Robo Home | Changes | Preferences | AllPages

Curve Flattening Challenge

A part of the MovementChallenge. This is the setup: Additionally:
  1. Challenger is not allowed to fire any bullets at all
  2. The ratio "Challenger score / reference bot score" * 100 is the Challengers CFC Index

Also see

/Index - /Chat

SandboxDT 1.91 is now available and can operate as the reference bot (set the behavior value in the perperty file in DT's data directory). When acting as a reference bot it does not read any data files containing robot stats.

I suggest running for 1000 rounds for stability purposes and to ensure DT can not find a flaw in the movement (it learns fast - but not always) -- Paul Evans

I suggest running 500 rounds for balance with the AntiPatternMatcherChallenge and because 1000 rounds takes epochs on my Java. -- PEZ

It seems to make matter a lot how many rounds we choose. CassiusClay gets 74% over 1000 rounds and 83% over 500 rounds. Part of this might be chance but it seems unlikely that it should be the only factor with such a big difference. Rather it is probably that DT learns more about CC's movement over time than CC learns about DT's targeting. Anyway, for the purposes of comparability we should settle for either 500 or 1000 rounds. I still suggest 500 because it takes its time to run 1000 rounds and also because 500 rounds is used in the TargetingChallenge. Besides, a gun should be able to learn fast enough for 500 rounds to be enough. =) -- PEZ

Robo Home | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited October 3, 2004 20:38 EST by PEZ (diff)