Toorkild/VersionHistory

From Robowiki
Jump to navigation Jump to search

0.4.5: codesize 718

  • Shrinking, make sure nothing's broken

0.4.4: codesize 724, APS 84.30

  • Back to 50 matches
  • Use MEA to help inform decision on weight of flattener hits

0.4.3: codesize 728, APS 84.01 (Common APS tied with 0.4.1)

  • Remove distance segment
  • Remove extra term from StopAndGo
  • Make the number of desired matches dependant on the amount of data collected

0.4.2: codesize 744, APS 83.79

  • Add distance segmentation

0.4.1: codesize 733, APS 83.93

  • Caught a bug with energy tracking with onHitByBullet (subtract difference instead of add)
  • Make flattener more likely to enable (tuned against UnfoolableNano, Quarrelet etc)

0.4.0: codesize 733, APS 83.60

  • Remove bullet power rolloff
  • Add my extra term to Stop And Go distance
  • Add back energy tracking to onHitByBullet (my thoughts were that if I was getting hit anyways, clearly the StopAndGo wasn't working, oh well)
  • Keep track of number of bullet hits and use this to inform the flattener switch decision.

0.3.9: codesize 705, APS 83.67

  • Add back bullet power rolloff as my energy gets low.

0.3.8: codesize 694, APS 83.69

  • Shrink a little more
  • Revert the aiming change

0.3.7: codesize 698, APS approx 83.5

  • Shrink and see what little, possibly useless features I can remove. This release is to make sure they weren't actually important.
  • I was aiming 1 bin off to the side. Whoops.

0.3.6: codesize 741, APS 83.86

  • Give forward-wall another shot

0.3.5: codesize 710, APS 83.48

  • Remove binsmoothing and interpolation
  • Better bullet fire detection
  • Correctly round bin locations

0.3.4: codesize 745, APS 82.47

  • Add back binsmoothing
  • Shoehorn in sub-bin-accuracy interpolation

0.3.3: codesize 712, APS 83.52

  • Back to 0.2.4b
  • Remove binsmoothing
  • Add the structured break to reduce my codesize a little

0.3.2: codesize 715, APS 83.29

  • Fix the wall-segment-using-my-latVel bug

0.3.1: codesize 726, APS 82.xx projected(pairings incomplete)

  • Add a forward wall segment to the pattern matching symbol
    • Aaargh! Using my latVel for the wall direction instead of theirs...

0.3.0: codesize 726, APS 83.38

  • Shrink right down to 689 bytes by:
    • Remove binsmoothing
    • Change booleans to ints for mode variables
    • Add structured break
    • Delete some extras
  • Reverse on enemy death if I have my flattener enabled
  • More matches
  • More bins
  • Aim from next location

0.2.4b: codesize 749, APS 83.61

0.2.5: codesize 744

  • Built off of 0.2.4
  • Match off of regular adv-vel instead of halved value

0.2.4 - accidental re-release of 0.2.2 - Score: 81.84

0.2.3: codesize 749

  • Remove the bin-smoothing of 0.2.0

0.2.2: codesize: 749 Score: 81.70

  • Reduce number of matches from 150 to 50

0.2.1: codesize: 749 Score: awaiting re-release results

  • Only match with half advancing velocity granularity

0.2.0: codesize: 744 Score: 81.53

  • Change number of bins from 230 to 255 (saves 2 bytes)
  • Added binsmoothing

0.1.10: codesize: 724 Score: 81.67

  • Slowly adding back features from 0.1.6: this release add back decreasing bullet power when enemy energy is low.

0.1.9: codesize: 715 Score: 81.25

  • Back to the old flattener enabler, but make it more likely to enable
  • Steal and improve Thorn's variable distance stop-and-go code - it is now 100% optimum, ie. impossible to make improvements
    • This is actually being harsh on myself, I derived a functional equivalent, but did *8 instead of <<3.

0.1.8: codesize: 723 Score: bad

  • A variation on the flattener enabler

0.1.7: codesize: 723 Score: 80.71

  • Implement a more reliable flattener-enabler - might still need some tuning though
  • Shoot with bullet power 1.99
  • Shoot against rambots with 2.99
  • Increase number of matches from 50 to 150

0.1.6: codesize: 698 Score: 80.955

  • Removed lots of little things that I believe don't add much to the score.

0.1.5: codesize: 748 Score: 81.08

  • Always shoots 1/16th of a pixel just to one side (no more random)
  • No longer matches across rounds (required black magic and bad practices to get the shrinking to work!)
  • Reduced the granularity on the lateral-velocity to 2 pixels-per-tick instead of 1 to follow what has been most successful with GF/VCS guns

0.1.4: codesize: 740 Score: 81.19 (drift downwards, scored approx. same as 0.1.3 at release)

  • Random factor is now +- 1/16th of a pixel

0.1.3: codesize: 733 Score: 81.33 APS (took common pairings APS with Thorn)

  • Made the random factor exactly +-1 pixel instead of a random value somewhere between -4 and +4 pixels

0.1.1CT: codesize: 733 score: 80.9 APS

  • Matches across rounds again for more codesize
  • CT: Catch This! Introduces a random factor in aim to prevent enemy from shooting down bullets

0.1.2: codesize: 749 score: pulled

  • Somehow managed to squeeze the codesize down even further
  • Shoots 3 power bullets if the last tick revealed a match over 30 ticks long
    • There was a bad bug somewhere...

0.1.1: codesize: 749 score: 2008

  • Switched to bit-shifting to store both adv-vel and lat-vel in one char, instead of a different char for each. Prevents 'half-off' matches that don't get rebuilt properly.
  • Doesn't match or rebuild across rounds
  • Utilized the Arrays.sort and Arrays.binarySearch to reduce my codesize - so now depends on Java 1.5+ (as does Robocode, so it shouldn't be a problem)

0.1: codesize: 742 score: 2006 after 1576 First release!