The open source your ant bot topic

[Nionita]

Hi,

If someone cares to see a Haskell entry, here is mine:

git@github.com:nionita/MyAnts.git

It is good, but not the best (position 105 or so in the end unofficial top). Not very well commented, but anyway...

Nicu

[pezzotto]

Here is mine:

https://github.com/lorenzoriano/AI-Challenge

I had written a longer post describing my strategy, but it never got published, and I can't re-write it right now. I'll blog post it one day

[meduza]

Hi, this is mine

https://github.com/nypi/ants

I really hope to see it at the first page of the final rankings
Thank you all for the great contest!

[lazarant]

Here is mine. Was ranked 2nd in the end.

[GreenTea]

Hi,
This is my code and jar. Enjoy!!
http://sourceforge.net/projects/ants2011/

[BenJackson]

Profile: http://aichallenge.org/profile.php?user=3473
Source: https://github.com/bjj/ants2011

All the submission revisions are tagged. The final one is definitely top 20, maybe top 10.

Sorry there aren't many comments. I'm happy to answer questions on the forum or in email. I will try to illustrate my combat at some point because it's going to be hard to understand it as is. It would help to look at the first version (which was less complex but not as sophisticated) before I bandaged it awkwardly to do "focus" battle resolution properly.

[utoxin]

Here's mine. Not necessarily proud of the code, but it's my first serious attempt at writing Python, and definitely my first AI of any kind, so I'm quite pleased with how it turned out.

https://github.com/utoxin/AIChallenge-Ants

I'm also happy to answer questions, or get suggestions on my approach. I went with what I call a 'scent map' for the primary driving force behind my AI. Combat was tricky with that approach, but I was getting pretty close to a solid approach at the end.

[RVeerdonk]

Doing my part,

This is my final submission (from Nov 10). Life interfered afterwards and many ideas for improvements went unimplemented. This bot made it to third spot around Thanksgiving and finished the regular season in 13th spot [hoping for top-30 in the finals].

It's a C++ bot with a module for combat (Mini-Max for up to 8 moving ants, and "all-or-nothing" beyond that based on local power balance). It has a food collection strategy as discussed at IRC in the beta time-frame, updated to treat enemy hills as food. Finally, it has exploration code based on a multi-component heuristic score that I found very difficult to tweak properly and became ever more complex (more ingredients) over time. However, this heuristic function is where a lot of strength/weakness appears to come from. I implemented it as a weighted integer sum of distances to a variety of targets (found by BFS). A real value and a distance function would have been the obvious next steps there.

I have made some smallish "improvements" in all of these modules since and integrated visualizer code, but that never consistently resulted in a version that beat this version in local testing. That probably means I introduced some bugs along the way or tweaking the heuristic function became even more fragile. I would need to work on some RL-type of approach there and a better testing strategy, but I did not have the time to work on it. If someone is interested, I would be willing to share that code as well.

René

http://aichallenge.org/profile.php?user=5

[Hippo]

I don't use java regularly ... I was a bit fighting with it ... here is my final code.

The most interesting is pondering ... (not the DistTo/DistFrom thing ... actually used just to cause crashes, but the symmetry detection and use).

I was 14th with 7th version (with single dimensional symmetries and bugs in prevent 1-0 exchanges and with "naive attacking turn") I was debugging a lot at the end, finishing with more frequent attacking turns, but attacking only with groups of "frontsize" at least 4. I hope this will improve it a lot.
Distance to frontline approach was added at latest versions as well.

You can see there is a long TODO list of ideas ... . The search method using too many (slightly modified) BFS's was not changed to search from all my ants with "unsearch" mainly due to not having time for debugging. Current version has complexity problems. Often a lot of work should be skipped just to prevent timeouts. You can see boolean variable choice ... the prepared switch to the other search method.

I have not implemented "narrow corridors" guarding ... and I don't have ideas how to do so. I hope to outnumber opponents due to fast food discovery before space fighters have time to build their gardens. My bot does not plan to survive long fight. It has to be agressive enough to make big loses to the opponnets trying to space fight it.

One more thing ... when gathering food is not helpful for my bot (as it will not be converted to my ants) the food gathering gains very less priority so my bot becomes very very agressive.

Due to complexity issues I had not start alpha-beta like fights. I don't lose my ants immediately, but combination requiring 2 or more turns will kill them. This is why my ants will be eliminated from the gardens easily. So the bot have small chances to get into the top 5. I hope not to be much under top 30.

[antysocial]

Here is my source code. Before the reset, I was around #26 or so.

http://aichallenge.org/profile.php?user=9469