|
would be cool to set up some live games like the nova league, but for goons. someone else do it though im just an ideas man
|
#
?
Jan 24, 2016 18:18
|
|
|
|
| # ? May 25, 2024 23:51 |
|
|
Hey getting my poo poo together again, and so getting back into Go. Used to play on KGS with goons, everybody moved to some website based thing now, right? E: oh hey, it's just online-go.com I guess, I'm smart. surc fucked around with this message at 18:26 on Jan 24, 2016 |
|
#
?
Jan 24, 2016 18:18
|
|
|
 After all these years... I finally got to a solid 1d rank on kgs! Now to see if I can actually hold it  Also since the top 3 players in the itgo cup seem to be around the same rank, I dropped out. Hopefully you guys can have a good competition!
|
|
#
?
Jan 24, 2016 18:49
|
|
|
it finally happened https://googleblog.blogspot.com/2016/01/alphago-machine-learning-game-go.html quote:After all that training it was time to put AlphaGo to the test. First, we held a tournament between AlphaGo and the other top programs at the forefront of computer Go. AlphaGo won all but one of its 500 games against these programs. So the next step was to invite the reigning three-time European Go champion Fan Hui—an elite professional player who has devoted his life to Go since the age of 12—to our London office for a challenge match. In a closed-doors match last October, AlphaGo won by 5 games to 0. It was the first time a computer program has ever beaten a professional Go player.
|
|
#
?
Jan 27, 2016 19:28
|
|
|
games: http://www.usgo.org/news/2016/01/alphago-beats-pro-5-0-in-major-ai-advance/ Not surprised at all tbh. Previous go research was mostly done by some random dudes in their own basement. Once someone who knows what they are doing decided to throw down some money it was bound to happen.
|
|
#
?
Jan 27, 2016 19:29
|
|
|
Cool that they used machine learning rather than mc
|
|
#
?
Jan 27, 2016 20:10
|
|
|
I've dabbled with Go for a long time, but I've never had anyone to play with, so I've really ever Gotten It. I'm fine with losing, but I'm getting tired of not having feedback so I can understand why I'm losing. Is the OP current and KGS is where things still happen? Or have people moved elsewhere?
|
|
#
?
Jan 27, 2016 21:35
|
|
|
online-go.com is the most friendly place for beginners, i've found. and it doesnt need a client to be installed either. there is an SA group there someone can invite you to
|
|
#
?
Jan 27, 2016 22:17
|
|
|
pointsofdata posted:Cool that they used machine learning rather than mc They used machine learning to guide MC. MCST is still at the core of the engine. Details of the algorithm published today in Nature, and the article can be found here: http://deepmind.com/alpha-go.html. It's really nice work, and the games are really good. Importantly: "AlphaGo is scheduled to play Lee Sedol 9-dan, the Korean pro widely recognized as the strongest player of the last decade, in a 5-game series in March."
|
|
#
?
Jan 27, 2016 22:22
|
|
|
Borachon posted:They used machine learning to guide MC. MCST is still at the core of the engine. Details of the algorithm published today in Nature, and the article can be found here: http://deepmind.com/alpha-go.html. Without using mc they still beat all the current programs
|
|
#
?
Jan 27, 2016 22:57
|
|
|
totino boy posted:online-go.com is the most friendly place for beginners, i've found. and it doesnt need a client to be installed either. Awesome. I just registered there as mikeycp. If someone could invite me that would be much appreciated.
|
|
#
?
Jan 27, 2016 22:58
|
|
|
Fan Hui won two of the five 'informal' games with shorter time rules. It would be interesting to see those as well, but unfortunately they didn't release them.
|
|
#
?
Jan 27, 2016 23:44
|
|
|
Is the ITGO room the default goon room? If so can someone invite HamOfTruth?
|
|
#
?
Jan 28, 2016 03:21
|
|
|
pointsofdata posted:Without using mc they still beat all the current programs Only if they give current programs a 5s/move time limit. MC programs are highly sensitive to how much computational power they get. I would be surprised if they did nearly so well against current programs at reasonable time limites.
|
|
#
?
Jan 28, 2016 03:25
|
|
|
I think the point of that part of the paper (?) is to show that what they've added initializes the program better for each turn than MC does. Not that what they've done is superior to MCST in all aspects.
|
|
#
?
Jan 28, 2016 03:53
|
|
|
Overwined posted:Is the ITGO room the default goon room? If so can someone invite HamOfTruth?
|
|
#
?
Jan 28, 2016 08:20
|
|
|
Xom posted:Done. Thanks. I used to play a lot of Go in college. I started on IGS way back when it was only ASCII with no persistent board. You punched in your move and it spit back the board at you. I took the game back up about 5 years ago but only really established nothing. Back in the day I got to the 10-12k ranking range, but I'm not sure how that translates now. I've played a couple games against the easier bots and trounced them by 200ish points each time.
|
|
#
?
Jan 28, 2016 15:32
|
|
|
Uncle Jam posted:I think the point of that part of the paper (?) is to show that what they've added initializes the program better for each turn than MC does. Not that what they've done is superior to MCST in all aspects. The paper does *lots* of interesting things. From a quick glance, the main things are that it uses a DNN to weight/initialized MCST searches, it truncates rollouts and uses a different DNN to evaluate those abbreviated rollouts, and it started training these DNNs against a large database of 6d-9d KGS games, and then against its own games later. I expect they have been and will be training against Lee Sedol games. THey also use a lot of horsepower - 1200 cores and 170 GPS - for this.
|
|
#
?
Jan 28, 2016 15:45
|
|
|
This is really cool and none of my friends really care so I'm a bit sad.
|
|
#
?
Jan 28, 2016 20:52
|
|
|
Borachon posted:The paper does *lots* of interesting things. From a quick glance, the main things are that it uses a DNN to weight/initialized MCST searches, it truncates rollouts and uses a different DNN to evaluate those abbreviated rollouts, and it started training these DNNs against a large database of 6d-9d KGS games, and then against its own games later. I expect they have been and will be training against Lee Sedol games. THey also use a lot of horsepower - 1200 cores and 170 GPS - for this. Yeah I hadn't been following this but I looked at the games it played vs Fan Hui and they are bonkers. It clowns him so bad, honte as gently caress I feel like you only got a shadow of its full force vs Fan Hui because it had him so under control. I want to see games against the boss man.
|
|
#
?
Jan 29, 2016 00:38
|
|
|
Borachon posted:The paper does *lots* of interesting things. From a quick glance, the main things are that it uses a DNN to weight/initialized MCST searches, it truncates rollouts and uses a different DNN to evaluate those abbreviated rollouts, and it started training these DNNs against a large database of 6d-9d KGS games, and then against its own games later. I expect they have been and will be training against Lee Sedol games. THey also use a lot of horsepower - 1200 cores and 170 GPS - for this. I think the only part I don't get is training against 6d kgs games. There are tons of kifu for pro games so why not use those? Well, maybe with deep learning it doesn't matter if it trains enough. I dunno.
|
|
#
?
Jan 29, 2016 05:56
|
|
|
Uncle Jam posted:I think the only part I don't get is training against 6d kgs games. There are tons of kifu for pro games so why not use those? Well, maybe with deep learning it doesn't matter if it trains enough. I dunno. Was probably just a large dataset in a convenient format.
|
|
#
?
Jan 29, 2016 14:36
|
|
|
On the AGA YouTube channel there is a video of Kim reviewing the alphago games, it's good but a lot of the video is Kim demanding to know how the program counts ko threats or other things.pointsofdata posted:Was probably just a large dataset in a convenient format. I scraped some Japanese geocity type sites that had like the sgf for all preliminary games of every tournament for a pro search database before, but maybe there is some usage rights they have to be careful about.
|
|
#
?
Feb 1, 2016 17:19
|
|
|
Uncle Jam posted:On the AGA YouTube channel there is a video of Kim reviewing the alphago games, it's good but a lot of the video is Kim demanding to know how the program counts ko threats or other things. When he was commenting on the games and what AlphaGo did well and poorly or well, it was really interesting. When he started to speculate in ways that made assumptions about how the underlying program worked, though, things were much less useful. The whole discussion on the value of sente in a particular position was really interesting and why AlphaGo's play in the first game was problematic was really interesting.
|
|
#
?
Feb 1, 2016 19:11
|
|
|
While I think understanding sente value is important for humans playing the game, and I enjoyed that discussion, I think this sente value as a whole is just a specialized case of whole board branch searching. Sente value is just a thing that gets inbuilt naturally to the billions of play out result of the tree search. Hopefully it'll be one of the things analysis programs can show us in the future though.
|
|
#
?
Feb 1, 2016 21:17
|
|
|
Can someone please invite adius to the ITGO room?
|
|
#
?
Feb 3, 2016 05:08
|
|
|
Done.
|
|
#
?
Feb 3, 2016 06:31
|
|
|
Goodness, it's been quite a while. Started playing Go and joined OGS back in 2010, lost interest for a while until I remembered it in 2012, played for a bit and then forgot about it again until finding this thread recently. Made a new OGS account as, while I remember the rules, I've forgotten most of the strategy involved. Maybe I can stick with it this time and manage to improve at the game? Let's see! It'd help if there was a group in my area to play in person, but alas. Username's ZackAter, would someone invite me to the SA group?
|
|
#
?
Feb 9, 2016 05:55
|
|
|
Invitation sent, welcome friend
|
|
#
?
Feb 9, 2016 11:18
|
|
|
I learned how to play about ten years ago but then promptly never really played. I recently found there's a local go club and have started going, but they only meet once or twice a week. I made an account on OGS and KGS, anybody willing to give me pointers and tell me how bad my shape is? Also, could I get an invite to the SA group? User name's totopi
|
|
#
?
Feb 13, 2016 15:30
|
|
|
I added you to ITGO on KGS.
|
|
#
?
Feb 13, 2016 17:52
|
|
|
tonberi posted:I learned how to play about ten years ago but then promptly never really played. I recently found there's a local go club and have started going, but they only meet once or twice a week. I made an account on OGS and KGS, anybody willing to give me pointers and tell me how bad my shape is? Also, could I get an invite to the SA group? tonberrytoby
|
|
#
?
Feb 14, 2016 18:02
|
|
|
Can I get an invite to the OGS group? My username on there is kirtar. Time to see if I can find my stones and books. For some reason I only know where my I put my board.
kirtar fucked around with this message at 21:08 on Feb 15, 2016 |
|
#
?
Feb 15, 2016 20:39
|
|
|
its okay, we all lose stones when we dont play for a while
|
|
#
?
Feb 15, 2016 22:07
|
|
|
totino boy posted:its okay, we all lose stones when we dont play for a while
|
|
#
?
Feb 15, 2016 22:14
|
|
|
toodles posted:I've lost my marbles
|
|
#
?
Feb 15, 2016 23:18
|
|
|
Can I also get added to ITGO on KGS? Username on there is kirtar. I've just been using OGS for now since I can do correspondence games on the same server if needed. I did a few 9x9 games to get out of provisional. Two of them were pretty awkward since I killed all of my opponents groups.
|
|
#
?
Feb 16, 2016 03:23
|
|
|
totino boy posted:its okay, we all lose stones when we dont play for a while I'm still 3k and don't tell me otherwise.
|
|
#
?
Feb 16, 2016 05:16
|
|
|
i am 100d no pro has ever beaten me
|
|
#
?
Feb 16, 2016 07:15
|
|
|
|
| # ? May 25, 2024 23:51 |
|
|
"misterfish" would like to play in the itgo room. I'm terrible and managed to lose ranking to bots... I'm at 27k now. I figure there is nowhere to go but up.
|
|
#
?
Feb 16, 2016 08:19
|
|
