…thout them, i found out accidentally.
if you have a..b.a and the left a moves right, then the right a is alive again. this wasn't being handled.
… defense scoring helps a bit
… to be a nice improvement.
The incoming order was biased. This seems to help a lot. also tuned some parameters and documented my thought process some more.
…her than frequency of superiority there's an exploration tiebreaker in here, but it's disabled as it worsens play
(very unlikely to change any actual behavior at this point, but I'm looking to remedy that)
dependent ants were in a random sampled state, not the maximum likelihood state, when picking moves. now it follows the chain of dependencies, and marks ants committed as necessary. also, elided the unused addCombatAnt stuff.
…eneral problem. each ant's move is conditioned on the moves of one closest ant "above" and one closest ant "left", so all formations are sampled together in a chain (and unrelated ants are sampled together but this is usually not an issue). also, fixed a major scoring bug. scoring needs a lot of cleanup.
the "dependent" ant generation is just a stub now and only works because the ants are ordered by coordinate; ideally, we'd be grouping them more intelligently around enemy ants. also, i think move scoring is completely broken. it seems to behave mostly right and is a pretty strong bot, but the numbers it outputs make no sense at all.
seems to work and ought to scale to the whole game.
- need to use <= radius2, not < - allow our ants to stay still on the second iteration through their moves