Gaff/VersionHistory

1.44: Surfer killer?

• Updated targeting -- same as 1.43g which topped the heap in the AS Challenge
  • Now uses a slightly different input data set for NNs (10 variables)
  • Features approximately doubled in number
• Many, many small framework tweaks carried over from Pris and Holden development
  • Now adjusts aim for tank movement
  • Hopefully fixes radar lock bug

1.42: Revamped movement

• Restored the orbital dodging mode which was keeping scores up vs simple targeters.
• Dropped the older random orbit movement in favour of the newer Raiko-style one.
• Tweaked some distances and fixed a minor distancing bug.
• Expanded my test bot set and this one worked out better than 1.34.
• Ranked 91 in RoboRumble: 69.9 APS, 1546 ELO, 1869 Glicko2, 2841 battles (14-Aug-2009)

1.40: Improved movement

• Created a pretty slick, low cost orbital movement to replace the ugly iterative one in previous versions
• Smoothed out dynamic distancing
• Replaced the buggy orbital dodge movement mode with another random movement inspired by Raiko's algorithm.
• Ranked 126 in RoboRumble -- 65.10 APS, 1573 ELO, 1803 Glicko2, 2089 battles (30-May-2009)
  • Not-so-improved movement? My tests showed a noticeable increase vs top bots, the rumble showed that I neglected to test against simpler bots, which ate Gaff alive using HOT!

1.34: Combined + improved upon 1.32 targeting methods

• After 15 seasons, squeaks into top spot of fast-learning TC2K7
• Uses targeting methods from 1.32 and 1.32rm and combines the outputs
• Improved input data classification by using radial basis functions
• Tweaked movement to avoid corners somewhat
• Ranked 91 in RoboRumble -- 69.66 APS, 1646 ELO, 2188 battles (26-May-2009)

1.32rm: Alternate version of 1.32 with improved random mover targeting

• Same input data as 1.32, using NN training scheme designed to learn vs. random movers.
• Ranked 96 in RoboRumble -- 69.16 APS, 1675 ELO, 2241 battles (13-May-2009)

1.32: Featuring new & improved anti-surfer targeting

• New input data method to NN targeting scores very high against surfers in TC2K7.
  • Targeting data inspired by data segmentation in Skilgannon's DrussGT
• Fixed bug where waves were not cleared at the end of every round.
• Improved min/max reachable GF calculation and fixed related bug in NN training.
• Ranked 95 in RoboRumble -- 69.32 APS, 1689.8 ELO, 2239 battles (12-May-2009)

1.30: Reworked movement

• Orbital movements now take into account max. reachable angle
• Added dynamic distancing based on energy differential
• Removed cornering movement
• Ranked 96 in RoboRumble -- 69.29 APS, 1714.9 ELO, 3705 battles (9-May-2009)

1.28: Minor tweaks to cornering and movement selection

• Cornering no longer selected when Gaff is already cornered.
• Damage differential now reset every round
• Ranked xx in RoboRumble == 68.45 APS, 1819 ELO, 2175 battles (17-Dec-2009)

1.26: Tweaked movement to improve performance against worst ProblemBots and a selection of top bots. No targeting changes.

• Cornering movement now used only when Gaff has a significant energy advantage
• Random and dodging orbital styles selected based on current damage differential; now switches with same round if needed.
• Bullet predictor decay rate decreased dramatically for faster adaptation.
• Distancing adjusted to make orbital modes similar and decrease advantage of segmenting on distance (makes a big difference).
• Fixed some bugs in the random mode which resulted in strange moves.
• Ranked 106 in OneOnOne (1805.94pts, 2059 battles, 20-Aug-2008)

1.24: Mostly movement updates

• Smoothed out dodging orbit movement mode
• Added a random orbit (non-dodging) movement mode
• Movement selection now settles on new random mode if other two modes lose too often
• Increased bullet predictor decay rate
• Minor tweaks to targeting NN to adapt faster and handle lower bullet powers better (but no challenge score improvement)
• Ranked 97 in OneOnOne (1816.7pts, 2061 battles, 17-Aug-2008)

1.20: Further targeting improvements, many movement bug fixes

• Targeting network no longer uses hidden layers.
• Network is retrained at a slower rate but more often.
• Fixed many, many bugs in the orbital movement and bullet dodging classes.
• Removed the "oddball" movement style
• Changed the movement style selection method
• Ranked 117 in OneOnOne (1800.74pts, 2122 battles, 6-Aug-2008)

1.10: Continued improvements to targeting and some new movement

• Targeting method now trains multiple samples per firing wave
• Fixed a major bug in the bullet dodging code
• Credits for both of the above go to Rednaxela
• Added two new movement styles (orbital and "oddball") for situations where cornering isn't working out.
• Ranked 163 in OneOnOne (1768.96pts, 2225 battles, 31-Jul-2008)

1.04: Major improvements to targeting after spending some time with the targeting challenges and RoboResearch

• Now uses only one targeting method based on a small neural network with only one hidden layer
• Added GuessFactor range limiting using my FutureMove class
• Fixed a nasty bug in FutureMove that also affected dodging
• Ranked 229 in OneOnOne (1688pts, 2007 battles, 27-Jun-2008)

1.00: Initial release

• Ranked 313 in OneOnOne (1608pts, 23-Jun-2008)