[Home]History of Lukious/Rewrite

Robo Home | Changes | Preferences | AllPages


Revision 44 . . September 29, 2007 19:10 EST by Voidious [1.19.32 results and notes]
Revision 43 . . (edit) September 29, 2007 0:07 EST by Voidious [50 seasons of 1.19.24 fast learning]
Revision 42 . . (edit) September 28, 2007 22:34 EST by Voidious [had something backwards]
  

Difference (from prior major revision) (author diff)

Changed: 11c11,12
| Lukious 1.19.24 | DC/GF | | | | | | |87.79 | 77.84 | 84.14 | 77.65 | 81.27 | 81.74 | | 3 seasons
| Lukious 1.19.24 | DC/GF | | | | | | | 87.79 | 77.84 | 84.14 | 77.65 | 81.27 | 81.74 | | 3 seasons
| Lukious 1.19.32 | DC/GF | | | | | | | 88.25 | 78.40 | 84.78 | 77.47 | 80.30 | 81.84 | | 8 seasons

Changed: 33c34
* 1.19.24: Dropped 5 attributes: distance last 15 ticks, distance last 25 ticks, time since velocity changed, time since max velocity, and turn rate. Dimension weights used by the gun are a running total of weights that are calculated once per shot, trying to take mutual information into account.
* 1.19.24: Dropped 5 attributes: distance last 15 ticks, distance last 25 ticks, time since velocity changed, time since max velocity, and turn rate. Forgot I'd left this in there - using Cosinus distribution instead of Gaussian for the KernelDensity. (Not sure this matters much at all...) Dimension weights used by the gun are a running total of weights that are calculated once per shot, trying to take mutual information into account.

Added: 37a39
* 1.19.32: Turned off all weighting, to serve as a baseline comparison for 1.19.33. Custom scaling for many attributes - e.g., third root of AdvancingVelocity, the portions from 0-1 and 7-8 count double in LateralVelocity, and decelerating values for accel are halved. Returned to Gaussian distribution in KernelDensity. Adjusted the GuessFactor bandwidth in the KernelDensity calculation to try and account for an issue David made me aware of. Returned to Gaussian distribution for KernelDensity.

Robo Home | Changes | Preferences | AllPages
Search: