WeeklongObsession/VersionHistory - Robo Wiki -= Collecting Robocode Knowledge =-

[MicroBot 1v1 Ratings] [MicroBot 1v1 Premier League] RoboRumble/Participants

? - ?/?/? [details] [compare x.x.x]

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

3.4.1 - 9/19/2007 [details] [compare 3.2.1]

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

3.4.0 - 9/16/2007 [details] [compare 3.2.1]

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 [details] [compare 3.2.1]

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 [details] [compare 3.2.1]

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 [details] [compare 3.2.1]

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 [details] [compare 3.2.0]

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 [details] [compare 3.0.0]

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 [details] [compare 3.1.0] [compare 3.0.0] [compare 2.4.0]

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 [details] [compare 3.0.0] [compare 2.4.0]

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 [details] [compare 2.4.0]

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 [details] [compare 2.3.0]

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 [details] [compare 2.2.4]

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) [details] [compare 2.2.3]

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 [details] [compare 2.2.2] [compare 2.0.0]

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

2.2.2 - 8/24/2007 [details] [compare 2.1.0] [compare 2.0.0] [compare 1.8]

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 [details] [compare 2.1.0] [compare 2.0.0] [compare 1.7.1]

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 [details] [compare 2.1.0] [compare 2.0.0] [compare 1.7.1]

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 [details] [compare 2.0.0] [compare 1.8]

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 [details] [compare 1.8]

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 [details] [compare 1.7.1]

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 [details] [compare 1.7] [comapre 1.6.3c]

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 [details] [compare 1.6.3c]

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 [compare 1.6.3c]

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