Congratulations a1k0n!

[Fritzlein]

What a dominating performance. Victory by 90 Elo points! The next 90 points spanned all the way from second to fourteenth. I am in awe.

Congratulations, Andy, and thanks again for explaining in the forum how you did it. I'm sure you're going to get a lot more questions about your secret sauce now that the results are in.

Yahoooooooo!

[Savaron]

Congratulations! What are you doing at Yahoo exactly?

[afri]

Congrats to the pwnz0r!

[analyst74]

Very impressive victory! And who should I send my resume to if I want to work with you? :p

[hebbie]

Indeed. Well done a1k0n.

If we cross paths again, I hope I can get revenge!

Hebbie

[Janzert]

Yes, Congratulations a1kon. Quite a nice gap between you and the rest of the field.

Janzert

[a1k0n]

Thanks everyone. I'm totally blown away, as I didn't think my approach was all that much better.

I'll write a blog entry with all the deets but I'm behind enough on work as it is... My code is of course still . The commit logs might be somewhat entertaining. The contest version is tagged a1k0nbot-2.18.2. (I lost count of version numbers so I have some -3.x tags... ignore them). I have a VPS host running my contest version bot on if you want to play against it.

I work for Yahoo! Mail in the service engineering organization currently, writing deployment tools for coordinating software upgrades to tens of thousands of machines.

[DjinnKahn]

Congrats a1k0n! : )

[aramadia]

Haha, didn't even think I'll come close to the top 20 then it turns out I was beaten only by you. Good job a1k0n.

I was going for simplicity as opposed to expert game-specific strategy (I really just ported a chess engine I had wrote to Tron/C++). I didn't even bother checking if I was walled off from an opponent and then solving the longest path problem. Instead the only interesting "feature" I had was storing a transposition table and using it to order moves. My evaluation function could evaluate 200k Nodes/s (using iouri_s tree of chambers) which probably gave me the advantage over other implementations running 20kN/s. Search averaged 20 ply in 15x15.

Overall, a great contest though.

[a1k0n]

Holy cow. Excellent job aramadia. On the whole I found that deeper searches with imperfect board evaluations were more "self-deluded" about what direction the game was going to go in the future, so I didn't worry too much about search depth and tried instead to keep my evaluation as useful to lower search depths as possible. But a transposition table would have been a really good idea, and not too hard to implement with Zobrist hashing. It would be really interesting to see what a difference that makes though.