Google AI Challenge 2011
Posted: Wed Oct 19, 2011 4:06 pm
Looks like a lot of fun!Gian-Carlo Pascutto wrote:http://aichallenge.org/
I'm writing mine in CoffeeScript, if only to learn that language and JavaScript. I was hoping to get something close to Python but with reasonable speed, but that's not really the case (it's quite a lot slower than C/C++ and not as good as Python ).Daniel Shawul wrote:I experimented a bit with a bot derived from java starter package. It seems that the AI is dominated by path finding (Djkistra & A*) , despite my expectation that I would be able to use some multi-player algorithms. I don't think a regular turn-based depth first search is applicable here at all since the map is usually too big with many ants. Also I don't think speed of program matters that much (I maybe wrong). What language and algorithms do you use ?
I'm using C and Collaborative Diffusion. I've tried A* but it is expensive computation-wise for every ant and storing the path for the succeeding turns introduces complexity.Daniel Shawul wrote:I experimented a bit with a bot derived from java starter package. It seems that the AI is dominated by path finding (Djkistra & A*) , despite my expectation that I would be able to use some multi-player algorithms. I don't think a regular turn-based depth first search is applicable here at all since the map is usually too big with many ants. Also I don't think speed of program matters that much (I maybe wrong). What language and algorithms do you use ?
Perhaps Genie is what you're looking for. It's a language that resembles Python and transpiles into C code. It's about the speed of C++.Gian-Carlo Pascutto wrote: I was hoping to get something close to Python but with reasonable speed
the problem is comparable from evaluation viewpoint with international checkers. We programmed a base there. So for example advancing with just 1 stone is not so clever, you want to do it the 'napoleontic' manner and advance with entire base. You can use distance based heuristics there and when your base is dominating a lot of terrain that's a winning advantage.Daniel Shawul wrote:I experimented a bit with a bot derived from java starter package. It seems that the AI is dominated by path finding (Djkistra & A*) , despite my expectation that I would be able to use some multi-player algorithms. I don't think a regular turn-based depth first search is applicable here at all since the map is usually too big with many ants. Also I don't think speed of program matters that much (I maybe wrong). What language and algorithms do you use ?
There is a similar effort at stanford (for general games) http://games.stanford.edu/ that also provide starter packages. The games are provided at run time in declarative programming such as PROLOG. It requires a lot of effort to get up a random mover so they also provide starter packages in java. You can start coding alpha-beta right away. I am a bit hesitant to use the starter packages for one I want to learn declarative languages. The packages does too much for you it is like having Houndini as starter package for chess