Difference between revisions of "Midboss/Version History"

From Robowiki
Jump to navigation Jump to search
(Midboss 1l)
(Version 1q)
 
(5 intermediate revisions by the same user not shown)
Line 7: Line 7:
 
----
 
----
 
<!-- {{subst:rumblestats:roborumble|ags.Midboss x|RumbleStatsDefault}} -->
 
<!-- {{subst:rumblestats:roborumble|ags.Midboss x|RumbleStatsDefault}} -->
 +
 +
* '''1q''' - 19/01/2010
 +
** Increase depth of 'precise prediction' (very different meaning than surfing) in the score prediction code, to be somewhat greater than version 1k (versions after 1k had it reduced a fair bit). Some tests indicate that this should fix the score loss since version 1k finally.
 +
** Sparsely allocate gun buffers as needed, instead of allocating all '''164250''' VCS buffers (each having 45 bins of <code>double</code> type) at once...
 +
*** Let's see... 164250*45*8 = 59130000 bytes = 56MB + Java's overhead per array... Ouch :P
 +
*** This '''greatly''' reduces memory usage, to 1/10th of prior usage, since in a typical battle only about 10% of those VCS buffers actually get written to anyway ;)
 +
 +
* '''1p''' - 19/01/2010
 +
** {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=roborumble&name=ags.Midboss+1p|rumble=RoboRumble|scorelabel=APS|score=83.06|rank=11th|win=728|loss=23|plrank=18th|glicko2=2091.9|score2label=Survival|score2=90.85}}
 +
** Fix some bugs in the 'firepower selection'/'score estimation' code.
 +
 +
* '''1o''' - 18/01/2010
 +
** {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=roborumble&name=ags.Midboss+1o|rumble=RoboRumble|scorelabel=APS|score=82.86|rank=13th|win=731|loss=20|plrank=16th|glicko2=2087.8|score2label=Survival|score2=91.97}}
 +
** Improve the hitrate estimator algorithm notably
 +
** Make some weird rare NaN bug in the surfing not cause infinite loop
 +
 +
* '''1n''' - 18/01/2010
 +
** {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=roborumble&name=ags.Midboss+1n|rumble=RoboRumble|scorelabel=APS|score=82.83|rank=13th|win=721|loss=31|plrank=31st|glicko2=2088.3|score2label=Survival|score2=92.15}}
 +
** Anti-bullet-shielding tweak
 +
** Other minor tweaks... Seeing how much of score loss was random chance
 +
 +
* '''1m''' - 16/01/2010
 +
** {{RumbleStatsDefault|link=http://darkcanuck.net/rumble/RatingsDetails?game=roborumble&name=ags.Midboss+1m|rumble=RoboRumble|scorelabel=APS|score=82.89|rank=13th|win=725|loss=25|plrank=20th|glicko2=2088.9|score2label=Survival|score2=92.07}}
 +
** Fix various major bugs in 1l, should be good now...
  
 
* '''1l''' - 16/01/2010
 
* '''1l''' - 16/01/2010
−
** ''Rating Pending''
+
** ''Pulled''
 
** Improve accuracy of score prediction by using real statistical approximation with normal distributions for the approximation that's used when unable to simulate deeply.
 
** Improve accuracy of score prediction by using real statistical approximation with normal distributions for the approximation that's used when unable to simulate deeply.
  

Latest revision as of 07:15, 20 January 2010

Midboss Sub-pages:
Midboss - Version History
  • 1q - 19/01/2010
    • Increase depth of 'precise prediction' (very different meaning than surfing) in the score prediction code, to be somewhat greater than version 1k (versions after 1k had it reduced a fair bit). Some tests indicate that this should fix the score loss since version 1k finally.
    • Sparsely allocate gun buffers as needed, instead of allocating all 164250 VCS buffers (each having 45 bins of double type) at once...
      • Let's see... 164250*45*8 = 59130000 bytes = 56MB + Java's overhead per array... Ouch :P
      • This greatly reduces memory usage, to 1/10th of prior usage, since in a typical battle only about 10% of those VCS buffers actually get written to anyway ;)
  • 1p - 19/01/2010
    • RoboRumble ‒ APS: 83.06% (11th), PL: 728-23 (18th), Survival: 90.85%
    • Fix some bugs in the 'firepower selection'/'score estimation' code.
  • 1o - 18/01/2010
    • RoboRumble ‒ APS: 82.86% (13th), PL: 731-20 (16th), Survival: 91.97%
    • Improve the hitrate estimator algorithm notably
    • Make some weird rare NaN bug in the surfing not cause infinite loop
  • 1n - 18/01/2010
    • RoboRumble ‒ APS: 82.83% (13th), PL: 721-31 (31st), Survival: 92.15%
    • Anti-bullet-shielding tweak
    • Other minor tweaks... Seeing how much of score loss was random chance
  • 1m - 16/01/2010
    • RoboRumble ‒ APS: 82.89% (13th), PL: 725-25 (20th), Survival: 92.07%
    • Fix various major bugs in 1l, should be good now...
  • 1l - 16/01/2010
    • Pulled
    • Improve accuracy of score prediction by using real statistical approximation with normal distributions for the approximation that's used when unable to simulate deeply.
  • 1k - 13/01/2010
    • RoboRumble ‒ APS: 83.37% (11th), PL: 733-19 (16th), Survival: 91.54%
    • Noticed issues against rambots happening again, and rethought some overcomplicated parts of the firepower calculation. Should work better now.
  • 1j - 13/01/2010
    • Pulled
    • Fix some minor little bugs/inaccuracies in the score estimation code, particularly for the endgame and infrequently firing bots
    • Add a new "Low segmentation slow rolling" gun buffer
      Old gun buffers were:
      • "High segmentation no rolling"
      • "High segmentation fast rolling"
      • "Surfing buffer simulator" (ends up getting negative weight against surfers)
  • 1i - 12/01/2010
    • RoboRumble ‒ APS: 82.98% (12th), PL: 732-19 (16th), Survival: 90.95%
    • Fix yet another minor bug in firepower selection
    • Improve enemy wave detection to have a check for gunheat like RougeDC does, this seems to fix Rambot performance
  • 1h - 11/01/2010
    • RoboRumble ‒ APS: 82.78% (15th), PL: 731-20 (16th), Survival: 91.19%
    • Add fancy code to ensure that static data is garbage collected at end of battle, because apparently Robocode doesn't do this right
    • Fix some bugs/inaccuracies in the score estimation code.
  • 1g - 10/01/2010
    • Bad results
    • Fixing various flaws that remained in version 1f.
  • 1f - 09/01/2010
    • Bad results
    • Fixing a terrible bug in version 1d's new code.
  • 1e - 09/01/2010
    • Bad results
    • Adding some new bullet power selection.
  • 1d - 09/01/2010
    • RoboRumble ‒ APS: 82.73% (15th), PL: 732-20 (16th), Survival: 91.55%
    • Added a couple more segments.
Retrieved from "http://robowiki.net/w/index.php?title=Midboss/Version_History&oldid=14560"