[Icfp04-discuss] Ant entries

[Andreas Bogk]

Guillaume Marceau <gmarceau at cs.brown.edu> writes:

> The parameters for all the coin flips in our ants where tuned using
> a distributed genetic algorithm (to generate a pool of successful
> parameters), and tournament elimination (to trim the pool down to one
> best parameter-set).

We did this too, but hadn't have enough time left to actually make use
of it.  I'm running the optimizer on one of the ants we've submitted,
and with tuned flips, we're beating some of the ants that beat us with
the un-optimized ant we've submitted.

In hindsight, to be successful in this competition, you had to:

- write a simulator, fast enough for decent tournamens
- write an ant language
- write a visualizer (debugging without it is impossible)

Then of course you had to write an ant, and start early enough working
on this to have it debugged at the end. The ant needed to be able to:

- find food and bring it back home (some people figured out the trick
  of marking in concentric circles, we did too)
- be able to defend your home (otherwise you'll get robbed blind)
- be able to attack the enemy base in some way

Having GA for optimizing flip settings would have been a big benefit.

Andreas