Difference between revisions of "Holden/VersionHistory"

Latest revision as of 22:47, 5 September 2009

Widened number of bins updated on a bullet hit
Fixed surfing algorithm to allow overshoot of GF+/-1 locations (was slowing down too soon)
Normally I release only even numbered versions (odd numbers are for testing), but I've been unable to repeat this version's performance after some recent refactoring!

RoboRumble ‒ APS: 80.27% (35th), PL: 721-16 (14th), Survival: 89.77%
Replaced my crude anti-ramming strategy with a slightly better one inspired by Skilgannon's description of how DrussGT does it. Holden doesn't do much surfing in this mode, but the movement planning is far better than the orbit-the-arena-center strategy used since 1.06.
Tweaked flattener to add minimal flattening after first loss
Primes GF0 as a dangerous location at the start of the round
Adjusted bot width calculation, fixed but in state recording
Version 1.14 is already ready & waiting in the wings once this version stabilizes... need to break into the top 20!
Retired 5-Sep-2009

RoboRumble ‒ APS: 79.81% (35th), PL: 718-18 (14th), Survival: 89.6%
Rewrote surfing class to make it more precise and extensible
Now uses same algorithm for prediction and movement
Bullet hits now use actual bullet heading (surprisingly I'm so used to targeting that I didn't realize this was possible!)
Improved bot width calculation and accounts for multiple tick exposure to waves at movement extremes
Now surfs in a true orbit around enemy, not around the wave origin
Shortened orbit radius to make better use of Holden's targeting advantage
Some bullet power tweaks to improve energy usage, especially during the endgame

RoboRumble - APS: 78.69% (41st), PL: 711-26 (23rd), Glicko-2: 2007.4, Survival: 87.09%
Now uses Gaff's latest anti-surfer gun.
Added rambot detection & avoidance
Tweaked lazy flattener to kick in only after two lost rounds; no longer decreases after a win.
Reduced preferred distance by 50
Switched from variable bullet powers to prefer 2.0
Retired 19-Aug-2009

RoboRumble - APS: 77.21% (47th), PL: 711-26 (23rd), Glicko-2: 1983, Survival: 87.6%
Added a flattening scheme which basically logs low-valued hits to the segmented buffers. This kicks in when Holden loses a round, increases if he keeps losing and decreases if he starts to win. Simple but effective.
Retired 16-Aug-2009

RoboRumble 1v1: 47th, 77.2% APS, 1713 ELO, 1984 Glicko2
Adjusted a few segments (more lateral velocity, less accel and wall distance).
Original version used bot width when calculating danger values; this one also uses width when storing hits.
Increased minimum distance.
Retired 25-Jun-2009

@@ Line 1: / Line 1: @@
+;1.13a - Wider surfing
+* Widened number of bins updated on a bullet hit
+* Fixed surfing algorithm to allow overshoot of GF+/-1 locations (was slowing down too soon)
+* Normally I release only even numbered versions (odd numbers are for testing), but I've been unable to repeat this version's performance after some recent refactoring!
 ;1.12a - Better anti-ramming + Bugfixes
+* {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=roborumble&name=darkcanuck.Holden+1.12a|rumble=RoboRumble|scorelabel=APS|score=80.27|rank=35th|win=721|loss=16|plrank=14th|glicko2=2037.5|score2label=Survival|score2=89.77}}
 * Replaced my crude anti-ramming strategy with a slightly better one inspired by [[Talk:Ramming_Movement#Strategy | Skilgannon's description]] of how [[DrussGT]] does it.  Holden doesn't do much surfing in this mode, but the movement planning is far better than the orbit-the-arena-center strategy used since 1.06.
 * Tweaked flattener to add minimal flattening after first loss
@@ Line 5: / Line 12: @@
 * Adjusted bot width calculation, fixed but in state recording
 * Version 1.14 is already ready & waiting in the wings once this version stabilizes...  need to break into the top 20!
+* Retired 5-Sep-2009