Life In 19x19

Posted: **Mon Sep 11, 2017 8:20 pm**

Marathon wrote:In creating software to play go, it can be difficult to create code that can accurately decide which groups are alive and which are dead at the ends of games. One way to help a bot decide is to have the bot play inside its own territory until each eye is only 1 or 2 points in size.

Click Here To Show Diagram Code: [go]$$Bcm1 dead? $$ --------------------------------------- $$ | . . . . . . . . . . . . . . . . O X . | $$ | . . . . . . . . . . . . . . . . O X . | $$ | . . . . . . . . . . . . . . . . O X 3 | $$ | . . . , . . . . . , . . . . . , O X 1 | $$ | . . . . . . . . . . . . . . . . O X X | $$ | . . . . . . . . . . . . . . . . O O O | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . , . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ ----------------------------------------[/go]

Posted: **Mon Sep 11, 2017 8:44 pm**

$$B var 1. B dead?
$$ ---------------
$$ | . . 3 1 X O .
$$ | X X X X X O .
$$ | O O O O O O .
$$ | . . . , . . .
Click Here To Show Diagram Code
[go]$$B var 1. B dead? $$ --------------- $$ | . . 3 1 X O . $$ | X X X X X O . $$ | O O O O O O . $$ | . . . , . . .[/go]

If (a)...(d) are the only legal moves remaining on the board:

Click Here To Show Diagram Code: [go]$$B $$ --------------- $$ | a b c d X O ? $$ | X X X X X O ? $$ | O O O O O O ? $$ | ? ? ? ? ? ? ?[/go]

- Randombot has exactly 25% chance to play

on one of them (if Pass is not an option) ;
( or exactly 20% chance of one of [a,b,c,d,Pass] )
- AlphaGo tends to perform slightly better.

One way to help a bot decide is to have the bot play inside its own territory until each eye is only 1 or 2 points in size.

Depending on the bot, this may or may not be part of the logic, and there may be additional criterias to decide a move. These other criterias may work in conjunction with the above technique, or they may override it in certain situations. So yes, var 1 is entirely possible with a Randombot.

Posted: **Mon Sep 11, 2017 9:16 pm**

Thanks, Ed.

My post was partly sarcastic, and partly in hopes to trigger a response that gave more details about this class of algorithm:-)

Posted: **Mon Sep 11, 2017 9:23 pm**

Hi Kirby, You're welcome.

It can be tricky sometimes to tell the tone of voice online.
Also, there may be a range of different levels in understanding of programming in the viewers, so sometimes it's OK to go back to the basics.

Posted: **Mon Sep 11, 2017 9:46 pm**

EdLee wrote:Hi Kirby, You're welcome.

It can be tricky sometimes to tell the tone of voice online.
Also, there may be a range of different levels in understanding of programming in the viewers, so sometimes it's OK to go back to the basics.

Agreed. Now, with a serious tone, does anyone know how a smart bot might select which moves to play within its own territory in order to determine life and death? It's not straightforward to me unless the algorithm does a search to iterate to a terminal state.

Posted: **Mon Sep 11, 2017 10:00 pm**

Now, with a serious tone, does anyone know how a smart bot might select which moves to play within its own territory in order to determine life and death? It's not straightforward to me...

Not only you; seems to be a non-trivial problem: at least CGoban and SmartGo sometimes get confused about the life-and-death (and dames), and still need human intervention in the final scoring. IGS' life-and-death marking is also not 100% automated and needs human clicking. Does anyone know how AlphaGo, Zen, Leela, etc. solved this ?

Posted: **Tue Sep 12, 2017 6:30 am**

Leela, at least, still makes errors marking life and death (though it's getting better), so I don't think it's a solved problem.

Posted: **Tue Sep 12, 2017 9:27 am**

Life and death status assessment by sampling (as done by NN/MC programs) does not "solve" problems. Problems are solved if the solution is verified or verifyable by a mathematical proof or mathematically proven algorithm.

Posted: **Tue Sep 12, 2017 2:53 pm**

Kirby wrote:
Marathon wrote:In creating software to play go, it can be difficult to create code that can accurately decide which groups are alive and which are dead at the ends of games. One way to help a bot decide is to have the bot play inside its own territory until each eye is only 1 or 2 points in size.

$$Bcm1 dead?
$$ ------------
$$ . . O X . |
$$ . . O X . |
$$ . . O X 3 |
$$ . , O X 1 |
$$ . . O X X |
$$ . . O O O |
$$ . . . . . |
$$ . . . . . |
$$
$$ -----------
Click Here To Show Diagram Code
[go]$$Bcm1 dead? $$ ------------ $$ . . O X . | $$ . . O X . | $$ . . O X 3 | $$ . , O X 1 | $$ . . O X X | $$ . . O O O | $$ . . . . . | $$ . . . . . | $$ $$ -----------[/go]

Well, some bots might actually do that. But, more likely, a bot would split single eyes of 3 or more points, so would start with a stone at T7 or T6.

Posted: **Mon Oct 23, 2017 4:44 am**

A better yet about-the-same-difficulty-to-code algorithm will be let opponent play inside first. Then this side try to capture them.

Posted: **Mon Oct 23, 2017 5:34 am**

zinger wrote:The main disadvantage of Chinese rules is that counting the score is more difficult.

What? Seriously?

I've been playing over-the-board games with Japanese scoring for a decade and I still get extremely nervous when counting a close game -- particularly when my opponent seems unwilling to slow the heck down and take care!

It is so trivially easy to make a mistake when moving stones about between territories of the same colour and, in my experience, not everyone is willing to be pedantic or even logical -- often leaving borders saying, "this was a black territory so it's fine if *most* of the border stones are still black. We'll just remember it was black's."

Nah. I hate that. And I cannot comprehend how players can spend three hours on a tournament game and then rush the counting but even at the low SDK ranks in tournaments, they still do.

Posted: **Mon Oct 23, 2017 8:02 am**

@Charlie: Have you ever counted the Chinese way?

Basic procedure:

Reshape territory into rectangular areas.
Note: During this procedure you may throw any and all stones (alive or dead) you want into the bowls, or take stones from bowls and add them to the board, or move stones between areas, to help make rectangular shapes.
Players do not always take care to keep borders one color (just remember whose territory it is) and do not always take it slowly and carefully.
Count the territory and remember this number.
Arrange the stones on the board into piles of 10.
Count the stones
Add the result of 2 and 4

Generally only one color is counted, as the other player then has the remainder of the 361 points. If there is seki, you can fill the seki intersections 50/50 with black/white stones, if there are an odd number of seki intersections, you have to leave one empty and remember that it gives both players half an additional point.

So basically the same risks as territory counting on step 1, but with the additional concern that moving to/from bowls can make it harder to reconstruct things.

And then there's the additional risk that the number from step 2 might be disagreed about (with no way to recount after step 3), or the seki might be forgotten.

In practice, both methods rarely have issues and in 99.9% of cases no dispute arises from the counting though.

Posted: **Mon Oct 23, 2017 8:35 am**

"The Chinese" is a metaphor for "area scoring" and need not include "Chinese counting for area scoring". There are other counting methods, including such that do not alter the position, include a consistency check and allow recounting easily: http://home.snafu.de/jasiek/int.html#Counting

Posted: **Mon Oct 23, 2017 3:06 pm**

Seems a lot easier to me. You only need to count one colour and you can add or remove stones at your convenience so reshaping the territory is a lot easier if it is a diagonal region. (Edit: logically, this could also help make the first number a round number so easier to remember, or to remove as many stones as possible, removing them from the next counting step.)

But I agree, in general. The problem is not the rule set. It's rushed counting. That could happen with both rule sets.

I still think it is wrong to say that counting is *harder* with Chinese rules. Different problems - maybe - but it's not like Japanese rules are not prone to the issues I point out.

Posted: **Tue Oct 24, 2017 1:22 am**

HermanHiddema wrote: Note: During this procedure you may throw any and all stones (alive or dead) you want into the bowls, or take stones from bowls and add them to the board, or move stones between areas, to help make rectangular shapes.
Players do not always take care to keep borders one color (just remember whose territory it is) and do not always take it slowly and carefully.

This all makes me feel a bit queasy.

Life In 19x19

Why do people still use the Chinese when it's clearly bad?

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re:

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba

Re: Why do people still use the Chinese when it's clearly ba