DeveloperCreated Wednesday, March 16th 2016 @ 17:34:01
Hi dear coders,
Today we release our new competition: Go!
This game is a very old Asian board game, but very popular in AI research because it is so difficult for computers to play (well), as there are countless game states. Recently popularized by Google's bot AlphaGo, who challenged the world's best human players and managed to beat them most in most games. As it so happened we were just working on releasing this game, so here it is!
Good luck with this competition (you will need it). The competition is still in Beta, so there might be a few bugs here and there. If reported they will be fixed asap.
As always, have fun coding! -Jim
- Created Wednesday, March 16th 2016 @ 18:21:32
This is great news, Jim! Thank you for adding this.
- Created Wednesday, March 16th 2016 @ 23:16:23
- Created Thursday, March 17th 2016 @ 00:29:13
Very cool! Curious to see what everyone comes up with :)
- Created Thursday, March 17th 2016 @ 15:38:56
awesome! thanks a lot for adding this game, I'm sure it will bring a very competitive and nice environment!
- Created Thursday, March 17th 2016 @ 16:59:16
Hey, do you really think it is realistic that this competition is won by skill? Its more likely the most lucky bot (with a decent amount of skill) wins this.
I think reducing the field size could result in a more fair and interesting competition.
- Updated Thursday, March 17th 2016 @ 20:39:44
+1 for reducing the size.
9x9 would still be a worthy challenge.
- Created Thursday, March 17th 2016 @ 21:20:57
-1 for reducing the size... With 9x9 you can start using search techniques like mcts, all pretty well documented. With the limited time available that won't work on 19x19, so you need to be more creative yourself, which I think makes the competition more interesting. The game level will be much lower, but the best programmer will win.
Hey, alphago can do it, so surely you can do it too...
My only minor concern is that each game will take a lot of moves, potentially costing a lot of game server time.
- Created Thursday, March 17th 2016 @ 21:39:14
Rly mcts with over 8 * 10^94 possibilities?
- Created Thursday, March 17th 2016 @ 22:00:29
AlphaGo has way more than 500 ms per move, file I/O, lots of memory, which we don't. 15x15 boards still have a huge number of board states (10^107, more than there are atoms in the universe ; in comparision, Ultimate Tic Tac Toe only has 10^38), but a lot less than 19x19 boards, and that would be much more manageable in 500ms.
- Created Friday, March 18th 2016 @ 06:08:26
But alphago also has a neural network that without any lookahead is just as strong as top MCTS programs. The last competition's (block battle, four in a row and ultimate tic tac toe) are very much brute force search oriented. I think it is refreshing to see a competition that is not about searching. Let's give this format a chance.
- Created Friday, March 18th 2016 @ 08:19:05
I do agree, it is a good thing we'll have to find other methods. But I keep thinking 19x19 is just monumental given the other requirements, and we'd also have to go for something else than search with 15x15 boards.
- Created Friday, March 18th 2016 @ 11:48:00
19x19 (even 15x15) is unrealistic you'll see as the elo ranking system won't work properly if too much luck (or lucky moves) are involved. I bet you, elo of the best bot wont exceed 1900 by much.
- Created Friday, March 18th 2016 @ 12:25:29
Wow-wow-wow! Great! Is there a good referee in games? It must be strong enough ;)
- Created Friday, March 18th 2016 @ 16:05:42
And curious on how the limitations will affect game behaviour :-)