[Home]PatternMatching/PatternRating

Robo Home | PatternMatching | Changes | Preferences | AllPages

For people who don't look for best matches rather than exact matches (like the Symbolic variety) in the pattern-matchers, how do you rate the matches? If I remember right, MogBot takes the absolute value of the difference in acceleration and heading change for each turn, but it seems like that might be misbalanced. How do you other top pattern-matchers rate the strength of a pattern? -- Kawigi

Usually I use the absolute difference divided by the range of the variable (a kind of normalized distance). When I consider more than one variable, I just add the resulting distances. -- Albert

So if you're using heading-change, would you also factor in the differences in possible heading change (like divide by the maximum heading-change for their current velocity), or would you divide by the absolute maximum heading change (i.e. when they aren't moving)? -- Kawigi

MicroAspid is dividing by the maximum heading change, but MicroAspidPMC is dividing by the maximum heading change at the enemy's velocity, and seems to work better (but, I have not tested against real bots, only against PatternBot). -- Albert

Interesting... looking back at Teancum (PMC edition?), I may be able to get away with not 'normalizing', since both my paramters (LateralVelocity and AdvancingVelocity) have a range of -8 to 8. However, it may even be more relevant to find the difference of sqrt(latv2+advv2) and atan2(latv, adv) and normalize them. -- Kawigi


Robo Home | PatternMatching | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited June 17, 2003 21:53 EST by Kawigi (diff)
Search: