WeeklongObsession/Version History

From Robowiki
< WeeklongObsession
Revision as of 02:50, 28 July 2009 by Voidious (talk | contribs) (migrating from old wiki)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search
WeeklongObsession Sub-pages:
WeeklongObsessionMC2K7 - Version History - Archived Talk 20071128

MicroBot 1v1 Ratings RoboRumble/Participants

? - ?/?/?

  • Micro Rating: ? (?), PL ? (?)
    • Mini Rating: ? (?), PL ? (?)
    • General Rating: ? (?), PL ? (?)
  •  ?
  • Codesize: ?

3.4.1 - 9/19/2007 details

  • Micro Rating: ? (?), PL ? (?)
  • Actually use 2 for the flattener this time, too.
  • Codesize: 685 (including execution speed code)

3.4.0 - 9/16/2007

  • Micro Rating: pulled
  • Adding back only the distancing part of the movement code.
  • Codesize: 686 (including execution speed code)

3.3.1 - 9/17/2007

  • Micro Rating: 1926 (5th), PL 176/0/8 (5th or 6th)
  • Oops, let's try actually reverting the flattener to using 2 this time.
  • Codesize: 651 + execution speed code

3.3.0 - 9/12/2007

  • Micro Rating: 1912 (5th), PL 171/0/15 (10th)
  • Revert flattener to using 2.
  • Some great simplifications to the movement, just to see how much all the codesize is really worth.
  • Codesize: 652 + execution speed code

3.2.2 - 9/8/2007

  • Micro Rating: 1950 (4th), PL 177/0/8 (6th)
  • Tweak to flattener: base flip probability factor from 2->2.5.
  • Codesize: 748

3.2.1 - 9/6/2007

  • Micro Rating: 1964 (3rd), PL 175/0/4 (3rd or 4th)
  • Minor bugfix in matching algorithm (in some cases it could use an older, 1-shorter or same length match).
  • Codesize: 747

3.2.0 - 9/5/2007

  • Micro Rating: 1964 (3rd), PL 172/0/8 (3rd or 6th) I seriously want to know how a major re-build of the matching algorithm yields SUCH similar results. Shouldn't learning twice as fast help?? I guess not against random movement...
  • Re-did the matching algorithm, this time hopefully without the major logical flaw of 3.1.x.
  • Unlimited match length.
  • No more don't-match-when-gun-is-hot conveniences.
  • Codesize: 749

3.1.1 - 9/4/2007

  • Micro Rating: pulled
  • Back to 48 move length for StopNGo.
  • Squeezed a lot of codesize.
  • Only match when the gun is cool (execution time optimization, for release to all rumble clients).
  • Codesize: 748

3.1.0 - 9/3/2007

  • Micro Rating: 1967 (3rd), PL 175/0/6 (3rd or 4th)
  • Matching clockwise as counterclockwise.
  • Oops - also decreased StopNGo move length to 40 (from 48).
  • Codesize: 748

3.0.0 - 8/29/2007

  • Micro Rating: 1963 (3rd), PL 174/0/6 (4th)
  • Replaced the PatternMatching/SingleTick matcher with a standard pattern matcher. So far not matching out-of-bounds, but also not matching clockwise as counterclockwise.
  • Codesize: 719

2.4.0 - 8/29/2007

  • Micro Rating: 1972 (2nd), PL 175/0/6 (4th or 5th)
  • Don't shoot if the gun is still turning. This should be the last step before 3.0.0, where the gun will be replaced with a standard pattern matcher and tweaked with results from Simonton/PMResearch
  • Codesize: 746

2.3.0 - 8/27/2007

  • Micro Rating: 1968 (2nd), PL 174/0/4 (2nd or 3rd)
  • Simplified energy management to min(base firepower, getEnergy() / 16) (which I got from Komarious).
  • Codesize: 736

2.2.4 - 8/27/2007 (barely)

  • Micro Rating: 1966 (2nd), PL 175/0/4 (2nd or 4th)
  • Reduce base firepower from 1200/distance to 900/distance.
  • Codesize: 744

2.2.3 - 8/25/2007

  • Micro Rating: 1957 (3rd), PL 171/0/9 (7th)
  • StopNGo always retreats at 15 degrees.
  • Codesize: 744

2.2.2 - 8/24/2007

  • Micro Rating: 1948
  • Ah, HA! Found the nasty little bug creeping around today - in some testing I set desired distance to 6000. Now it's back to 600.
  • Squeezed codesize.
  • Back to 1/2 velocity matching, since I have the codesize again.
  • Codesize: 744

2.2.1 - 8/24/2007

  • Micro Rating: pulled
  • Ah, I had a bug in 2.0's distancing: it was using 90-35 (=55) for max_retreat, when in was supposed to use 35. Unwittingly fixed in 2.2, but did MUCH worse, so we'll try using 55.
  • Codesize: 749

2.2.0 - 8/24/2007

  • Micro Rating: pulled
  • Revert to matching on full retreating velocity (for 1 codesize).
  • Restore maximum retreating angle.
  • Codesize: 749

2.1.0 - 8/23/2007

  • Micro Rating: 1947 (4th), PL 168/0/10 (7th)
  • StopNGo movement now always travels perpendicular to the enemy.
  • Non-StopNGo movement has no maximum retreating angle.
  • Codesize: 746

2.0.0 - 8/21/2007

  • Micro Rating: 1954 (2nd), PL 169/0/9 (7th)
  • Experimenting with a new movement system that works well in the MC2K6. We'll see how it does against bots that move ... I'm not super optimistic.
  • Codesize: 744

1.8 - 8/21/2007

  • Micro Rating: 1949 (3rd), PL 169/0/10 (7th)
  • Squeezed out a little CodeSize.
  • Match on 1/2 retreating velocity.
  • Codesize: 693

1.7.1 - 8/20/2007

  • Micro Rating: 1951 (2nd), PL 170/0/7 (3rd)
  • A possible bugfix in the way symbols were created/interpreted.
  • Codesize: 697

1.7 - 8/18/2007

  • Micro Rating: 1946 (3rd), PL 169/0/7 (5th)
  • Trying out matching on advancing/lateral velocity (reduces codesize a lot).
  • Codesize: 696


1.6.3

  • MicroBot 1v1 - Rating: 1971 (1st)
    • 1951 (2nd) after a long sabbatical (wow the field changed!)
  • more exact rounding for the symbolic pattern matcher.
  • Codesize: 749

1.6.2

  • Codesize: 747
  • MicroBot 1v1 - Rating: 1975 (1st)
  • squeezed 2 bytes.
  • fixed a logical bug in the in translating 1.5's multi-mode chooser to the almost-equivalent-but-smaller 1.6 version.

1.6

  • Codesize: 748
  • MicroBot 1v1 - Rating: 1935 (?)
  • squeezed codesize.
  • adding enemy energy when he hits me, for stop-and-go accuracy.

1.5

  • Codesize: 749
  • MicroBot 1v1 - Rating: 1977 (1st)
  • squeezed codesize.
  • put back in the niceness of multi-mode choice from 1.2.
  • put back in picky energy management.

1.4

  • Codesize: 749
  • MicroBot 1v1 - Rating: 1957 (2nd)
  • traded a little energy management for *some* of the niceness back in the stop-n-go mode.
  • re-set tuner variables to version 1.0's, to get a good comparison if the matcher makeover was worth its codesize (movement is always more important than I think).

1.3

  • Codesize: 747
  • MicroBot 1v1 - Rating: 1914
  • the pattern matcher got a big makeover. It really squeezes the info into memory now. In fact, we don't even need to limit the size of the HashMap anymore! (for 35 rounds)
  • sacrificed a lot of niceness in the stop-n-go mode.

1.2

  • Codesize: 749
  • MicroBot 1v1 - Rating: 1718
  • brought that disired distance back up ... hoping that's why it wasn't doing as well.
  • found & fixed a couple bugs in the pattern matcher.
  • eliminated a pattern matcher feature that wasn't worth itscodesize.
  • improved the radar lock.

1.1

  • Codesize: 749
  • MicroBot 1v1 - Rating: 1895
  • reduced the average bullet power a touch (by bringing in the desired distance from the enemy a smidge)
  • dropped the memory usage by almost 1/2 (the real reason for this release is that I was seeing OutOfMemoryExceptions in the rumble).
  • sped things up by about 3x (by adding "PATTERN_INCREMENT").

1.0

  • MicroBot 1v1 - Rating: 1944
  • Codesize: 749