Page 2 of 6

Re: LZ help for position

Posted: Sat Oct 20, 2018 4:28 am
by Kirby

Re: LZ help for position

Posted: Tue Oct 23, 2018 4:34 pm
by Pippen
Click Here To Show Diagram Code
[go]$$B
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . X . . X . . |
$$ | . . . X . . . . . , . . . . . X . O . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . , . . . . . , . . . . . a . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . . . . . , . . . . . O . . . |
$$ | . . . . . O . . . X . . . X . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
Black to play. How does LZ see Black and what's LZ favourite continuations? I played at 'a' by the way.

Re: LZ help for position

Posted: Tue Oct 23, 2018 5:30 pm
by Tryss
Game is still equal (black winrate is at 46.9%)

LZ wants to attach at C4, and her second choices are to make a shimari in the top left (both directions are considered C13, C14, F17 or G17).

Your move is not considered at all. When played, it's a -3% winrate drop, and LZ expects white to tenuki. My guess is that the right side is not very important right now (it's kinda flat?)

LZ first main line
Click Here To Show Diagram Code
[go]$$Bc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . X . . X . . |
$$ | . . . X . . . . . , . . . . . X . O . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . , . . . . . , . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . O . . |
$$ | . . c . . . . . . . . . . . . . . . . |
$$ | . . 1 O . . . . . , . . . . . O . . . |
$$ | . 3 2 . . O . . . X . . . X . . . . . |
$$ | . a b . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
After that, white has several options at a, b or c. In the original sequence (when evaluating the board before 1) she choose b, but after over 15k playouts before playing 4, the first choice switched to a.

Final position (with possible white following as x) :
Click Here To Show Diagram Code
[go]$$Bc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . X . . X . . |
$$ | . . . X . . . . . , . . . . . X . O . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . 7 . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . , . . . . . , . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . x . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . X . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . 5 1 O . . . . . , . . . . . O . . . |
$$ | . 3 2 6 . O . . . X . x . X . . . . . |
$$ | . 4 . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
The board is still equal (47% for black)

Re: LZ help for position

Posted: Tue Oct 30, 2018 7:53 pm
by Pippen


I am interested in the sequenze 8-32 purely for White.

1. Where is the first time Leela would have played differently as White.
2. After 32, if Black plays F8 White cannot kill the 6 black stones inside IMO. What sequenze would Leela play to F8 and how does it see White then?

Re: LZ help for position

Posted: Wed Oct 31, 2018 1:38 pm
by Waylon
Pippen wrote: I am interested in the sequenze 8-32 purely for White.

1. Where is the first time Leela would have played differently as White.
2. After 32, if Black plays F8 White cannot kill the 6 black stones inside IMO. What sequenze would Leela play to F8 and how does it see White then?
Starting with move 8, the picture shows which moves Leela Zero would play differently for White. (net #184)
The main line suggested by Leela can be seen in the small board on the left.
Analyse.jpg
Analyse.jpg (1.23 MiB) Viewed 17400 times

Re: LZ help for position

Posted: Sat Nov 03, 2018 6:50 am
by Pippen
Here's a question:

I run a virtual machine on the Google Cloud. I downloaded Lizzie, but it won't start though I've installed the latest Java. Any idea what could be the problem? Otherwise I try Sabaki because that is running, but takes more time to install Leela.

Re: LZ help for position

Posted: Thu Jan 10, 2019 9:52 am
by Pippen
What I find interesting is that LZ doesn't think at all that 'a' is a must-move, but in fact LZ tends to tenuki there while I see basically NO good players doing it.
Click Here To Show Diagram Code
[go]$$B
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . a . . |
$$ | . . . . . . . . . . . . . X . . X O . |
$$ | . . . X . . . . . , . . . . . X O O . |
$$ | . . . . . . . . . . . . . . . X X O . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . O . . . |
$$ | . . . . . . . . . X . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Posted: Thu Jan 10, 2019 10:09 am
by EdLee
Hi Pippen,

Happy 2019. :)

I have no access to any bot except for DeepLeela.com on an ancient iphone5. :mrgreen:

DL: (a) 49.2%; (b) 49.6%
Another instance: (a) 50%; (b) 49.4%
Click Here To Show Diagram Code
[go]$$B (a)
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . 3 . . |
$$ | . . . . . . . . . . . . . X . . X O . |
$$ | . . . X . . . . . , . . . . . X O O . |
$$ | . . . . . . . . . . . . . . . 1 X 2 . |
$$ | . . 8 9 . . . . . . . . . . . 4 O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . 6 . . . . . . . . . . . . . . . |
$$ | . . O , . . 7 . . , . . . . . O . . . |
$$ | . . . . 5 . . . . X . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
Click Here To Show Diagram Code
[go]$$B (b)
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . 6 . . . |
$$ | . . . . . . . . . . . . . X . 4 X O . |
$$ | . . . X . . . . . , . . . . 5 X O O . |
$$ | . . . . . . . . . . . . . 0 . 2 X 1 . |
$$ | . . . . . . . . . . . . . . . 8 O 3 . |
$$ | . . . . . . . . . . . . . . . 9 7 . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . O . . . |
$$ | . . . . . . . . . X . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]
I'm curious if you get similar results on your device(s) ? :)

Re: LZ help for position

Posted: Thu Jan 10, 2019 10:44 am
by jlt
LZ157 doesn't consider "b", but once played, "a" and "b" seem roughly equal.
a.JPG
a.JPG (90.22 KiB) Viewed 12868 times
b.JPG
b.JPG (91.37 KiB) Viewed 12868 times
b2.JPG
b2.JPG (100.71 KiB) Viewed 12868 times

Re: LZ help for position

Posted: Thu Jan 10, 2019 10:59 am
by Pippen
So basically it's a coin flip? Also: How good or bad are 42-43 for Black? With Leela 11 it would indicate a bad position for black.

@EdLee: My Leela 11 on my computer gets for 'a' = 50.99 and 'b' = 50.08 with like a 2h calculation, so it differs a little bit.

Re: LZ help for position

Posted: Thu Jan 10, 2019 1:12 pm
by jaca
Lizzie reminds me of my mother. When i was little, she often would tell me what i should have done instead of what i did.

But if i ever asked "why should i have done that instead of this?", i only ever got the only answer that Lizzie gives:

"BECAUSE I SAY SO!!!!"

Re: LZ help for position

Posted: Thu Jan 10, 2019 1:53 pm
by Pippen
What I find interesting is that LZ doesn't think at all that 'a' is a must-move, but in fact LZ tends to tenuki there while I see basically NO good players doing it.
Click Here To Show Diagram Code
[go]$$B
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . a . . |
$$ | . . . . . . . . . . . . . X . . X O . |
$$ | . . . X . . . . . , . . . . . X O O . |
$$ | . . . . . . . . . . . . . . . X X O . |
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . O , . . . . . , . . . . . O . . . |
$$ | . . . . . . . . . X . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]

Re: LZ help for position

Posted: Thu Jan 10, 2019 4:20 pm
by jaca
Pippen wrote:What I find interesting is that LZ doesn't think at all that 'a' is a must-move,..
'a' is yose - both black and white are strong and not surrounded with plenty of open space to live in, so neither of them needs 'a'.

the only reason you would want to play a gote yose move is that it's the biggest move on the board.

you could try comparing the positions after either side plays 'a' to figure out how big it is

and then compare the overall position after black 'a' with that of a different move somewhere else

maybe influence maps would help you see the big picture

Re: LZ help for position

Posted: Fri Jan 11, 2019 12:55 am
by jlt
Pippen wrote:How good or bad are 42-43 for Black? With Leela 11 it would indicate a bad position for black.
On an empty board, LZ157 thinks that Black's winrate is 46.4%, so 42% or 43% during the opening is slightly bad. According to her, the first mistake (about -2%) is :b5: at K3, she prefers to approach the lower left corner.
Pippen wrote: What I find interesting is that LZ doesn't think at all that 'a' is a must-move, but in fact LZ tends to tenuki there while I see basically NO good players doing it.
That's because the lower left corner hasn't been approached yet. My impression is that LZ values approaching komokus, or making a shimari from a komoku, more than humans.

Re: LZ help for position

Posted: Fri Jan 11, 2019 3:27 am
by Uberdude
Pippen wrote:What I find interesting is that LZ doesn't think at all that 'a' is a must-move, but in fact LZ tends to tenuki there while I see basically NO good players doing it.
Click Here To Show Diagram Code
[go]$$Bc
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . a . . |
$$ | . . . . . . . . . . . . . X . . X O . |
$$ | . . . X . . . . . , . . . . . X O O . |
$$ | . . . . . . . . . . . . . . . X X O . |
$$ | . . . . . . . . . . . . . . . b O . . |
$$ | . . . . . . . . . . . . . . . . c . . |
$$ | . . . . . . . . . . . . . . . . e . . |
$$ | . . . . . . . . . . . . . . . . d . . |
$$ | . . . , . . . . . , . . . . . , . . . |
$$ | . . . . . . . . . . . . . . . . . . . |[/go]
I see black playing tenuki, after first exchanging b for c, quite a lot recently in both AI and pro games (e.g. http://www.go4go.net/go/games/sgfview/81834). You can see the evolution of thinking in this joseki over the last few years:
- before AI s16 slide instead of r16 attach was common, AI made slide practically disappear (it's easy to tenuki and weak shape)
- old thinking for the attach joseki was after black extends at a (it's not just yose as jaca said: it is a base for both players) white group needs a base (if tenuki black pincer at r12 is powerful) so white extends to d or e.
- if white extended to d then sooner or latter black can exchange b for c. He might not want to do it too early because it loses some other aji (e.g. if pincer at r10 then s13 peep can be an aim), but it's more likely to be a black sente because white pushing there is quite easy to ignore for black and also makes white's extension too close and inefficient shape.
- (sometimes white would extend just one space to e, then she can answer push with hane, but it's a bit tight)
- so white pushes at b instead of an extension is a move bots liked and humans play a lot too now. The idea is to take that key sente point for black and attempt to build a bigger and more efficient right side. black can play a pincer on the right or peep s13 and then there can be complex fighting (taking the liberty makes q17 cut a useful resource for white to exploit) with weak groups for both but at least white isn't in an inefficient slack shape without black having to do anything.
- so now black says "hey! q14 is a good move for you isn't it, it's a good move for me too" and plays it instead of descending in the corner. This makes the most sense when black is developing the top side. Because the push extend exchange helps white make a base on the right side black descend in the corner is no longer as important: black will usually tenuki or continue pushing / knight move in the centre.

(jlt also made a good point about there being an unapproached 3-4 being available, that being bigger than 4-4 corners is a basic opening theory lesson from LZ)