LifelongObsession/Version History

From Robowiki
Jump to navigation Jump to search
Sub-pages:
LifelongObsessionVersion History

  • 0.5.1
    • 1v1 – Rating: 2005 (22nd), PL 490/0/19 (21st)
    • Removes the first "feature" mentioned below.
  • 0.5.0
    • 1v1 – Rating: pulled
    • Calculates wave dangers with exactly the guess factors that could still hit (except for some small hidden bug).
    • Desired distance is set in a new way:
      1. If I am NW of the enemy, target distance is the greater of the distance between the enemy and the N or W wall.
      2. If I am SW of the enemy, target distance is the greater of the distance between the enemy and the S or W wall.
      3. Etc.
    • Distancing is generally running toward desired distance at 20 degrees, except when getting very close to the target distance (when it smooths down).
  • 0.4.0
    • 1v1 – Rating: 2006 (22nd), PL 490/0/19 (21st)
    • Added anti-head-on-targeting-without-adjusting-gun-for-robot-turn movement.
  • 0.3.5
    • 1v1 – Rating: 2004 (22nd), PL pulled
    • Ok, looks like that segmented pattern matcher was worth a few points.
    • Revert to 0.3.3
    • LifelongObsession dynamically adjusts the weight of each movement on the fly.
  • 0.3.4
    • 1v1 – Rating: 1993 (25th), PL pulled
    • I was accidentally running a segmented pattern matcher all this time. And I don't think that should be as good as a straight one, like WeeksOnEnd.
  • 0.3.3
    • 1v1 – Rating: 1999 (25th), PL pulled
    • Worked the "missing feature" mentioned in 0.3.1 into the overhauled movement (now when a bullet is found enemy waves that are already in the air are updated).
  • 0.3.2
    • 1v1 – Rating: 1999 (25th), PL pulled
    • Good, 0.3.1 was doing much worse than 0.3.0.
    • Revert to 0.3.0.
    • Simple targeting avoidance is given a higher weight (.75).
  • 0.3.1
    • 1v1 – Rating: pulled
    • This version takes all the new movements out & just tests how much the buggyness & the lacking feature was affecting 0.3.0's result. Hmm ... 0.2.1 also was running at least some on a Robocode 1.1.5 client, which affected its score. Darn, too many variables to control.
  • 0.3.0
    • 1v1 – Rating: 2000 (24), PL pulled
    • 1/2 way complete overhaul of the movement system. It now combines the same WaveSurfing with AntiPatternMatching, anti-head-on-targeting, anti-linear-targeting-when-the-bot-shoots-within-the-walls, and anti-circular-targeting-when-the-bot-shoots-within-the-walls, with varying weights.
  • 0.2.1
    • 1v1 – Rating: 2020 (19th), PL 494/0/17 (19th)
    • Looks like that fold was worth a couple points. I guess the score diff between this an WeeksOnEnd is due to other lurking bugs, or that running only on my machine was better for WeeksOnEnd for some reason (which I don't like, nor anyone else).
    • Revert to 0.1.5
    • Tweak to firepower management: much sooner & slower powering-down when getting low on energy, forget powering down when opponent is low on energy (this is something I can do in WeeksOnEnd for less codesize).
  • 0.0.2 (in rumble as 0.2)
    • 1v1 – Rating: 2014 (22nd), PL 493/0/14 (16th)
    • I am very surprised that folding the PM in half again (see version 0.1) didn't help my ratings. This is a revert back to the WeeksOnEnd style. It should be almost exactly functionally equivalent, except for execution speed optimizations. We'll see what happens.
  • 0.1.5
    • Pulled
    • Bugfix: uhhh.... let's try that last bugfix again.
  • 0.1.4
    • Pulled
    • Bugfix: hopefully fixed an ArrayIndexOutOfBounds exception at start of round, when bots are up close & I'm not yet perpendicular (where GF like 3 is possible).
  • 0.1.3
    • Pulled
    • Bugfix: the movement conduit didn't reset "turnRemaining", so the distancing and precise prediction was slightly messed up.
    • Bugfix: one of the execution speed enhancements didn't work (fixed).
  • 0.1.2
    • 1v1 – Rating: 2017 (22nd), PL 493/0/14 (16th)
    • Bugfix: removed the use of Rules.* methods for rumble compatability.
  • 0.1
    • Pulled
    • This is functionally equivalent to WeeksOnEnd, except for 1 gun improvement: it folds the pattern matcher in half again. It matches all these equivalently: enemy running clockwise forward, clockwise backward, counter-clockwise forward, counter-clockwise backward.