Google AI Challenge 2011

[diep]

Sorry to say so but Dijkstra is a tad old algorithm isn't it?

[diep]

Aha i see in the faq:

Are there any prizes?

"Maybe. We're working on that. There is of course the glory and bragging rights that come with doing well in the best programming contest ever made."

So i guess the winners title is gonna be: "Best AI Amateur".

[MikeGL]

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.

Edsel,

Great performance by your entry TwistedLogic, I noticed its currently ranked 190th out of more than 6,000 participants in aichallenge.

I replayed one of TL games played Nov 14th 9:58pm, against Vitaly.Arbuzov and lavalamp. Too bad lavalamp razed TL hive early in the game, and lavalamp hive crushed by Vitaly who eventually won that game.

I'm just a beginner programmer and don't code much, but I noticed some weakness in TL.ant while observing the game.

a. Food gathering:
at Turn 71/402, Food at row:121 col:83, 2 TL ants returned to collect the food when there are already 2 approaching, I first thought it was just consolidating and regrouping before advancing, but after less than 35 turns they just split again, without encountering any enemy ants.

b. Hive defense:
TL hive was attacked early as the lone TL ant that encountered the lavalamp platoon didn't call for backup. I noticed in some top 10 entries that their ants are:
(i) holding ant position if it is = or slightly more
(ii) quickly retreat and call for backup when outnumbered
(iii) attack/move forward only when the number of ants are almost doubled than the opposing ants.

Good luck in your entry and future TL.ant versions

[Edsel Apostol]

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.

Edsel,

Great performance by your entry TwistedLogic, I noticed its currently ranked 190th out of more than 6,000 participants in aichallenge.

I replayed one of TL games played Nov 14th 9:58pm, against Vitaly.Arbuzov and lavalamp. Too bad lavalamp razed TL hive early in the game, and lavalamp hive crushed by Vitaly who eventually won that game.

I'm just a beginner programmer and don't code much, but I noticed some weakness in TL.ant while observing the game.

a. Food gathering:
at Turn 71/402, Food at row:121 col:83, 2 TL ants returned to collect the food when there are already 2 approaching, I first thought it was just consolidating and regrouping before advancing, but after less than 35 turns they just split again, without encountering any enemy ants.

b. Hive defense:
TL hive was attacked early as the lone TL ant that encountered the lavalamp platoon didn't call for backup. I noticed in some top 10 entries that their ants are:
(i) holding ant position if it is = or slightly more
(ii) quickly retreat and call for backup when outnumbered
(iii) attack/move forward only when the number of ants are almost doubled than the opposing ants.

Good luck in your entry and future TL.ant versions

Hi Mike,

Thanks. It's still too few games for the current version to consider the rankings. The last version has climbed to rank 74 after 28 games and the current version has only 10 games for now. It needs at least around 50 games to stabilize it's ranking.

Some of the top bots there have been developed for months now when the competition is still on the beta stage so they have months ahead in development.

As for food gathering, it's a flow in the technique used. It's not that critical though as the current method is still competitive so will have to find fix for that later.

As for hive defense I'm still improving it. Major issue right now for my bot is the tactical fighting skills. It's decent but not good enough.

Here's some of my latest version's game on one of the tcp server:

http://ants.fluxid.pl/replay.34897
http://ants.fluxid.pl/replay.35478
http://ants.fluxid.pl/replay.35448

Bot name is AntDroid.

[MikeGL]

Thanks for the links.
will check the games of AntDroidv16.

[Gian-Carlo Pascutto]

I was hoping to get something close to Python but with reasonable speed

Perhaps Genie is what you're looking for. It's a language that resembles Python and transpiles into C code.

It depends on what you consider "resemble", I guess. I only see some very superficial syntax similarity. It's not even a dynamic language (which is of course why it transpiles). I'd rather use C++ in that case.

[Ron Murawski]

I was hoping to get something close to Python but with reasonable speed

Perhaps Genie is what you're looking for. It's a language that resembles Python and transpiles into C code.

It depends on what you consider "resemble", I guess. I only see some very superficial syntax similarity. It's not even a dynamic language (which is of course why it transpiles). I'd rather use C++ in that case.

I worked in Genie for a couple of months and I found it quite easy to make the Python-to-Genie transition. The generated C code is quite clear.

Shedskin is a transpiler that is much closer to the Python language. ( http://code.google.com/p/shedskin/ ) But it only supports a subset of Python and seems quite restricted when compared to Genie.

If you are looking for a fully-compatible, dynamic Python language, then you need to give up your requirement for 'reasonable speed'.

The fastest Python is PyPy ( http://pypy.org/ ), but it is not (yet) orders of magnitude faster than CPython.