It is currently Thu Mar 28, 2024 8:55 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 32 posts ]  Go to page Previous  1, 2
Author Message
Offline
 Post subject: Re: LZ v Golaxy
Post #21 Posted: Mon Apr 23, 2018 11:37 am 
Lives with ko

Posts: 136
Liked others: 47
Was liked: 21
Rank: KGS 6 dan
macelee wrote:
It turns out that Golaxy has another aspect that is more similar to human than to other bots. Instead of giving a winning percentage, Golaxy gives its positional judgement in points (e.g. black leads by 6.2 points) as if it can count.
So it turns out that if you like territory and points you should play 3-3 in all the corners you can. Which makes sense :)


This post by johnsmith was liked by 2 people: abcd_z, Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #22 Posted: Wed Apr 25, 2018 7:34 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Golaxy is playing a series of 30 no-komi games against pros (I guess on yikeweiqi server), macelee has put them in go4go. Humans won 2 of the 27 he has (well done Kang Dongyun and Yun Chanhee!). It beat Park Junghwan, I think the final game will be against Ke Jie?
http://www.go4go.net/go/games/byplayer/1992


This post by Uberdude was liked by: sorin
Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #23 Posted: Wed Apr 25, 2018 8:58 am 
Dies in gote

Posts: 23
Liked others: 1
Was liked: 9
Golaxy, while adopting AlphaGo‘s basic architecture, has brought innovations into feature combination, model structure, and MCTS calculation methods。 It manages to complete model training with less computing resources or training samples, yet with the potential to achieve the highest possible Go level by AI。

Golaxy tries to play Go in a style that is easier for human players to understand and to learn。 It can play Go on boards of all sizes and all komi settings。 Golaxy will not concede even when leading。


This post by Hane was liked by 2 people: Bill Spight, thirdfogie
Top
 Profile  
 
Offline
 Post subject:
Post #24 Posted: Wed Apr 25, 2018 9:03 am 
Honinbo
User avatar

Posts: 8859
Location: Santa Barbara, CA
Liked others: 349
Was liked: 2076
GD Posts: 312
Quote:
Golaxy will not concede even when leading。
Seems typo: when losing.

Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #25 Posted: Wed Apr 25, 2018 9:07 am 
Lives in gote

Posts: 388
Liked others: 416
Was liked: 198
Uberdude wrote:
Golaxy is playing a series of 30 no-komi games against pros (I guess on yikeweiqi server), macelee has put them in go4go. Humans won 2 of the 27 he has (well done Kang Dongyun and Yun Chanhee!). It beat Park Junghwan, I think the final game will be against Ke Jie?
http://www.go4go.net/go/games/byplayer/1992


I noticed in the Yun Chanhee that Golaxy's opening was very unusual- two moves in 5x7, and two moves in 6x6, while black's moves were all orthodox - I wonder if it was the human operator choosing the first 4 bot moves manually just to make the game more interesting?

_________________
Sorin - 361points.com

Top
 Profile  
 
Offline
 Post subject: Re:
Post #26 Posted: Wed Apr 25, 2018 9:09 am 
Lives in gote

Posts: 388
Liked others: 416
Was liked: 198
EdLee wrote:
Quote:
Golaxy will not concede even when leading。
Seems typo: when losing.


I think it is meant to be: "will not play slack/stupid-looking moves when ahead".

_________________
Sorin - 361points.com

Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #27 Posted: Wed Apr 25, 2018 9:16 am 
Dies in gote

Posts: 23
Liked others: 1
Was liked: 9
source : http://sports.sina.com.cn/go/2018-04-23/doc-ifzqvvrz7515857.shtml

Top
 Profile  
 
Offline
 Post subject:
Post #28 Posted: Wed Apr 25, 2018 9:43 am 
Honinbo
User avatar

Posts: 8859
Location: Santa Barbara, CA
Liked others: 349
Was liked: 2076
GD Posts: 312
Thanks, Sorin; makes more sense that way.
Quote:
Golaxy will not concede even when leading。
An issue here is the definition of concede:
- in terms of the final points difference; or
- in terms of the winrate.

They're implicitly (and selectively) using the former meaning, which is quite human-centric, and a pre-AlphaGo idea.
Whereas AFAIK AlphaGo/AlphaZero-class bots also don't concede in the latter meaning, when ahead. ( AG's winrate monotonically non-decreasing, when AG is ahead. )

Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #29 Posted: Thu Apr 26, 2018 1:51 am 
Lives in sente

Posts: 923
Location: UK
Liked others: 72
Was liked: 479
Rank: 5 dan
KGS: macelee
sorin wrote:
I noticed in the Yun Chanhee that Golaxy's opening was very unusual- two moves in 5x7, and two moves in 6x6, while black's moves were all orthodox - I wonder if it was the human operator choosing the first 4 bot moves manually just to make the game more interesting?


I heard that there was some software configuration issues that resulted the rather unusual opening. Also note that there was a very large-scale seki - not sure if that caused difficulties to the AI.

Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #30 Posted: Thu Apr 26, 2018 4:21 am 
Lives in sente

Posts: 923
Location: UK
Liked others: 72
Was liked: 479
Rank: 5 dan
KGS: macelee
macelee wrote:
sorin wrote:
Also note that there was a very large-scale seki - not sure if that caused difficulties to the AI.


Golaxy has completed the 30-game series. The total score is 28-2.

Apparently Golaxy has no problem with seki! In fact in game 30 against Zhou Ruiyang, they managed to create a 4-group seki situation. This remind me a recent LG cup game viewtopic.php?p=229331#p229331

Click Here To Show Diagram Code
[go]$$W
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . O . . |
$$ | . . . X . . . . . . . O O O X . X O . |
$$ | . . X X O . . X . O . X O X X X O . . |
$$ | X . O O X X X X O , X . X . X X O . . |
$$ | X X . X O O O X O . . X X X O X O . . |
$$ | X O X X X . . O O . X . . . O O X O . |
$$ | X O O O X O . . . . X O . O . . X O . |
$$ | O O O X O . . . . . X O . O X O X . . |
$$ | O X X X O O . . . O O X . X O . X . . |
$$ | O O . X O . . . . , . . X . O , . . . |
$$ | O X X X O . . . . . . . . . . . . . . |
$$ | O O X X O . . . . . . . . . . . . . . |
$$ | . X . X O O O . . . . . . . . X X . . |
$$ | X X X O O X O . . . . . . . . . O . . |
$$ | X O X X O X O . . . . . . . . . . . . |
$$ | X O O O X X O . . , . . X . . , . . . |
$$ | X O . O X O O . . X . . . . X . O . . |
$$ | O O X O X O O X . . . . . . . . . . . |
$$ | . X X X X O . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


This post by macelee was liked by: thirdfogie
Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #31 Posted: Thu Apr 26, 2018 11:35 am 
Beginner

Posts: 12
Liked others: 5
Was liked: 0
Rank: 15k
Hane wrote:
Golaxy, while adopting AlphaGo‘s basic architecture, [...] manages to complete model training with less computing resources or training samples


You know what that says to me? He used transfer learning from Leela Zero's weights.

Top
 Profile  
 
Offline
 Post subject: Re: LZ v Golaxy
Post #32 Posted: Thu Apr 26, 2018 4:36 pm 
Gosei

Posts: 1590
Liked others: 886
Was liked: 527
Rank: AGA 3k Fox 3d
GD Posts: 61
KGS: dfan
abcd_z wrote:
You know what that says to me? He used transfer learning from Leela Zero's weights.

I don't know about the weights, but I think at the very least it would be silly not to use Leela Zero's millions of quality self-play games when training any network.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 32 posts ]  Go to page Previous  1, 2

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