SilverSurfer/Version History

From Robowiki
Jump to navigation Jump to search
SilverSurfer Sub-pages:
SilverSurferVersion History - Wave Suffering - Help Requests - Archived Talk 20060602

2004-11-29: version 2.66
Based v2.65
Heavy segmentation: DISTS = 3; VELS = 9; ACCS = 3; HEADS = 4; WALLS = 3;
EnemyWave hitsRA splitted in 3 distance segments (0-200,200-400,400-).
Danger graph for WeakGun only with Hits.(no allHits)
No waves readyToGo avaliation for WeakGuns.
Result:

2004-11-28: version 2.65
Based v2.64
Shortening escape envelope for HOTs if and only if bullet power > 1.7.
back with old Multi-Wave weightening:(1 - ((1D / 55D) * Math.pow(arrive, 2)));
removed no-wall-smoothing for WeakGuns (2.64).
Result:2043.58

2004-11-28: version 2.64
Based v2.63
Removed the historyHits.
Multi-Wave weightening:(1 - ((1D / 20D) * Math.pow(arrive, 2)));, was:(1 - ((1D / 55D) * Math.pow(arrive, 2)));
Using 9 velocity segments, instead of 5.
Tweaked WeakGun movement to a behaviour very similar to anti-HOT moving (without shortening escape envelope).
Result:2035.45

2004-11-27: version 2.63
Based v2.62
Removed "rammerKiller" of 2.62
hitsAvg returns 0.0 if hitsCount < 10.
Added a historyHits to danger graphing: allHits+Hits+0.5*visits+0.25*hitsHistory (hitsAvg >= 0.07)
Result:2052.26

2004-11-25: version 2.62
Based v2.61
if rammer and distance < 300, firePwr=3.0
Result:2045.46
(Strange diff for such a small stuff...)

2004-11-24: version 2.61
Based v2.60
hits depth set from 3 to 1. allHits depth set from 3 to 1. visits depth set from 20 to 10. selecting danger graph (3 types) based in hits avg:
.allHits+longAllHits(HOT)
.allHits+Hits (hitsAvg < 0.07)
.allHits+Hits+0.5*visits (hitsAvg >= 0.07)
Result:2058.94

2004-11-24: version 2.60
Based v2.59
back to holding fire at pwr 8.
hits depth set from 3 to 1.
Result:2052.17

2004-11-23: version 2.59
Based v2.53.33
Holding fire at pwr 17.
Result:2048.57

2004-11-22: version 2.58
Based v2.53.33
Back to 2 dangers graphs (Hot & non-Hot).
Result:2055.96

2004-11-21: version 2.57
Based v2.53.33
Back to 2.53.33. selecting danger graph (3 types) based in hits avg.
Result:2054.72

2004-11-16: version 2.56
Based v2.55
Attempt to an auto-select of danger graph.
Result:2034.80

2004-11-14: version 2.55
Based v2.54
Tired of those boring little tweaks, and of this score hunting, it´s time for a fun-time version!
I call this an "Ashtray" version ;) (not for me, is suposed...). Probably it would reflect also somewhere, but this test is also to find if.
Also "corrected and optimized" firePower.
Result:2048.53

2004-11-13: version 2.54
Based v2.53.33
recalled down to version 2.53.33, something went wrong after it, or it was only luck... anyway, back with it.
Better ramming detection, bullet power explicitly 3.0 against rammers. Particularly better against "oportunity rammers", like Cigaret.
Result:2058.23

2004-11-11: version 2.53.39
Based v2.53.38
Gun:
Back with bonus for last hits:

if(PatternSample.resetHits){
   hits=0;
}else if(hits>0){
   rate /= 1D+(1D - Math.pow((double)hits/(double)MAX_HITS,2 ));
   hits--;
}

Back with old latest samples bonus formula, now:

rate /= 1D + ((difAge < AGING_LIFE)?1D*(1D-Math.pow(((double)difAge/(double)AGING_LIFE),2D)):0);

old:

rate /= 1D + ((difAge < AGING_LIFE)?3D*(1D-Math.pow(((double)difAge/(double)AGING_LIFE),0.5)):0);

Result:2053.8

2004-11-10: version 2.53.38
Based v2.53.36
Gun:
distance-to-corner stronger(1 factor) at corners, weaker(null, factor 0) at middle field.

double cdiff = REFERENCE.distanceCorner - this.distanceCorner;
...		
double minCornerDist = Math.min(REFERENCE.distanceCorner,this.distanceCorner);
double x = minCornerDist/(AxeBot .getIt().getFieldDiagonal()/4D); 
double cornerWeight = 1-(x*x); // 1-((x/250)^2)
cornerWeight = (cornerWeight<0)?0:cornerWeight;
...
penalty += (rate * (cdiff * cdiff / 20000.00) * cornerWeight);

PatternSample rating: time-since-deccelerating , distance , heading, distance-to-wall & distance-to-corner weightened.
Result:2059.22

2004-11-09: version 2.53.36
Based v2.53.35
Gun:
Back to distance-to-corner (old). PatternSample rating: time-since-deccelerating , distance , heading, distance-to-wall & distance-to-corner.
Result:2056.54

2004-11-09: version 2.53.35
Based v2.53.34
Gun:
Changed the mirror-distance scheme for distance-to-center (PEZ again). PatternSample rating: time-since-deccelerating , distance , heading, distance-to-wall & distance-to-center.
Result:2057.22

2004-11-08: version 2.53.34
Based v2.53.33
Gun:
Changed distance-to-corner for a mirror-distance scheme (distance to the closest "mirrored" reference point). PatternSample rating: time-since-deccelerating , distance , heading, distance-to-wall & mirror-distance.
Result:2059.64

2004-11-07: version 2.53.33
Based v2.53.32
Gun:
Added distance-to-corner (old). PatternSample rating: time-since-deccelerating , distance , heading, distance-to-wall & distance-to-corner.
Result:2063.94 ** BEST SCORE EVER **

2004-11-07: version 2.53.32
Based v2.53.31
Gun:
back to AGING_LIFE=100; in:

rate /= 1D + ((difAge < AGING_LIFE)?3D*(1D-Math.pow(((double)difAge/(double)AGING_LIFE),0.5)):0);

Result:2059.50

2004-11-05: version 2.53.31
Based v2.53.30
Gun:
At PatternSample.rate():

//PEZ dicovered this "bug"(?)...
//I was using rate += (rate * ... without using a bonus
//variable.
//The rate calculation was not what i expected,
//not sure if it would make any difference, but 
//at least gives me a different rate...
//Dunno however if it will make any dif, or even
//break it, but i´m curious...
//So, all Blame or Shame to PEZ...
double bonus = 0;
bonus += (rate * (tsdeaccdiff * tsdeaccdiff / 200.00));
bonus += (rate * (ddiff * ddiff / 20000.00));
bonus += (rate * (hdiff * hdiff / 500.00));
bonus += (rate * (wdiff * wdiff / 20000.00));
rate += bonus;

Result:2061.27

2004-11-05: version 2.53.30
Based v2.53.28
Gun:
an inverse curve comparing to previous. AGING_LIFE=200:

rate /= 1D + ((difAge < AGING_LIFE)?3D*(1D-Math.pow(((double)difAge/(double)AGING_LIFE),<b>0.5</b>)):0);

Result:2058.97

2004-11-04: version 2.53.29
Based v2.53.28
Gun:
AGING_LIFE=200; was 100. in:

rate /= 1D + ((difAge < AGING_LIFE)?3D*(1D-Math.pow(((double)difAge/(double)AGING_LIFE),2D)):0);

Result:2056.04

2004-11-04: version 2.53.28
Based v2.53.27
Gun:
Removed the bonus to the hitted in samples.
Removed the bonus to the last-hitted in samples.
Stronger bonus for newer samples:

rate /= 1D + ((difAge < AGING_LIFE)?3D*(1D-Math.pow(((double)difAge/(double)AGING_LIFE),2D)):0);

Result:2060.94

2004-11-03: version 2.53.27
Based v2.53.25
Cleaned old code (some) Gun:
Adding a bonus to the hitted samples:

rate /= (hit)?2D:1D;

Result:2047.58 - buggy?

2004-11-03: version 2.53.25
Based v2.53.23
Gun:
Back to old vector search in PM:
{1, 2, 3, 4, 5, 6, 7, 10, 15, 20, 30, 40};
Adding a bonus to the 10 last hit samples:

if(hits>0){
   rate /= 1D+(1D - Math.pow((double)hits/(double)MAX_HITS,2 ));
   hits--;
}

Result:2062.18

2004-11-02: version 2.53.23
Based v2.53.22
Gun:
Trying a more "wide" vector search in PM:
now:{1, 2, 3, 4, 5, 7, 10, 15, 20, 30, 40, 50};
Result:2053.56

2004-11-01: version 2.53.22
Based v2.53.21
Gun:
Trying a more "fast" vector atualization in PM:
now:{1, 2, 3, 4, 5, 6, 7, 10, 15, 20, 25, 30};
before:{1, 2, 3, 4, 5, 6, 7, 10, 15, 20, 30, 40};
Result:2059.34

2004-11-01: version 2.53.21
Based v2.53.20
Gun:
Back with the old shooting blocking scheme... Added a attempt to faster-learning gun, rating (AGING_LIFE=100):

rate /= 1D + ((difAge < AGING_LIFE)?(1D-Math.pow(((double)difAge/(double)AGING_LIFE),2D)):0);

Result:2061.15

2004-11-01: version 2.53.20
Based v2.53.19
Gun:
Trade the shooting block for bullet power restriction:

if ((lifeAfter <= 8) && (lifeAfter < hisLife) && (target.getDistance() > 150)) {
   pwr= 0.1;
}

Result:2056.15

2004-10-30: version 2.53.19
Based v2.53.18
Gun:
Fixed(?) a positional bug(?) at gun. if i calculate 1 tick earlier, the center of firing should be the next(t+1), right? Extrapolating shooting pos at getFuturePos().
Result:2059.21

2004-10-29: version 2.53.18
Based v2.53.17
Gun:
MultipleChoice grain set back to 20 pixs.
MultipleChoice back to smoothed 1/(n^2).
Result:2058.13

2004-10-29: version 2.53.17
Based v2.53.16
Movement:
Added a simple distance control system: "retreats" if (3*me.getLife() < 2*enemy.getLife())
Result:2052.62

2004-10-29: version 2.53.16
Based v2.53.15
Gun:
Initial value for rating back to 0.
MultipleChoice grain set from 20 pixs to 15.
MultipleChoice unsmoothed (if PEZ says that neither him or Jam use smoothin´ in their GF guns, whom am i to disagree? - Credits, credits...).
Result:2044.96

2004-10-27: version 2.53.15
Based v2.53.14
Gun:
PatternSample rating: time-since-deccelerating , distance , heading & the distance-to-wall scheme.
Initial value for rating set from 0 to 10.
Result:2049.91

2004-10-26: version 2.53.14
Based v2.53.13
Gun:
Back to old-fashioned vector normalization. PatternSample rating: time-since-deccelerating , distance , heading , the new distance-to-wall scheme & distance-to-corner.
Result:2048.23

2004-10-26: version 2.53.13
Based v2.53.12
Gun:
Crazy experiment vith the vectors orientation (normalizing not now in 0dgs, but in angle_to_me).
Result:2049.41

2004-10-25: version 2.53.12
Based v2.53.11
Gun: PatternSample rating: time-since-deccelerating , distance , heading & a new distance-to-wall scheme.
Result:2052.74

2004-10-25: version 2.53.11
Based v2.53.10
Gun: PatternSample rating: time-since-deccelerating , distance & heading.
Result:2049.92

2004-10-24: version 2.53.10
Based v2.53.09
Gun: PatternSample rating: time-since-deccelerating , distance & distance-to-wall.
Result:2044.36

2004-10-23: version 2.53.09
Based v2.53.08
Gun: PatternSample rating: time-since-deccelerating & distance .
Result:2046.78

2004-10-23: version 2.53.08
Based v2.53.07
Gun: Trying PEZ´s sugestion of time-sinvce-vel-changed (credits to him)... included time-since-vel-changed in PatternSample rating (took out time-since-inverted, time-since-accelerating, time-since-deccelerating).
Result:2040.97

2004-10-22: version 2.53.07
Based v2.53.06
Gun: included time-since-inverted in PatternSample rating.
Result:2040.75

2004-10-21: version 2.53.06
Based v2.53.05
Gun: re-include the time-since-accelerating in PatternSample rating.
Result:2042.40

2004-10-20: version 2.53.05
Based v2.53.04
Gun: re-include the time-since-deccelerating in PatternSample rating.
Result:2042.84

2004-10-19: version 2.53.04
Based v2.53.03
Gun: Trying a more "dense" vector search in PM:
now:{1, 2, 3, 4, 5,6,7, 10, 15, 20, 30, 40};
before:{1, 2, 3, 5, 10, 20, 40};
Result:2041.97

2004-10-19: version 2.53.03
Based v2.53.02
Gun: Trying:

double maxPwr = (BulletTracker.getHitAvg(target.getDistance())>0.35 )?3.0:1.9;
bulletPower = (distance<150)?3.0:maxPwr;

Result:2041.39

2004-10-19: version 2.53.02
Based v2.53.01
Gun: I just can´t resist to make things more complex, don´t I? Now trying:

double maxPwr = (BulletTracker.getHitAvg(target.getDistance())>0.25 )?3.0:1.9;
bulletPower = (distance<150)?3.0:maxPwr;

Result:2038.53

2004-10-18: version 2.53.01
Based v2.53
Gun: Trying PEZ (credits to him, obvious) sugestion for bullet power control: bulletPower = (distance<150)?3.0:1.9;.
Result:2039.78 (Thanks, PEZ!)

2004-10-17: version 2.53
Based v2.52
Gun: Total re-construction. Starting with vectorial PM + MultipleChoice(100 samps).
Result:2031.09

2004-10-17: version 2.52
Based v2.51
Gun: Added my last vector movement to rating.
Result:2030.94

2004-10-16: version 2.51
Based v2.50
Gun: MultipleChoice with 100 samples (50, before). Rating incremental (rated & aged).
Result:2034.98

2004-10-14: version 2.50
Based v2.48
Faster learning experiment.
Result:2039.12

2004-10-13: version 2.49
Based v2.48
Single Wave Surfing.
Result:2033.42

2004-10-12: version 2.48
Based v2.47
removed rating PatternSample by fireTime.
tweaked all-waves rating formula.WaveFired generating recalculation of Course - only if arrivalDelta <= 7.
Result:2036.72

2004-10-11: version 2.47
Based v2.46.4
Gunnery test: including fireTime as info for PatternSample, rating by it also.
Result:2036.29

2004-10-11: version 2.46.4
Based v2.46.3
WaveFired generating recalculation of Course - only if arrivalDelta <= 10.
Result:2035.38

2004-10-11: version 2.46.3
Based v2.46.2
Took out the "waves danger zoning" tweaking for HOA (back to same as 2.46.1).
WaveFired generating recalculation of Course.
Result:2024.90

2004-10-10: version 2.46.2
Based v2.46.1
Lil "waves danger zoning" tweaking for HOA (same as 2.46). Protection against SecurityExceptionBug, a pull-the-pin-and-swallow-the-granade strategy, idea inspirated in Jonathan´s similar idea (credits to him). Special saving (more precise & smaller, saving only allHits) for HOTs.
Result:2038.23

2004-10-10: version 2.46.1
Based v2.46
Back with random +-0.5 to chosen gf. Waves danger zoning tweaked.
Result:2041.83

2004-10-09: version 2.42RLD
v2.42 reloaded.
Version pos-BigCrash.
Result:2034.17

2004-10-08: version 2.46
Based v2.42.
Version pos-BigCrash. LOTS of new stuff...
Result:2017.94

2004-09-22: version 2.42
Keeping 2.35 cfg.
New "Ramming Detection" algo: if((attackAngle < 15) && (distance < 450) && (Math.abs(vel)==8){ ramming=true; }else{ ramming=false; }
Initializing direction to avoid walls at begin of rounds.
Simplified chooseFuture() method
Fixed some more nasty & stupid bugs (discovered with WaveSurfingChallenge):
AxeVector class giving crazy angles in setModule(0) cases. Protected now, if setModule(0), module==0.00001.
No anti-mirror if enemy is HOT (also avg value for tetection changed from 30 to 20).
Result:2042.98

2004-09-18: version 2.41
Keeping 2.35 cfg.
Fixed nasty & stupid bug #1 (discovered with WaveSurfingChallenge):
My "Ramming Detection" algo was: if((attackAngle < 25) && (distance < 450)){ ramming = true; }else...
This lead into situations that SS was anti-ramming when it souldn´t (particularly at WaveSurfingChallenge, because of the stopped opponent, SS sometimes was diving into situations that the conditions for detecting rammers where attended).
New "Ramming Detection" algo (condensed): if((attackAngle < 25) && (distance < 450) && (Math.abs(vel)){ ramming=(rammCount++ > 10)? true?false; }else...
Result:2037.79

2004-07-12: version 2.35.4
Keeping 2.35 cfg. Movement route prefered angle back to 90dgs. Added time-since-v0 to rating samples at PM gun.
Result:2027.07

2004-07-12: version 2.35.3
Keeping 2.35 cfg. Back with AntiMirrorSystem . Movement route prefered angle set to 110dgs.
Result:2028.27

2004-07-11: version 2.35.1
Keeping 2.35 cfg. Took out AntiMirrorSystem. "aware 100% scoring" only in the last 2 rounds.
Result:2035.31

2004-07-11: version 2.35
Back to 2.33.3 cfg.
Result:2039.10

2004-07-04: version 2.40
Keeping 2.33.3 cfg . Great change in routing movement, using the BotSimulator to choose the route, in order to be more precise and maximize escape envelope. Expected to be slow and somewhat buggy yet...
Result:2028.21

2004-06-27: version 2.34.2
Keeping 2.33.3 cfg . Trying a faster learning segmentation.
Result:2033.66

2004-06-20: version 2.33.3.3
Keeping 2.33.3 cfg . Fixing bug in 2.33.3.2.
Result:2033.94

2004-06-19: version 2.33.3.2
Keeping 2.33.3 cfg . Removed wall segment. HO detection function. HOA particularization, if HeadOn enemy, movement intent to be more precise and shorter escape envelope.
Result:2031.14 ** BUGGY **

2004-06-19: version 2.33.3.1
Keeping 2.33.3 cfg (think it´s broken, but it´s only for comparisson). Added a single simple distance-to-wall segment for moving (0-100,100-).
Result:2029.78

2004-06-18: version 2.33.3
Keeping 2.33.1 cfg. Back to SingleWaveSurfing. Trying a "real" bot simulator, that gives(i think...) me accurate navigation even when smoothing hard.
Result:2031.70

2004-06-15: version 2.33.2
Keeping 2.33.1 cfg. Trying AllWavesSurfing.
Result:2024.89

2004-06-15: version 2.33.1
Keeping 2.33 cfg. Correction of bug in 2.33v: "attempt to get a more wide escapvelovelope". using bot dim 36 to calculate consecutives. Only fast-learning see-poe-populated with 2 shots.Artificial wave shot when opp. dies - attempt to avoid "last breath" bullets. Moving too close try to escape if he is closer than u to the wall.
Result:2033.10

2004-06-13: version 2.33
Keeping 2.31 cfg. Attempt to "get more wide" the escape envelope by choosing a newwave sooner if vel=0. using bot dim 50 to calculate consecutives. Only fast-learning seg pre-populated with 2 shots.
Result:2029.29 ** BUGGY **

2004-06-13: version 2.32
Keeping 2.31 cfg. HOA effort again, now also segmented populated, added also a "aware 100% scoring" RRServer ignores 0 points results (thnx PEZ again). using bot dim 50 to calculate consecutives.
Result:2031.94

2004-06-12: version 2.31
Keeping 2.27 cfg. HOA effort (thanks PEZ words), in the process detected and fixed(?) a major bug in movement (pray that i had not introduced new ones). Pre-popolulating gf0 with 2 shots.
Result:2035.60 ** Best score ever **

2004-06-08: version 2.30
Keeping 2.27 cfg. Radical experiment in MultipleChoice gun´s feature. Added also attackAngle to PatternSample rating.
Result:2029.84

2004-06-08: version 2.27.5
Keeping 2.27 cfg. Anti-ramm tweak again (hopefully without bug of 2.27.4.* series).
Result:** RECALLED, UPLOADED WRONG VERSION!! **

2004-06-08: version 2.27.4.2
Keeping 2.27 cfg. Anti-ramm re-re-tweaking, actually took off the anti-ramm tweaking, to confirm my suspicion that some bug was introduced after 2.27.3 (there is something very rotten in the state of denmark).
Result:2022.23

2004-06-08: version 2.27.4.1
Keeping 2.27 cfg. Anti-ramm re-tweaking.
Result:2020.13

2004-06-07: version 2.27.4
Keeping 2.27 cfg. Anti-ramm tweaking.
Result:2022.74

2004-06-07: version 2.27.3
Keeping 2.27 cfg. Gun tweaking (added time_since_accelerating & time_since_deccelerating to rating function).
Result:+99.97(2028.74)

2004-05-30: version 2.29
Keeping 2.27 cfg. Another (hopeless?) experiment, playing around with start moving time.
Result:+74.55(2002.69) *trashy... recalled

2004-05-25: version 2.28
Keeping 2.27 cfg. Playing around with MaxVelocity.
Result:+77.71(2005.51) *stronger against PMs, but much weaker against the others....

2004-05-24: version 2.27
2.21 back, i must be missing something... faster reading data; compatible with JRE 1.3.1.
Result:+90.93(2018.73)

2004-05-24: version 2.26
Keeping 2.21 cfg. back to old firePwr; Lots of gf measuring bug correction (hope didn´t trashed anything); Gun firing bug corrected (starting game); Forcing fire if rammed;
Result:+69.70(1998.54)

2004-05-23: version 2.21.3
Keeping 2.21 cfg. keeping FirePwr increased; back to RollingAverages; back to consecutives = 36; faster reading data; compatible with JRE 1.3.1.
Result:+78.79(2006.42)

2004-05-22: version 2.21.2
Keeping 2.21 cfg. FirePwr increased; fastAverage n =4; consecutives set to 51 (instead of 36).
Result:+72.32(2000.25)

2004-05-20: version 2.21.1
Keeping 2.21 cfg. A variant of 2.21, trying a alternative formula for RollingAverage (PEZ sugestion) for more fast adaptation.
Result:+81.79(2010.93)

2004-05-19: version 2.25
Keeping 2.21 cfg. More bugs: Route orientation when vector´s module == 0; close-range with wave null now go 120 degrees; other minor bugs.
Result:+72.40(2000.48)

2004-05-18: version 2.23
** BUGGY ** Keeping 2.21 cfg. Two bugs (at least that i found) from 2.22: Anti-Mirror buggy, bulletInterception adding -12 (break) - Corrected both.
Result: * RECALLED

2004-05-17: version 2.22
** BUGGY ** Keeping 2.21 cfg. Lots of bugs/malfunctions corrections: revised getGF() methods, some buggy; wallSmoothing function improved & optimized .Cleaned some code.
Result:+63.31(1992.33)

2004-05-12: version 2.21
Manteining 2.20.1 cfg. Navigation more precise.
Result:+89.19(2019.62)

2004-05-11: version 2.20.1
Manteining 2.17.2 cfg. Smoothing factor set to 1.
Result:+91.09(2017.56) *First version ever to reach 2nd in general 1X1

2004-05-10: version 2.20
Manteining 2.17.2 cfg. Smoothing factor set to 2.
Result:+75.85(2003.05)

2004-05-07: version 2.17.5
Manteining 2.17.2 cfg. Variant of 2.17 rating form ((1*segmented_segment) + (1*unsegmented_segment)+(1*flattener)).
Result:+82.05(2009.56)

2004-05-06: version 2.17.4
Manteining 2.17.2 cfg. Variant of 2.17 rating form ((1*segmented_segment) + (1*unsegmented_segment)+(0.5*flattener)).
Result:+88.62(2015.59)

2004-05-05: version 2.19
Manteining 2.17.2 cfg. Trying an approach based on PEZ´s idea of "mirroring" the edges for edge protection under smoothing.
Result:+74.79(2002.24)

2004-05-03: version 2.18.1
2.18 variant. Lightweighted cannon 10000 samples max.
Result:+81.76(2009.67)

2004-05-03: version 2.17.3
as PEZ requested: Variant of 2.17 rating form ((0*segmented_segment) + (1*unsegmented_segment)).
Result:+49.85(1977.76)

2004-05-03: version 2.18
Manteining 2.17.2 cfg. Lightweighted cannon 20000 samples => 5000 samples max.
Result:+82.20(2011.41)

2004-05-02: version 2.17.2
Variant of 2.17 rating form ((1*segmented_segment) + (1*unsegmented_segment)).
Result:+84.99(2013.56)

2004-05-02: version 2.17.1
Variant of 2.17 rating form ((0.5*segmented_segment) + (1*unsegmented_segment)).
Result:+76.26(2004.95)

2004-05-02: version 2.17.
Removed bullet power segments (actually set to 1 seg). Trying rating form %2mentmented_segment + (0.5*unsegmented_segment)) instead of (segmented_segment + unsegmented_segment+(0.1*visits)).
Result:+82.17(2010.80)

2004-05-02: version 2.16.
Adding bullet power segment (3:0-1;1-2;2-3).
Result:+83.89(2012.41)

2004-05-02: version 2.15.
5 absolute velocities instead of 3 (now:0;1-2;3-4;5-6;7-8).
Result:+84.37(2013.79)

2004-05-01: version 2.14.
Trying a FastAdaptation solution, that is a merge of ABC´s across-segments smoothing solution and PEZ´s "unsegmented segment" approach (credits for them both).
2.07 config was not so "the same" in 2.12 & 2.13, corrected the min dist. parameter (100 => 150).
Corrected a bug in the loading, was using readByte that could give me negative values for things that was to be positive, replaced with readUnsignedByte.
Result:+82.82 (2012.08)

2004-04-28: version 2.13.
2.07 config. Saving data but no pre-saved.
Result:+59.45

2004-04-27: version 2.12.
Back to 2.07 config. Trying a version without saving data (Let´s see if ABC´s observation works here).
Result:+57.16

2004-04-25: version 2.11.
Took out DinamicDistancing. Trying an experimental wave weightening (no surprises if it degrade the rating).
obs: due to size restrictions, this version will not include source.source in 2.10r> Result:+67.70

2004-04-24: version 2.09.
Trying for the first time dinamic distancing, maybe buggy...
obs: due to size restrictions, this version will not include source.source in 2.08
Result:+64.42

2004-04-23: version 2.07.
Seems that the "tweaked" anti-ramm pull me down somewhat like 8 pts. Don´t really know the problem, just giving a blind shot.
obs: due to size restrictions, this version will not include source.source in 2.06
Result:+81.02

2004-04-22: version 2.05.
removed the "Lightened" gun back to slow and heavy (perhaps in the future, i´ll work more on it).
obs: due to size restrictions, this version will not include source.source in 2.04
Result:+67.09

2004-04-21: version 2.03.
Tweaked the anti-ram. "Lightened" the gun (10000 samples, grain 30, 10 samps in multiple choice).
obs: due to size restrictions, this version will not include source.source in 2.02
Result:+65.04

2004-04-19: version 2.01.
Tweaked a little the weight for hits rollingAvgs.
obs: due to size restrictions, this version will not include source.source in 2.00
Result:+71.37

2004-04-16: version 1.22.
AM data: Introduced velocity and acceleration segmentation. segmentation now: distance - 3, acc - 3, vel - 3.
obs: due to size restrictions, this version will not include source.source in 1.21
Result:+70.06 (first version breaking the 2K barrier)

2004-04-13: version 1.20.
Bug in the wave identification method.
obs: due to size restrictions, this version will not include source.source in 1.19
Result:+53.51

2004-04-11: version 1.18.
More realistic (strongest) consecutive gf setting to compensate bot size. Improved bullet detection (bulletHitBullet).Included a lot of pre saved data.
obs: due to size restrictions, this version will not include source.source in 1.17
Result:+51.70

2004-04-09: version 1.16.
More bug corrections. Included consecutive gf setting to compensate bot size.
obs: due to size restrictions, this version will not include source.source in 1.15
Result:+47.18 !!

2004-04-06: version 1.14.
A lot of bug corrections.
Result:+26.59

2004-04-03: version 1.13.
Movement based on triangle, wall smooth. Bug correction on rating (now counting hits correctly). Tweaked rating a little
obs: due to size restrictions, this version will not include source.source in 1.12
Result:+25.53

2004-03-27: version 1.11.
Trying a move-to rating inspired on Jamougha´s gf rating.
Result:+17.02

2004-03-24: version 1.10.
More deep inside the restrictions age. Tweaked with PEZ´s TGB.
Result:-12.25

2004-03-22: version 1.09.
Roots, man, roots. Attempt to at least get same score as Okami.
obs: due to size restrictions, this version will not include source.
Result:-2.60

2004-03-16: version 1.07.
Complete restructuration of AM AI (again!). Measuring GF by angle ok.
Result:-33.00

2004-03-16: version 1.06.
Complete restructuration of AM AI. New measuring GF by angle.
Result:-27.55

2004-03-14: version 1.05.
Fixed bug from last version in AM AI. Movement angle & distancing tweak. Mainteining 1.02 alteration.
Result:-11.89

2004-03-14: version 1.03.
Fixed bug in last visiteds filter. First priority set to last hit GF. Mainteining 1.02 alteration.
Result: -18.60

2004-03-13: version 1.02.
using Math.asin(MAX_VEL/bulletVel) instead of MAX_VEL/bulletVel for calculation of the max escape angle.
Result:+4.75pts.

2004-03-11: version 1.01.
Trying pre-saved data (.gfs).
obs: due to size restrictions, this version will not include source.
Result:-4.54pts.

2004-03-09: version 1.00.
Bullet Power increased a lot.
Result: -5.82pts.