Difference between revisions of "DeBroglie/Version History"

Revision as of 03:41, 26 June 2012

rev0073 -- Narrower kernel width on targeting histogram. Curve normalization fix on movement histogram.
- rev0073 - Accidentally included experimentally turning off culling data points via displacement vectors.
- rev0073b - Turned point culling back on. RoboRumble ‒ APS: 81.98% (36th), PL: 886-17 (15th), Survival: 90.85%

rev0072 -- Complete rewrite of the data histogram class. Much much simpler code, major gain in performance. Doesn't look like the bot is skipping as many turns now (crossed fingers). Each data point's precise intersection width is used for its kernel width. Also went ahead and did Gaussian kernel density on targeting.

rev0069 through rev0071 -- Changing to Gaussian kernel density on surfing.

rev0064 through rev0068 -- Moved to Eclipse for development instead of Robocode's built in editor. (I had major problems with the new 1.7.4.0 editor)

rev0061 through rev0063 -- Hunting bugs in bot initialization that led to dropped rounds.

rev0057 -- License change from RWPCL to zlib. Modified zlib on some source files.

rev0056 -- Surfing two waves. RoboRumble ‒ APS: 80.14% (51st), PL: 877-21 (18th), Survival: 89.63%

rev0049 through rev0055 -- Data handling changes for surfing, distance control, very basic (bad) anti-ram management. Normalized hitrate and histograms for flattener.

rev0048 -- Bad surfing bugs squashed. This bot is now essentially like my own take on BasicGFSurfer with more segmentation and better data via precise prediction, precise intersection and dynamic clustering. The raw basics of the bot are complete. :)

rev0045 through rev0047 -- Tracking down major bugs in surfing implementation.

rev0044 -- First version that actually wave surfs (poorly). The bot is now using gun and movement written by me. Aristocles movement abandoned.

rev0026 through rev0043 -- Finishing work on the gun, playing around with Random Movement challenge and RoboResearch.

rev0025 -- Bot discards situations with displacement vectors that lead outside of the battlefield.

rev0024 -- Bullet collisions now delete relevant real waves. Precise wave intersection code steamlined for speed. Added Rednaxela's FastTrig (taken from DrussGT's source, since I had no internet access), also for speed. Some data gathering tweaks.

rev0016 -- Removed Real/Virtual wave from kd-tree segmentation.. added distance.

rev0012 -- Better (but still relatively naive) escape angles, and a more detailed set of segments for distance from the wall.

rev0008 -- Moved to using version numbers off of my Mercurial repository. Some improvements in Segmentation.

v 0.85 -- Rewritten from the ground up. Using the SqrEuclid version of Rednaxela's KD-tree implementation, so deBroglie counts as a DC bot now, I suppose. The movement is from PEZ's Aristocles. It's a placeholder for now while I iron out targeting issues.

v 0.66 -- 0.65 was significantly weaker against some bots than 0.55 was. Did some downloading and trial battles against a few of the biggest changes. Happened because I removed the ram behavior from my old oscillator. Put ram behavior back in, including some aggressive re-ramming and gun shooting for fun.

v 0.65 -- Bug in 0.6 fixed. I'd accidentally passed the bullet velocity to my escape angle predictor as "20 - 3/firepower" instead of "20 - 3*firepower." Oops! The wave surfer learns a little slow, but often does nice, tight dodging, so I've got a good foundation to build on, I think.

v 0.6 -- Implemented goTo wave surfing. Nice boost against top 25ish bots, but when it went into the rumble, a particularly ugly bug reared its head. Skipped turns badly against weak bullets.

v 0.55 -- v 0.5 was using stupid amounts of memory. Hanging on RoboRumble with DrussGT and a few others. Trimmed things down. Definitely need to implement a growable data structure of some kind for stats.

v 0.5 -- Rewritten angleToBin and binToAngle to compensate for clockwise/CCW movement. This allowed elimination the cw/ccw stats segment. Double the data density with half the memory use. Consequently added a few bins and timing segments. Still a net memory savings. Also included last velocity in cw/ccw calculation if current v == 0. Seemed wrong for v == 0 to always be clockwise.. the enemy might have a pattern to exploit when crossing v == 0 after all.

v 0.4 -- full max escape angle calculation instead of just the polar coordinate version.

v 0.2 -- First version. Gutted and rebuilt my old Virtual Bullets + Anti-gravity bot Newton.

@@ Line 1: / Line 1: @@
 *'''rev0073''' -- Narrower kernel width on targeting histogram.  Curve normalization fix on movement histogram.
 ** rev0073 - Accidentally included experimentally turning '''off''' culling data points via displacement vectors.
-** rev0073b - Turned point culling back on.
+** rev0073b - Turned point culling back on. {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=roborumble&name=tjk.deBroglie+rev0073b|rumble=RoboRumble|scorelabel=APS|score=81.98|rank=36th|win=886|loss=17|plrank=15th|glicko2=2002.1|score2label=Survival|score2=90.85}}
 *'''rev0072''' -- Complete rewrite of the data histogram class. Much much simpler code, major gain in performance. Doesn't look like the bot is skipping as many turns now (crossed fingers). Each data point's precise intersection width is used for its kernel width.  Also went ahead and did Gaussian kernel density on targeting.