It is currently Thu Mar 28, 2024 7:28 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 13 posts ] 
Author Message
Offline
 Post subject: Is not defending them effective against programs?
Post #1 Posted: Thu May 30, 2019 9:06 pm 
Beginner

Posts: 14
Liked others: 0
Was liked: 0
Rank: AGA 1 kyu
I have found that programs get confused when you do things like making approach moves and not responding to the programs reply in the corner, or putting in stone in the middle of the programs extension and then ignoring its reply. Also, sacrificing small numbers of stones seems to work.

The programs seem to get confused by stones they are not sure are live or not.

Are there weaknesses of programs? Do they still work against the versions that are beating professionals?

Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #2 Posted: Thu May 30, 2019 10:02 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
betgo wrote:
I have found that programs get confused when you do things like making approach moves and not responding to the programs reply in the corner, or putting in stone in the middle of the programs extension and then ignoring its reply. Also, sacrificing small numbers of stones seems to work.

The programs seem to get confused by stones they are not sure are live or not.

Are there weaknesses of programs? Do they still work against the versions that are beating professionals?


Yes, programs have weaknesses. See Beat Elf https://lifein19x19.com/viewtopic.php?f=15&t=16641 . Elf is one of today's top programs that beats pros.

However, the weaknesses you mention are unlikely to confuse today's top programs. In their training they have played against opponents that do those things. That is, against weak versions of themselves. :)

As for local life and death situations, even top bots can err, because they read the whole board. Given enough time they can solve all local life and death situations, but, depending on other options on the board that they explore, in any sufficiently complicated or unusual case they can err. There are also whole board positions where they will leave a local position undecided, because they deem it more important to play elsewhere. That doe not mean that they have made a mistake. :)

If you are running into all of these weaknesses, you have a program that is weak, given the conditions that Uberdude has mentioned.

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

Visualize whirled peas.

Everything with love. Stay safe.


This post by Bill Spight was liked by: Bonobo
Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #3 Posted: Fri May 31, 2019 2:56 am 
Gosei
User avatar

Posts: 1753
Liked others: 177
Was liked: 491
With which bots did you observe these weaknesses? Even games or handicap games? How many stones handicap, and how many playouts per move?

Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #4 Posted: Fri May 31, 2019 4:03 am 
Dies in gote

Posts: 53
Liked others: 3
Was liked: 33
Rank: KGS 2k
Bots can have weaknesses yes, but its important to also know in what kind of environment the weaknesses appear.

In a lost game, or high handicap game, the bot might just ignore and/or play random moves if it feels that there is no way to win.

For example, if the bot is loosing by 20 points, it might suicide its own group, making it loose by even more. This is because suiciding and counting are both just as sure losses in the eyes of the bot.


This post by Aram was liked by: Bonobo
Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #5 Posted: Fri May 31, 2019 4:22 am 
Beginner

Posts: 14
Liked others: 0
Was liked: 0
Rank: AGA 1 kyu
jlt wrote:
With which bots did you observe these weaknesses? Even games or handicap games? How many stones handicap, and how many playouts per move?


Not on the really strong configurations. Against regular leela and leelaz on my laptop at like 500 playouts/move at a 4 stone handicap. I would assume the programs / configurations beating top pros don't make obvious mistakes like that. However, I was wondering if plays like that might still give an advantage in causing the program to pay suboptimally. What I noticed was not usually blatant mistakes but that it would not play as well if it was confused about whether stones were live.

Regular leela has weaknesses like not understanding seki, and thinking it has territory.

Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #6 Posted: Fri May 31, 2019 4:28 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
I think the biggest weakness of Leela Zero wasn't mentioned: ladders, particularly at low playouts and the older networks.


This post by Uberdude was liked by: Bonobo
Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #7 Posted: Fri May 31, 2019 4:36 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
betgo wrote:
However, I was wondering if plays like that might still give an advantage in causing the program to pay suboptimally. What I noticed was not usually blatant mistakes but that it would not play as well if it was confused about whether stones were live.


Is your purpose to beat the program or to learn from it?

_________________
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: Is not defending them effective against programs?
Post #8 Posted: Fri May 31, 2019 5:04 am 
Dies in gote

Posts: 53
Liked others: 3
Was liked: 33
Rank: KGS 2k
betgo wrote:
jlt wrote:
With which bots did you observe these weaknesses? Even games or handicap games? How many stones handicap, and how many playouts per move?


Not on the really strong configurations. Against regular leela and leelaz on my laptop at like 500 playouts/move at a 4 stone handicap. I would assume the programs / configurations beating top pros don't make obvious mistakes like that. However, I was wondering if plays like that might still give an advantage in causing the program to pay suboptimally. What I noticed was not usually blatant mistakes but that it would not play as well if it was confused about whether stones were live.

Regular leela has weaknesses like not understanding seki, and thinking it has territory.



You should not make any assumptions on strengths or weaknesses when running at 4 handicap stones, the program simply is not designed to play at such handicap.
Also remember that the program is using Tromp-Taylor rules (similar to Chinese), not Japanese.

Handicap play is something that some other programs are currently trying to solve, but i do not think its currently easy to download any of them.

Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #9 Posted: Fri May 31, 2019 12:47 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
You may enjoy taking handicap stones from Deep Leela at https://www.deepleela.com . It is Leela 11, which is weaker than Leela Zero, but was trained on human play, so it plays handicap games in a more human manner. :)

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

Visualize whirled peas.

Everything with love. Stay safe.


This post by Bill Spight was liked by: Bonobo
Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #10 Posted: Fri May 31, 2019 5:24 pm 
Beginner

Posts: 14
Liked others: 0
Was liked: 0
Rank: AGA 1 kyu
Bill Spight wrote:
You may enjoy taking handicap stones from Deep Leela at https://www.deepleela.com . It is Leela 11, which is weaker than Leela Zero, but was trained on human play, so it plays handicap games in a more human manner. :)


Is there a way to configure deepleela not to show suggested moves for its opponent?

I would think the handicap issue would be important as a lot of people would like to play / train against a strong program, but not that many are 4D+ Japanese / US scale.

Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #11 Posted: Fri May 31, 2019 6:20 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
betgo wrote:
Bill Spight wrote:
You may enjoy taking handicap stones from Deep Leela at https://www.deepleela.com . It is Leela 11, which is weaker than Leela Zero, but was trained on human play, so it plays handicap games in a more human manner. :)


Is there a way to configure deepleela not to show suggested moves for its opponent?


I don't know. Maybe someone here does, or you could ask.

Quote:
I would think the handicap issue would be important as a lot of people would like to play / train against a strong program, but not that many are 4D+ Japanese / US scale.


Me, too.

_________________
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: Is not defending them effective against programs?
Post #12 Posted: Fri May 31, 2019 10:16 pm 
Lives in gote

Posts: 388
Liked others: 416
Was liked: 198
betgo wrote:
I have found that programs get confused when you do things like making approach moves and not responding to the programs reply in the corner, or putting in stone in the middle of the programs extension and then ignoring its reply.


How does the programs confusion manifest itself?

_________________
Sorin - 361points.com

Top
 Profile  
 
Offline
 Post subject: Re: Is not defending them effective against programs?
Post #13 Posted: Sat Jun 01, 2019 7:01 am 
Gosei

Posts: 1733
Location: Earth
Liked others: 621
Was liked: 310
No not defending is not effective against programs.

Effective in your sense is playing them on obsolete hardware.

It is like binding the legs of a world class runner and then cheer because you are walking faster than the athlete.

You can not make rational conclusions if you set unrational constraints.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 13 posts ] 

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