Life In 19x19
http://prod.lifein19x19.com/

Deep Zen vs. Three Japanese Pros in 9X9
http://prod.lifein19x19.com/viewtopic.php?f=18&t=7189		 Page 1 of 2
Author:  ez4u [ Tue Nov 13, 2012 9:36 pm ]
Post subject:  Deep Zen vs. Three Japanese Pros in 9X9
The next computer vs. pro event in Japan will take place 2012-11-25. Three pros: So Yokoku 8p, Ohashi Hirofumi 5p, and Ichiriki Ryo 2p will each play two even 9X9 games against Zen. Chinese rules, and komi will be seven points (I assume this means 3.5 points as we are used to expressing with Chinese rules). There will be twenty minutes main time followed by 30 seconds byo yomi.
Author:  Rémi [ Thu Nov 15, 2012 2:43 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
ez4u wrote:
The next computer vs. pro event in Japan will take place 2012-11-25. Three pros: So Yokoku 8p, Ohashi Hirofumi 5p, and Ichiriki Ryo 2p will each play two even 9X9 games against Zen. Chinese rules, and komi will be seven points (I assume this means 3.5 points as we are used to expressing with Chinese rules). There will be twenty minutes main time followed by 30 seconds byo yomi.


The komi is very probably 7. Matches between programs indicate that 7.5 gives more winning chances to White, and 6.5 gives more winning chances to Black. So 7 has become the preferred komi in computer-go tournaments.

Rémi
Author:  luigi [ Mon Nov 26, 2012 12:08 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
How did those games turn out? I checked the link, but I don't speak Japanese.
Author:  Cassandra [ Mon Nov 26, 2012 12:40 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
According to http://entcog.c.ooco.jp/entcog/event/event20121125.html Zen lost all the six games.
Author:  Splatted [ Mon Nov 26, 2012 12:44 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
Looks like Zen lost them all. You can see the actual games at

http://entcog.c.ooco.jp/entcog/event/event20121125.html

Edit: In case it's useful to know White = 白 and black =　黒
Author:  luigi [ Mon Nov 26, 2012 1:06 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
Thank you, guys.

Surprising results for 9x9. I had expected Zen to win at least a couple of games.
Author:  luigi [ Mon Nov 26, 2012 1:09 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
Which games were played by which players in the link you gave me?
Author:  SoDesuNe [ Mon Nov 26, 2012 3:27 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
If I remember correctly, Zen tends to lose more often on 9*9 compared to 19*19. I'm curious as to how this comes about. Seems as the machine isn't as "optimal" in reading as it is in chess? Or maybe some other things factor in?
Author:  Splatted [ Mon Nov 26, 2012 3:58 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
1+4. Ichiriki Ryo 2p
2+5.Ohashi Hirofumi 5p
3+6.So Yokoku 8p
Author:  Splatted [ Mon Nov 26, 2012 4:00 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
SoDesuNe wrote:
If I remember correctly, Zen tends to lose more often on 9*9 compared to 19*19. I'm curious as to how this comes about. Seems as the machine isn't as "optimal" in reading as it is in chess? Or maybe some other things factor in?


That's really wierd. Does anyone know why?
Author:  jts [ Mon Nov 26, 2012 8:08 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
Why are people surprised? Isn't zen rated 4 or 5d on kgs?
Author:  gowan [ Mon Nov 26, 2012 8:13 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
What sort of machine was Zen running on? It makes a difference in how strongly these programs play.
Author:  p2501 [ Tue Nov 27, 2012 1:08 am ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
jts wrote:
Why are people surprised? Isn't zen rated 4 or 5d on kgs?

Yeah but on 19x19 I guess. I think Computer-Engines are stronger on 9x9.
Author:  schawipp [ Tue Nov 27, 2012 3:32 am ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
Splatted wrote:
1+4. Ichiriki Ryo 2p
2+5.Ohashi Hirofumi 5p
3+6.So Yokoku 8p


Thanks for the information. In the first game, I do not really understand why white (Zen?) resigned. If after Black-33 white plays e. g. on J3 the LR corner should be simply dead, and 34. G7 - G8 doesn't seem to work for B. In result, W looks rather big. Obviously I am missing something crucial... ? Thank you for any explanations.
Author:  ez4u [ Tue Nov 27, 2012 3:59 am ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
gowan wrote:
What sort of machine was Zen running on? It makes a difference in how strongly these programs play.

From the user information of 'Zen19D' on KGS, by its author, Hideki Katoh:
"Zen19D is a version of Zen19, runs on a mini-cluster of 4 PCs (a dual 6-core Xeon X5680/4.4 GHz, a 6-core i7 3930K/4.2 GHz, a 6-core Xeon W3680/4 GHz and a 4-core i7 920/3.4 GHz) connected via a GbE LAN. Although Zen19 is a prototype of a commercial product, Zen19D is being developed for academic research by team DeepZen, a joint project of ZenAuthor and me."

This is the version used for these competitions.
Author:  logan [ Tue Nov 27, 2012 4:03 am ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
schawipp wrote:
Splatted wrote:
1+4. Ichiriki Ryo 2p
2+5.Ohashi Hirofumi 5p
3+6.So Yokoku 8p


Thanks for the information. In the first game, I do not really understand why white (Zen?) resigned. If after Black-33 white plays e. g. on J3 the LR corner should be simply dead, and 34. G7 - G8 doesn't seem to work for B. In result, W looks rather big. Obviously I am missing something crucial... ? Thank you for any explanations.

Click Here To Show Diagram Code
[go]$$Bm33 BP:3 WP:0
$$ -------------------
$$ | . . . . . . . . . |
$$ | . . . Q . . 5 . . |
$$ | . . X X Q Q , . . |
$$ | . . . . X Q 1 3 . |
$$ | . . . . X X O O O |
$$ | . . X X . O X X O |
$$ | . . O X . O O X 2 |
$$ | . . . . O . X 4 X |
$$ | . . . . . . X X X |
$$ -------------------[/go]

Black wins capture race.

Click Here To Show Diagram Code
[go]$$Wm34 BP:3 WP:0
$$ -------------------
$$ | . . . . . . . . . |
$$ | . . . O . . . . . |
$$ | . . X X O O 1 3 . |
$$ | . . . . X O X 2 7 |
$$ | . . . . X X O O O |
$$ | . . X X . O X X O |
$$ | . . O X 8 O O X 9 |
$$ | . . . 4 O 0 Y . Y |
$$ | . . . 6 5 . Y Y Y |
$$ -------------------[/go]

Black saves his bottom stones. Is ahead in points.
Author:  schawipp [ Tue Nov 27, 2012 5:32 am ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
logan wrote:
Click Here To Show Diagram Code
[go]$$Bm33 BP:3 WP:0
$$ -------------------
$$ | . . . . . . . . . |
$$ | . . . Q . . 5 . . |
$$ | . . X X Q Q , . . |
$$ | . . . . X Q 1 3 . |
$$ | . . . . X X O O O |
$$ | . . X X . O X X O |
$$ | . . O X . O O X 2 |
$$ | . . . . O . X 4 X |
$$ | . . . . . . X X X |
$$ -------------------[/go]

Black wins capture race.
...


Thanks, that's really interesting (for me at least :)). That B wins the capturing race, was not obvious for me. I thought of some variations as the following:

Click Here To Show Diagram Code
[go]$$Wm38 BP:3 WP:0
$$ -------------------
$$ | . . . . . . 8 7 . |
$$ | . . 4 O 3 . X 5 6 |
$$ | . . X X O O 1 2 . |
$$ | . . . . X O X X 9 |
$$ | . . . . X X O O O |
$$ | . . X X . O . . O |
$$ | . . O X . O O . O |
$$ | . . . . O . X O X |
$$ | . . . . . . X X X |
$$ -------------------[/go]


Click Here To Show Diagram Code
[go]$$Bm47 BP:3 WP:0
$$ -------------------
$$ | . . . . . . X 3 1 |
$$ | . . X O O . X 2 X |
$$ | . . X X O O O X 4 |
$$ | . . . . X O X X O |
$$ | . . . . X X O O O |
$$ | . . X X . O . . O |
$$ | . . O X . O O . O |
$$ | . . . . O . X O X |
$$ | . . . . . . X X X |
$$ -------------------[/go]


And after the squeeze the B stones have only 2 liberties against 3 liberties left. Thinking against about it, a better way is obviously the following:

Click Here To Show Diagram Code
[go]$$Wm38 BP:3 WP:0
$$ -------------------
$$ | . . . . . . . . . |
$$ | . . . O 3 . X 4 . |
$$ | . . X X O O 1 2 . |
$$ | . . . . X O X X . |
$$ | . . . . X X O O O |
$$ | . . X X . O . . O |
$$ | . . O X . O O . O |
$$ | . . . . O . X O X |
$$ | . . . . . . X X X |
$$ -------------------[/go]


Here the B stones have 6 vs. 4 liberties which should be easy. Are there other possibilities than black-41? Thank you in advance for correction of any flaws (I am starting to get serious doubts about my reading capabilities... :scratch: ).

Edit: If in my first diagram, B plays 43 at 44, this seems also a possible solution.

Quote:
(I am starting to get serious doubts about my reading capabilities... :scratch: )

No further comments on that ;-)
Author:  Mike Novack [ Tue Nov 27, 2012 7:36 am ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
ez4u wrote:
gowan wrote:
What sort of machine was Zen running on? It makes a difference in how strongly these programs play.

From the user information of 'Zen19D' on KGS, by its author, Hideki Katoh:
"Zen19D is a version of Zen19, runs on a mini-cluster of 4 PCs (a dual 6-core Xeon X5680/4.4 GHz, a 6-core i7 3930K/4.2 GHz, a 6-core Xeon W3680/4 GHz and a 4-core i7 920/3.4 GHz) connected via a GbE LAN. .......


No, Gowan's question remains unanswered until we confirm that the hardware Zen19D usually runs on was the hardware available for this trial.
Author:  Rémi [ Tue Nov 27, 2012 11:37 am ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
You can find more information in the computer-go mailing list:
http://dvandva.org/pipermail/computer-g ... 05544.html
http://www.asahi.com/igo/topics/TKY201211270576.html

Rémi
Author:  luigi [ Tue Nov 27, 2012 1:43 pm ]
Post subject:  Re: Deep Zen vs. Three Japanese Pros in 9X9
Rémi wrote:
You can find more information in the computer-go mailing list:
http://dvandva.org/pipermail/computer-g ... 05544.html
http://www.asahi.com/igo/topics/TKY201211270576.html

Rémi

From the first link:

Quote:
Zen had chances to draw in some games in Black but selected much
risky (actually losing) moves. I guess this caused by implementing
draws by adding a third value, 0.5, to UCB. To play a draw
move, all better looking moves (by prior) have to be refused, or proved
worse than 0.5. This could take pretty long time in some positions.
#Faking komi to 6.5 might help but a better solution possible?

I guess this explains why Zen performed worse than expected. After all, computers have already beaten pros on 9x9 before.
Page 1 of 2 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/