iDevGames Forums
Google AI Challenge - Printable Version

+- iDevGames Forums (
+-- Forum: Community Zone (/forum-4.html)
+--- Forum: Contests (/forum-19.html)
+--- Thread: Google AI Challenge (/thread-8179.html)

Google AI Challenge - OneSadCookie - Oct 6, 2010 09:16 AM

Who's entering:

There seems to be no way to search by participant's name or anything so I thought it'd be nice to collect iDG members' profile page links here.


RE: Google AI Challenge - GryphonClaw - Oct 6, 2010 09:20 AM

My profile is:

RE: Google AI Challenge - geezusfreeek - Oct 6, 2010 12:37 PM

I don't have a bot up yet, but here's my profile.

RE: Google AI Challenge - sealfin - Oct 6, 2010 01:08 PM

Woah, I hadn't heard of this - thanks for posting this OneSadCookie!

I've entered, but I've not submitted a bot yet; my profile page is here; annoyingly, there doesn't seem to be any way to edit your profile...

@geezusfreeek: your profile... Grin I'm reminded of Resurrection Man...

RE: Google AI Challenge - AnotherJake - Oct 7, 2010 12:44 AM

I tried to sign up three times, but their account system apparently doesn't work, so I it appears I don't get to play Annoyed

[edit] Holy cow! ... like 15 minutes after my very first attempt it finally emails me a confirmation link and it worked when I signed in this time! This is completely retarded coming from a "programming" club. Unbelievable. I got an activation email from my second attempt within seconds, and my first one arrives fifteen minutes later?! TF? I hope they don't decide to disqualify me now for having three accounts. [shakes head]

[edit2] Whoops, I got all torqued off and forgot to post my profile page:

RE: Google AI Challenge - PowerMacX - Nov 1, 2010 07:56 PM


My epic battle against OSC

RE: Google AI Challenge - AnotherJake - Nov 1, 2010 11:16 PM

Well at least you're in it PMX! I really wanted to join in but I got slammed with (paying) projects. I did come up with what I think was a pretty clever strategy on the first round, but that's as far as I got Sad

RE: Google AI Challenge - KittyMac - Nov 9, 2010 10:57 AM

Sweet. I remember going to Waterloo back when I was in college for these programming contests. Ah, the memories...

I'm in!

RE: Google AI Challenge - PowerMacX - Nov 20, 2010 12:06 PM

LOL, I was holding out uploading an update because KittyMac's score was very close to mine, so I hoped to see at least a match before updating. Of course, then I totally forgot about it and now I'm 1400th-ish and KittyMac 700th-ish! Rasp

Well, with the new score KittyMac got to play with OSC! I'll be uploading a new version this weekend to try and catch up.

RE: Google AI Challenge - OneSadCookie - Nov 20, 2010 12:08 PM

Wow, that wasn't much of a game :'(

RE: Google AI Challenge - KittyMac - Nov 20, 2010 12:12 PM

(Nov 20, 2010 12:08 PM)OneSadCookie Wrote:  Wow, that wasn't much of a game :'(

Oh, snap!

RE: Google AI Challenge - PowerMacX - Nov 28, 2010 08:47 AM


I finally managed to upload an updated version, and now my score is between KittyMac's and OSC, but that's only after 17 matches. Since all scores were reset, it will take a while for scores to stabilize... so I thought I would mention that now when I'm not losing Rasp

Still hoping to see another iDG vs iDG battle! Grin

RE: Google AI Challenge - EvolPenguin - Dec 2, 2010 01:21 PM

I found the post-mortem of the winner of the Google AI challenge.


RE: Google AI Challenge - PowerMacX - Dec 2, 2010 06:42 PM

(Dec 2, 2010 01:21 PM)EvolPenguin Wrote:  I found the post-mortem of the winner of the Google AI challenge.


Argh! 90% of what he mentioned was my approach too! Of course, about half of it ended up being TODOs comments in my final version... Rasp

He mentioned that the map change helped some bots and hurt others. I can definitely say it helped mine, because having more planets at the same distance from both players at the start is something my bot was rather good at. It uses a few "tricks":
1. Using a futureState(planet, turns), where turns == distance from planet to my starting planet-1, it finds all the neutral planets that are going to be taken over by the enemy (and gets how many enemy ships would be on it)
2. It sends an X amount of ships to those planets, with X equal or higher to: expected enemy ships at destination + destination growth rate (because I arrive a turn later) + 1
3. It doesn't send any ships on the first turn Sneaky (KittyMac's bot apparently does the same!)

That way, an enemy bot may send 61 ships to take over a 60 neutral with growth rate 5, and a turn later I send just 7 ships and take it over a few turns later, right after his ships arrive Cool

That part of the bot behavior is very clear in this game:

Another thing my bot does is figure out which of my planets are going to be taken over at or before turn X, where X is the distance from said planet to my "reinforcement" planet. Unlike with neutrals, in this case I don't want to purposely arrive "late", since if I arrive Z turns late I'd lose Z*planet growth rate*2 ships!

Nonetheless, that is the only real reinforcement aspect it has, since I didn't get to implement all my TODOs, the biggest of which was ranking planets (both mine and the enemy's) by weakness against a full-on all-planets attack. That would have made the reinforcement flow a lot more effective, by choosing reinforcement planets more wisely.