It is currently Wed May 07, 2025 9:30 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 24 posts ]  Go to page Previous  1, 2
Author Message
Offline
 Post subject: Re: Google AI Challenge
Post #21 Posted: Fri Nov 26, 2010 5:33 am 
Lives with ko
User avatar

Posts: 295
Location: Linz, Austria
Liked others: 21
Was liked: 44
Rank: EGF 4 kyu
GD Posts: 627
Li Kao wrote:
Did somebody write a bot which uses a variant of MCTS? It should be possible to adapt it to this game pretty well.


I have tried, but not very successfully.

The first problem is the huge game tree complexity. What I did here was adapt the general idea of UCT to include dynamically expanding tree width. That is, when initially creating a node, I populate it with a few heuristically generated moves. Then I dynamically add random moves when the node is visited often. The obvious flaw is that the node is only visited often when the heuristic moves are good, so an extremely good followup that's completely different than the heuristics is extremely unlikely to be found. But on the other hand, visiting each possible move at least once just isn't practical when there are millions of possible moves at each point ;)

The second problem is performance. My implementation just isn't able to search more than about 100k nodes a second, and that just doesn't seem enough to generate good moves.

The end result is that my MCTS/UCT bot is not really much better than my currently submitted "stupid" bot (which is currently ranked 485 after all ;) ).

Top
 Profile  
 
Offline
 Post subject: Re: Google AI Challenge
Post #22 Posted: Thu Dec 02, 2010 1:07 pm 
Gosei
User avatar

Posts: 2116
Location: Silicon Valley
Liked others: 152
Was liked: 330
Rank: 2d AGA
GD Posts: 1193
KGS: lavalamp
Tygem: imapenguin
IGS: lavalamp
OGS: daniel_the_smith
At least one person (forget who) in the top 50 was doing UCT/MCTS. I think it's pretty clear you can't use random playouts in this game, the move generator needs some intelligence. I did UCT to find the optimal order in which to achieve my goals.

My program was too complicated for the amount of time I had to write it, too many bugs and I didn't get even a small amount of what I wanted finished. Oh well, hopefully next time they'll pick something that doesn't interest me so I don't lose every second of my life for another three months...

Still, I guess #262 isn't a horrible finish.

_________________
That which can be destroyed by the truth should be.
--
My (sadly neglected, but not forgotten) project: http://dailyjoseki.com

Top
 Profile  
 
Offline
 Post subject: Re: Google AI Challenge
Post #23 Posted: Thu Dec 02, 2010 5:41 pm 
Lives in sente
User avatar

Posts: 921
Liked others: 401
Was liked: 164
Rank: German 2 dan
Here is the winner: http://www.zdnet.com/blog/burnette/hung ... ntest/2131

Here is his blog: http://quotenil.com/Planet-Wars-Post-Mortem.html

The source code is available and very nicely commented.

_________________
A good system naturally covers all corner cases without further effort.

Top
 Profile  
 
Offline
 Post subject: Re: Google AI Challenge
Post #24 Posted: Fri Dec 03, 2010 12:19 pm 
Gosei
User avatar

Posts: 2116
Location: Silicon Valley
Liked others: 152
Was liked: 330
Rank: 2d AGA
GD Posts: 1193
KGS: lavalamp
Tygem: imapenguin
IGS: lavalamp
OGS: daniel_the_smith
At least I beat space.invaders! :)

_________________
That which can be destroyed by the truth should be.
--
My (sadly neglected, but not forgotten) project: http://dailyjoseki.com

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 24 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