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

AI RYUSEI Tournament 2017, December 9-10
http://www.lifein19x19.com/viewtopic.php?f=18&t=15298		 Page 1 of 2
Author:  pookpooi [ Fri Dec 08, 2017 2:14 am ]
Post subject:  AI RYUSEI Tournament 2017, December 9-10
This tournament is the replacement of UEC Cup
Japanese rule, 30 minutes main time per player, no byoyomi.
300,000, 150,000, 50,000 yen for 1st, 2nd, and 3rd place

List of participant programs

AQ / Yu Yamaguchi (Japan)
Kifuwarabe / Satoshi Takahashi (Japan)
MARU / Atsushi Takeda (Japan)
Deep_ark / HEROZ IGO team (Japan)
otabot / Team Data Kernel (German)
akira / Junya Watanabe (Japan)
Abacus / Tsinghua Univ. (China)
DolBaram / Lim Jaebum (Korea)
ballade / Ichiro Uziie (Japan)
nlp / Kenji Iwai (Japan)
QinoaIgo / Qinoa (Japan)
Rayn / Rayn × Tripleize AI R & D Team (Japan)
Mayoigo / Masaki Murayama (Japan)
Fine Art / Tencent AI Lab (China)
Tianrang / Tianrang (China)
Kaminoitte / 神乎碁技 (USA)
Kugutsu / Tokumoto (Japan)
DeepZenGo / DeepZenGo Project (Japan)
Aya / Hiroshi Yamashita (Japan)
Katsunari / Shinichi Sei (Japan)

DeepZenGo-centric live stream at http://live.nicovideo.jp/gate/lv308903880

Tournament format
This competition consists of preliminary matches and final tournament.
Preliminary matches are held on the first day joined by all programs to choose 16 teams that proceed to the final tournament held on the second day. An anomalistic Swiss style will be used for the preliminary-game matching up to the sixth game, and the Swiss style for the seventh game (final match).
If the number of programs participating to the preliminary games is odd, GNU_Go will be added.
To decide the ranking of the preliminary games, after draws being converted to 0.5 wins and 0.5 losses, the following conditions are applied in the order of listing.
Competitors with the higher number of wins are ranked higher.
Competitors with the higher score of Solkoff (the total number of wins of all the opponents) are ranked higher.
Competitors with the higher score of SB (the total number of wins of opponents they have defeated) are ranked higher.
As for the competitors in the same rank based on (1) to (3), those with higher scores of DB (the number of wins minus the number of losses) are ranked higher.
Decided by lot.
The final tournament is conducted as shown in the tournament chart in accordance with the order of the preliminary-match ranking, and will decide the first to sixteenth competitors.
Image

Schedule (in Japan Time Zone)
December 9th 2017
09:00Preliminary league to choose 16 teams going up to the tournament for the second day.
09:10Start reception
09:20Participants' meeting time
09:30Opening ceremony
09:501st round
10:502nd round
11:50Lunch break
13:003rd round
14:004th round
15:005th round
16:006th round
17:007th round
18:00Accolade

December 10th 2017
09:30Final tournament for 16 teams.
09:40Start reception
10:00The first matches
11:00The second matches
12:00Lunch break
13:30The semmifinal matches
14:30The final match
16:00Closing ceremony

More information (in English language) at http://www.igoshogi.net/ai_ryusei/01/en/
Author:  Uberdude [ Fri Dec 08, 2017 3:22 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
So for which bot was 'define' a test account? :)
Author:  pookpooi [ Fri Dec 08, 2017 3:31 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
Uberdude wrote:
So for which bot was 'define' a test account? :)

My instinct says it's the latest FineArt version that beat 99-1 matches in Fox. DeepZenGo might send either 15.5 or 15.7 from its testing behavior in CGOS.
Author:  Uberdude [ Fri Dec 08, 2017 4:07 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
pookpooi wrote:
Uberdude wrote:
So for which bot was 'define' a test account? :)

My instinct says it's the latest FineArt version that beat 99-1 matches in Fox. DeepZenGo might send either 15.5 or 15.7 from its testing behavior in CGOS.

Do you have any of those game records (other than Ke Jie win one)? Define has quite a distinctive style, e.g. early 3-3s, come into mini-Chinese immediately, always take corner after attach and hane against 4-4 approach and knight move. Did FineArt on Fox do this?
Edit: found it myself in zip here viewtopic.php?p=225397#p225397
Edit 2: quickly looked at a few games of that FineArt, doesn't seem like define.
Author:  Uberdude [ Sat Dec 09, 2017 12:30 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
Apparently FineArt lost a game to weak bot Maru because it kept filling in its territory at the end despite being Japanese rules. I do find it funny how it can play moves that spank top professionals but cannot pass as well as a 25 kyu. A nice example of so-called artificial intelligences doing absurdly dumb things when outside their expected parameters.
Author:  Uberdude [ Sat Dec 09, 2017 1:07 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
Prelim results:
FineArt beat DolBaram
FA beat Abacus
Abacus beat Rayn (funny damezumari ko threat in double ko)
FA beat Rayn
FA beat Tianrang
FA beat Zen with big kill
Author:  pookpooi [ Sat Dec 09, 2017 1:35 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
Uberdude wrote:
FA vs Zen live

And Art won easily with just 165 moves
Author:  pookpooi [ Sat Dec 09, 2017 2:11 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
All game records on first day here

http://52.198.104.180/firstday.html

Look like DeepZenGo and FineArt will separate into different bracket, which is the good thing. We won't want them to fight earlier than Final.

Update: full result in table format of the first day here

http://www.igoshogi.net/ai_ryusei/01/result.html
Author:  johnsmith [ Sat Dec 09, 2017 4:11 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
Good play by FineArt. At move 74, Zen had evaluated position at 56% in its favor, while Leela says it's even. But after some moves in the upper right corner, FineArt gained influence and skilfully attacked Zen's big group.



Attachments:
DeepZenGo-FineArt-201712090832.sgf [1.53 KiB]
Downloaded 1341 times
Author:  Uberdude [ Sat Dec 09, 2017 4:24 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
I preferred black after the trade top side: FA's result is so clean (and top left corner to side became solid, I feel like white should have some skillful play to probe there are make some aji before black got so solid) whilst white's corner is full of bad aji. I wouldn't have found FA's sacrifice to get sente influence to support the attack, but that's exactly the kind of thing I thought black deserved out of the position.
Author:  johnsmith [ Sat Dec 09, 2017 4:33 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
Uberdude wrote:
I preferred black after the trade top side: FA's result is so clean (and top left corner to side became solid, I feel like white should have some skillful play to probe there are make some aji before black got so solid) whilst white's corner is full of bad aji. I wouldn't have found FA's sacrifice to get sente influence to support the attack, but that's exactly the kind of thing I thought black deserved out of the position.
I like how B kept the territory under his large star shimari. It's ususally full of holes, but B made it solid. Perhaps that's where B got the lead, because those stones were so effective. And O14 looks so ugly having in mind weak W group on the left side.
Author:  johnsmith [ Sat Dec 09, 2017 4:41 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
I let Zen analyze a bit longer and it found this move that would make it alive. Is it really working? Branch at move 100.



Attachments:
DeepZenGo-FineArt-201712090832 2.sgf [1.71 KiB]
Downloaded 1341 times
Author:  Bohdan [ Sat Dec 09, 2017 3:39 pm ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
johnsmith wrote:
Uberdude wrote:
I preferred black after the trade top side: FA's result is so clean (and top left corner to side became solid, I feel like white should have some skillful play to probe there are make some aji before black got so solid) whilst white's corner is full of bad aji. I wouldn't have found FA's sacrifice to get sente influence to support the attack, but that's exactly the kind of thing I thought black deserved out of the position.
I like how B kept the territory under his large star shimari. It's ususally full of holes, but B made it solid. Perhaps that's where B got the lead, because those stones were so effective. And O14 looks so ugly having in mind weak W group on the left side.


As for me it was a clear gain for white. Just look at the beginning of the battle. White invaded and just killed black stones. Zen just lacking of computing power compared to FA and thats why it couldn't live. It definitely shouldn't be killed.
Author:  johnsmith [ Sat Dec 09, 2017 4:32 pm ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
Bohdan wrote:
johnsmith wrote:
Uberdude wrote:
I preferred black after the trade top side: FA's result is so clean (and top left corner to side became solid, I feel like white should have some skillful play to probe there are make some aji before black got so solid) whilst white's corner is full of bad aji. I wouldn't have found FA's sacrifice to get sente influence to support the attack, but that's exactly the kind of thing I thought black deserved out of the position.
I like how B kept the territory under his large star shimari. It's ususally full of holes, but B made it solid. Perhaps that's where B got the lead, because those stones were so effective. And O14 looks so ugly having in mind weak W group on the left side.


As for me it was a clear gain for white. Just look at the beginning of the battle. White invaded and just killed black stones. Zen just lacking of computing power compared to FA and thats why it couldn't live. It definitely shouldn't be killed.
Do you know what are time settings and hardware? Though Deep Zen team should be awere that competition is bringing it all in.
Author:  Bohdan [ Sat Dec 09, 2017 6:01 pm ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
I guess there are no restriction on hardware. At least it never was in past. And I doubt that DeepZen team can compete with Tencent company on this. FineArt never published the hardware they are using.
Author:  pookpooi [ Sat Dec 09, 2017 7:03 pm ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
johnsmith wrote:
Do you know what are time settings

http://www.igoshogi.net/ai_ryusei/01/en/rules.html
Author:  pookpooi [ Sat Dec 09, 2017 10:27 pm ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
As expected (no pun intended), DeepZen meet FineArt in the final. This is basically Alien vs. Predator scenario because they demolish other programs with ease. Good thing FineArt blundered yesterday (Tianrang also) so the programmer can fix, today there's no room for any error anymore. FineArt is the favorite to win but can DeepZen pull an upset just like it won at World AI Go Championship in China?

Update: FineArt has very high chance to win, according to DeepZenGo. But no matter who wins, we lose
Image
Author:  pookpooi [ Sun Dec 10, 2017 12:25 am ]
Post subject:  Re: AI RYUSEI Tournament 2017, December 9-10
All second game records are here http://52.198.104.180/secondday.html
Author:  EdLee [ Sun Dec 10, 2017 2:01 am ]
Post subject: 
Does anyone happen to know how the developers in Japan (e.g. DeepZen) and outside of Japan (e.g. AlphaZero, FineArt, etc.) feel about implementing area scoring (e.g. AGA's) versus territory scoring ?

( How does Leela do scoring? )
Author:  Waylon [ Sun Dec 10, 2017 8:54 am ]
Post subject:  Re:
EdLee wrote:
Does anyone happen to know how the developers in Japan (e.g. DeepZen) and outside of Japan (e.g. AlphaZero, FineArt, etc.) feel about implementing area scoring (e.g. AGA's) versus territory scoring ?

( How does Leela do scoring? )

It seems that Leela uses area scoring.

I used SmartGo to create the four attached game files. One uses Japanese rules and komi 6.5, the other three are with 7.5 komi and Chinese, AGA territory and AGA area rules.

SmartGo's scoring function gives the expected results:
Jap = B+0.5, Chi = B+1.5, AGAter = B+1.5, AGAarea = B+1.5

Leela scores differently only with Japanese rules:
Jap = B+2.5, Chi = B+1.5, AGAter = B+1.5, AGAarea = B+1.5

Zenith Go 7 seems to use always territory scoring, regardless of the RU parameter in the sgf file:
Jap = B+0.5, Chi = W+0.5, AGAter = W+0.5, AGAarea = W+0.5

Crazy Stone Deep Learning gives the correct score for Japanese and Chinese rules, but seems to be wrong with AGA rules:
Jap = B+0.5, Chi = B+1.5, AGAter = W+0.5, AGAarea = W+0.5

(It seems I can only attach three files. You can edit the RU parameter in the sgf file with a text editor)

Attachments:
ScoreTestAGAter.sgf [242 Bytes]
Downloaded 623 times
ScoreTestChin.sgf [228 Bytes]
Downloaded 633 times
ScoreTestJap.sgf [231 Bytes]
Downloaded 613 times
Page 1 of 2 All times are UTC - 8 hours [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/