Difference between revisions of "RoboRunner/Version History"

From Robowiki
Jump to navigation Jump to search
(RoboRunner 1.2.1 - confidence interval on overall score)
m (fix formula)
Line 13: Line 13:
 
** Add -smart argument to enable smart battle selection. After running 2 battles against each bot (or set of bots), tries to select the battle that will increase the overall score accuracy in the least amount of time.
 
** Add -smart argument to enable smart battle selection. After running 2 battles against each bot (or set of bots), tries to select the battle that will increase the overall score accuracy in the least amount of time.
 
*** It calculates the expected decrease in the [[wikipedia:Standard error|standard error]] that the next battle would provide, divided by the average running time for that battle.
 
*** It calculates the expected decrease in the [[wikipedia:Standard error|standard error]] that the next battle would provide, divided by the average running time for that battle.
*** I.e.: <math>{{SE \over \sqrt{numBattles}} - {SE \over \sqrt{numBattles + 1}}} \over {avgBattleTime}</math>
+
*** I.e.: <math>{{SD \over \sqrt{numBattles}} - {SD \over \sqrt{numBattles + 1}}} \over {avgBattleTime}</math>
 
*** If number of bots is > number of threads, only one instance of a battle configuration is scheduled at a time.
 
*** If number of bots is > number of threads, only one instance of a battle configuration is scheduled at a time.
 
*** A small number of random battles are sprinkled in from among the bots with the fewest battles, to mitigate situations where the variance after 2 battles was randomly low or zero. Probability of running a random battle is: <math>.02 \over 2 ^ {minBattles - 2}</math>, and the battle is randomly chosen from the bots tied for the fewest battles. I'll make this more configurable if/when anyone cares.
 
*** A small number of random battles are sprinkled in from among the bots with the fewest battles, to mitigate situations where the variance after 2 battles was randomly low or zero. Probability of running a random battle is: <math>.02 \over 2 ^ {minBattles - 2}</math>, and the battle is randomly chosen from the bots tied for the fewest battles. I'll make this more configurable if/when anyone cares.
Line 24: Line 24:
 
** Support bot groups in challenges. If present, group scores will be printed and overall score will be average of group scores.
 
** Support bot groups in challenges. If present, group scores will be printed and overall score will be average of group scores.
 
** Print wiki formatted scores for [[Targeting Challenge]] (BULLET_DAMAGE) or [[Movement Challenge|Movement Challenge]] (MOVEMENT_CHALLENGE) scoring styles.
 
** Print wiki formatted scores for [[Targeting Challenge]] (BULLET_DAMAGE) or [[Movement Challenge|Movement Challenge]] (MOVEMENT_CHALLENGE) scoring styles.
** Print scores against each bot/set of bots and overall score after all battles are run. Run with 0 seasons to see challenge scores.
+
** Print scores against each bot/set of bots and overall score after all battles are run. Run with 0 seasons to see overall scores.
 
* 1.1.0 - 7/28/2012
 
* 1.1.0 - 7/28/2012
 
** Support for all basic scoring options: percent score, survival firsts, survival score, bullet damage (TC), and movement challenge.
 
** Support for all basic scoring options: percent score, survival firsts, survival score, bullet damage (TC), and movement challenge.

Revision as of 04:08, 15 August 2012

RoboRunner Sub-pages:
RoboRunnerVersion History - Help

  • 1.2.1 - 8/14/2012
    • Display approximate 95% confidence interval for overall score. Uses a faster/less accurate version (accurate to a range of ~.02) when showing overall score between battles, with a more thorough calculation in the final score (accurate to 2 decimal places).
  • 1.2.0 - 8/12/2012
    • Change data file format to use gzipped XML files with raw scores for every individual battle. Data files are about the same size as before (but only because they are zipped).
    • Display confidence intervals for individual bot scores. (E.g.: 89.15 +- 1.05 = 95% confidence the true average score is within 1.05 of 89.15.)
    • Add -smart argument to enable smart battle selection. After running 2 battles against each bot (or set of bots), tries to select the battle that will increase the overall score accuracy in the least amount of time.
      • It calculates the expected decrease in the standard error that the next battle would provide, divided by the average running time for that battle.
      • I.e.: <math>{{SD \over \sqrt{numBattles}} - {SD \over \sqrt{numBattles + 1}}} \over {avgBattleTime}</math>
      • If number of bots is > number of threads, only one instance of a battle configuration is scheduled at a time.
      • A small number of random battles are sprinkled in from among the bots with the fewest battles, to mitigate situations where the variance after 2 battles was randomly low or zero. Probability of running a random battle is: <math>.02 \over 2 ^ {minBattles - 2}</math>, and the battle is randomly chosen from the bots tied for the fewest battles. I'll make this more configurable if/when anyone cares.
  • 1.1.2 - 8/5/2012
    • Add -t argument to set number of threads, up to number of Robocode installs configured in roborunner.properties.
    • Add -wiki argument to force wiki formatted output, needed for challenges like MC2K7 that use regular scoring.
    • Configure JVM arguments and additional robot directories in roborunner.properties. Should gracefully upgrade existing installs.
    • Print percentage of bots faced until it's 100%.
  • 1.1.1 - 7/29/2012
    • Support bot groups in challenges. If present, group scores will be printed and overall score will be average of group scores.
    • Print wiki formatted scores for Targeting Challenge (BULLET_DAMAGE) or Movement Challenge (MOVEMENT_CHALLENGE) scoring styles.
    • Print scores against each bot/set of bots and overall score after all battles are run. Run with 0 seasons to see overall scores.
  • 1.1.0 - 7/28/2012
    • Support for all basic scoring options: percent score, survival firsts, survival score, bullet damage (TC), and movement challenge.
    • In Melee battles, also track and display scores against each bot.
  • 1.0.1 - 7/26/2012
    • Bug fix: when displaying overall score, only count bots in the current challenge, not all bots the challenger has faced.
  • 1.0.0 - 7/25/2012
    • Initial release.
    • Multi-threading with persistent Robocode engines.
    • 1v1, Melee, custom battle field sizes.
    • Collects all types of scoring (score, survival firsts, survival score, bullet damage) but can only display APS.
    • Supports RoboResearch challenge (.rrc) files, except for scoring groups.