Swim's Reviews

[djhbrown]

If you would like your game reviewed for free by an expert with commonsense, then post it here and Swim will review it for you.

You just need to tell me at which point in the game you discovered that you had gone wrong, and Swim will tell you what you should have done before that instead of what you did - and, most importantly - WHY you should have done it differently.

All levels welcome, from 30 kyu to 9 dan. The tougher the problem, the more Swim likes it; she has already discovered what Alphago should have done against Lee Sedol in the only public game it has ever lost.

Only one review per week; first come, first served.

[Kirby]

At move 36, I think I should have played Q9 instead. What does Swim think?

[djhbrown]

i will let you know asap; please hold your breath until i get back to you with her answer, and she will answer, even if it's a spoof

PS Q9 would be queer indeed, but as the beans boys say: "you know everything, Brian", even if you insist you are not the Messiah. anyhow, what you think is as irrelevant to Swim as what alfadoodle thinks - but she will listen to Leela's opinion, after she has formed her own. Let the jury decide on a majority verdict. Now if you will excuse me, the bike is calling.

PPS ah, you mean Q9 at move 38 i think i know what your problem is, even before asking Swim

PPPS: oops! i was looking at the wrong 36.... attention to detial has never been my strong point

PPPPS: i'm not going to ask Swim this, but i am asking you: it look to me like 34 is wrong direction - what do you think?

[Kirby]

I dunno- just giving you the benefit of the doubt; curious what to see an example of your swim thing applied to my game.

[djhbrown]

you're not giving me anything - except a chance. and that's all i asked, so you are giving me everything.

but i want more - because my question about 34 was a serious one - it genuinely does look to me like you attached on the wrong side.

but there's only one thing i know for sure about Go, and that is i don't know anything about anything for sure.

there is a principle that i think applies to 34, namely, starting off on the opposite side of the one you want, and i'm sure there's a Guo Juan lesson somewhere where she talks about midgame joseki, and the crosscut would be a key part of that, so maybe 34 was on the correct side.

i'm not strong enough to know or to remember, my own game never had a solid foundation because i started learning so late in life (about age 27) but if you're really 3D you ought to be, so you should be able to answer my question, or at least have a decent stab at it.

you do that for me, and - i promise - Swim will do something for you in return. She will tell you something you didnt know, or maybe something that you do know, but forgot temporarily when you were playing that game 36 is as decent a place to start as any, i guess, but i'm going to have her take a look at 34 anyway, because i'm now curious about it.

so please dont answer my question until i tell you her answer, because i dont want any subconscious bias pushing me towards one answer or another, even if i already have one to find a better move than 34!

And if Swim doesnt come up with anything that makes you want to fall on your knees and worship at her feet, i will eat my hat.

Discipline is the art of getting mindless fools (disciples) to follow you up the garden path

[Kirby]

Discipline is the art of getting mindless fools (disciples) to follow you up the garden path

You sure are an interesting character, aren't you...

[djhbrown]

ditto to you; i wouldn't talk to you otherwise.

now shut up and let me start swimming, or we'll never get anywhere

it's raining today, so i might get a bit further than this, which is the result of pass 1 of the colour map on move 33; i will get to 36, but 34 is clamouring for attention at the moment; if i'm right that your 34 is wrong direction, and attachment on the other side is too cramped, then tenuki or some other option might be preferable. Swim has a fondness for moyo expansion/reduction shoulder hits (which she learned from Jennie Shen way back in the beginning), so i'm half-expecting one of those, but we'll see.

who is this david campbell? he's not that bloke that dragged me down to Sydney on false premisses in 1999, just to make the numbers up? i recall that fellow had another empty aphorism as his byline: "the meaning of a message is the response you get". But to be fair to him, he was only doing what HR required him to do, so it wasn't his fault. And besides, if they had said yes, i wouldn't have gone to China and had all the interesting, enjoyable (and one strange, vaguely disquieting) experiences there.

i will post each stage of Swim's processes, so you can see how it develops. i often make slip-ups on the maps, so it would nice if you or someone could push it past pnprog's gomap and see if it concurs with this:

kirby.png (60.39 KiB) Viewed 17782 times

http://www.news.com.au/entertainment/tv ... 6042655ab5

http://www.theherald.com.au/story/42335 ... ark/?cs=12

[djhbrown]

(yes, Jennie Shen - i keep getting it mixed up; thanks, whichever moderator or hacker edited my spelling)

here's the end of colour map pass 2. capturing that one white stone makes a huge difference to black's upper cluster, as you can see (and we haven't even got to the shadows, yet).

the shadow map is not an optional extra, but a vital component of situation assessment, for without it, as here, it looks like game over because black has so much and white so little.

But it's not the whole story...

kirby2.png (106.27 KiB) Viewed 17771 times

[oren]

Your map is for move 34 and not 36 which seemed to be the question.

[djhbrown]

... the human instinct for cooperation and neighbourliness. We are pack animals, who have achieved our evolutionary success by working together

The most powerful drive in the ascent of man is the pleasure he derives in his own skill. He loves to do what he does well; and having done it well, he loves to do it better. You see it in his science, you see it in the magnificence with which he carves and builds. The loving care; the gaiety; the effrontery!

http://www.youtube.com/watch?v=rIcOpCxb ... h&index=10

the picture shows Swim's preliminary perceptions of the situation after move 33.

Before analysing her perceptions, Swim asks you to tell her your subjective beliefs about the relative worth of territory and shadow, and how much of the still neutral space you expect to be able to get.

To do this, Kirby, with reference to the code below, please can you choose values for f, g, and h.

f and g represent your opinion of the relative values of prospective territory and influence (shadow)

they are general functions, but for this experiment, we can consider them as simple weight factors; for example, f=g=1

similarly, h can be a weight factor, representing your general degree of optimism about to how much of the open space you can sequester.

since you are stronger than your opp, one simple objective heuristic you could use is n*(hisrank/yourrank) - this assumes you think you will end up with n points per stone from now on, but it doesn't weigh in your feelings about your or his chances of capturing weak groups - it would be possible to define a more general function for h, but i don't think that's necessary at this stage of experimentation, since our focus is on your appraisal of the effectiveness of Swim's reasoning methods, which will begin once we have decided values for f, g and h.

for an even game, the simplest choice is to just let f,g = 1 and h = 0.25 (which assumes it will be evenly shared; the open space is reduced each time a move is played - if stones get captured, they are more likely to become territory than open space).

Code: Select all

let lead =  value (me) - value (you); 
if lead + chance > 0 then playsafe    
else attack
       
value (x) = f (territory (x), g (influence (x)) 
                  + if white (x) then komi

territory (x) =   size (eyespace (x)) +  prisoners (x))

influence (x) =  if weak (x) then 0 
                 else g (size (shadow (x)) 
                 
chance   =   h (size (neutral space))

kirby3a.png (166.66 KiB) Viewed 17536 times

Code: Select all

let lead =  value (me) - value (you); 
if lead + chance > 0 then playsafe    
else attack

value(player) = sum(value, player.group)
       
value (x) = f (territory (x), 
               g (influence (x)) 
            + if white (x) then komi

territory (x) =   size (eyespace (x)) 
                  +  prisoners (x))
influence (x) =  if weak (x) then 0 
                 else g (size (shadow (x)) 
                 
chance   =   h (size (neutral space)) 

Swim's output shown in black
additional comments in brown

weak groups have no value.

white has 4 groups: top right is alive; upper left is strong because its territory+shadow is big.

the minimum size needed to make 2 eyes is 7x3 = 21 on the side, and 7x5 = 35 in the middle.

lower left hoshi stone is weak (even though it's not under direct attack)

white's right side group has a link to the side, but its eyespace of 12 points < 21, so it's weak.

if f=g=1:

value(black) = 57+80 = 137
value(white) = 11+62 = 73

if h = 0.25:

chance = 80/4 + 6.5 = 26.5

lead = -64

lead+chance = -64+26.5 = -37.5

therefore, white must attack

[djhbrown]

methods create alternative subgoals

Code: Select all

to attack do
   foreach you.group
   if weak(you.group) then attack(you.group)
   else foreach you.moyo invade(you.moyo)
# when reduction is not enough

to attack(group) do 
   if surrounded(group)
   then reduce(group.eyespace);
   and then [to be completed]
   else foreach group.cluster where width(group.cluster.link)<3
        cut(group.cluster.link)

kirby4.png (168.44 KiB) Viewed 17539 times

black has two groups; both are alive

a sector line between two groups is a straight line across empty intersections between outlier stones or colour-controlled edge points of two groups.

a moyo is an area bounded by sector lines and colour-controlled edge points.

kirby4moyo.png (168.75 KiB) Viewed 17539 times

black has one moyo: D18-D19-L19-M17-R16-T15-T1-R9-O3-Q3-R4-R5-T4-T1-M1-O3-D18

this includes the white stone at P9

to be continued

[alphaville]

You are not analyzing the actual move that Kirby asked about, not sure if you noticed someone else who already pointed that out earlier - if you did, please ignore my post.

From the analysis so far - Swim seems to come up with huge difference in estimated territories, that must be a bug in the algorithm (or is it in the implementation? )

[djhbrown]

huge difference in estimated territories

not estimated territories, estimated potential territory/influence.
There is a difference - for example, TM estimates her potential territory is not as big as she wants, so she has set herself the goal of making it bigger - unless, that is, she's just following orders from a higher-up.

[djhbrown]

573 views, and not one person has noticed the idiotic statement that weak groups have no value. Of course they have value. You lot are as useless for not noticing it as djhbrown is for saying it.

[oren]

All of these posts have little value is the problem.