Post mortem including source code by xathis

[fvannee]

Congratulations man, you deserve it. Looking at your explanation, it all seems so simple. Actually, most of it is very similar to what I did, but you perfected it a lot better. I greatly enjoyed watching your bot in action. Hope to see you next challenge!

[agent_smith]

[deccan]

[Hippo]

Wow, keep it simple:). Thanks you for sharing your code. It is fascinating how much the touch space exploration idea was powerfull. It is that powerfull that you don't need additional help by maintaining hivecount and or remembering food location.

Of course this is not the only thing which let your bot to dominate. There were used good programming practices as reducing the map graph not to test edge existence again and again. What fascinates me is the simplicity of the code.

I would complicate the code by TRYING to speed up the mission walk by chrisB teleporting (hive spawn controll) and by sharing the mission among ants to shift gardeners and complete the mission by the frontline one. The pessimistics maximin fight estimate let your bot dominate by conservative play. It is fast enough to let you a lot of spare time. I would probably try to break stalemates... when no fight move is generated at all (having enough spare time) by chosing one local fight per turn and computing whole fight matrix trying to find Nash like equillibrium for the fight.

Detecting opponent stillness is nice heuristics to break stalemates in a lot of cases.

There was that many areas where to go ... studying opponent's fight habbits, ... but you LET IT SIMPLE and WON.

Congratulations

[Corniel]

Well I didn't win (although I was both best Dutchman and best Tweakers.net participant ) I'm satisfied with my 72th place. I did write a blog about my bot:

[xathis]

Thanks everybody!

Well i'm glad i confused the hell out of you with these two traditional starting lines, i hope they'll be found in the next winner's post mortem.

inutard: haha you were not supposed to see this. I really thought i was going to loose after GreenTea was at the top 2,5h before the end.

Corniel: nice, i'll include a link to your blog post in my post mortem.

[Hippo]

Carefully studying the strategy ... I am confused by the fact you were sure the contest conditions would not change ... hard codded alpha, beta, gamma distances, turntime 500ms fixed ... but your guess worked well

[dimkadimon]

xathis congratulations on the win you really deserve it. Huge thanks for the detailed explanation. I have some questions. How long did it take you to implement version 1? At first I was depressed to see that you have such an awesome bot so early into the contest, but then I was glad to have something to aim for. Do you think version 3 is much stronger than version 1? Could you have won the contest with version 1 alone perhaps? Good luck and I hope to see you in the next contest.

[Parasprites]

There was no chance of winning with version 1, since it was already beaten before finals even started.
Version 2 was a major improvement over version 1, while version 3 was just minor bug fixes.

[xathis]

dimkadimon: As Parasprites said version 2 was far better than version 1. pguillory's latest version was better than my v1 if i remember correctly and he only ended up #12 in the finals. My two bigest changes for version 2 were:
The combat optimizations that allowed me to look at groups of more than 7 ants, this made more agressive behaviour possible.
Defence. v2 had no defence at all
Version 3 had some really small fixes only.

I have not kept track of the time i spent on the aichallenge... I started in the beta phase already when the game was different (there were no hills) and reached the first place in the beta in mid or end August. I had a few different versions in the beta, so my version 1 is not really my first upload, though i only had one upload before that was adapted to deal with hills.