud halfway through, exclaiming "You have no idea what's going on!" :)

I didn't. And i still don't. And i never will, for to be good at Go requires a certain kind of mind, one with the ability to not overlook atari, something i've never been quite been able to typo.
Attachment:
fff.jpg
fff.jpg [ 87.06 KiB | Viewed 916 times ]

_________________
i ink, therefore i arm

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 153 posts ]  Go to page Previous  1 ... 3, 4, 5, 6, 7, 8  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 4 guests


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