Pugilist/Version History

From Robowiki
Jump to navigation Jump to search
Pugilist Sub-pages:
PugilistVersion History - Help Requests - Guess Factor Chat - PugilistGL - PugilistGL/Code - Archived Talk 20040501 - Archived Talk 20090529

Newest events first:

  • Jan 29 2006 - Version 2.3 released - ? points
    • Movement - Old Wave handling restored. No "artificial" wall-bouncing (yeah, I've tried that like 1000 times, but now it's time again).
    • 82 bytes free
  • Jan 28 2006 - Version 2.2d released - 1978 points
    • Movement - Another test with simplified wave handling
    • 60 bytes free
  • Jan 27 2006 - Version 2.2c released - 1988 points
    • Movement - Old Wave handling restored, still no stop-position evaluation
    • 50 bytes free
  • Jan 27 2006 - Version 2.2b released - 1978 points
    • Movement - No stop-position evaluation and a lot of simplifications in Wave handling
    • 82 bytes free
  • Jan 26 2006 - Version 2.2 released - 1991 points
    • Movement - Now evaluates stop-position in surfing. Wave danger not vanishing as fast with distance from Pugilist. (If these changes don't give me a rating boost I'll scrap them and use the saved 70 bytes I just found for something else.)
  • Nov 24 2004 - Version 2.1.2 released - 1987 points and 100% wins in the PremierLeague
    • Gun - Waiting for alignment before fire.
  • Sep 2 2004 - Version 2.0.9 released - 1997 points
    • Movement - Updating stats for passing waves, which should provide better performance. Though there's that strange bug that I haven't found yet...
  • Sep 1 2004 - Version 2.0.8b released - 1967 points (Really, really strange. I think P has a big bug somewhere that shows itself very seldomly and this version had more bad luck than most of the others.)
    • Movement - Fixed bug generating NaN danger values for passing waves.
  • Aug 4 2004 - Version 2.0.8 released - 2004 points
    • Movement - Higher initial values when updating the hit stats.
  • June 15 2004 - Version 2.0.2 released - 1994 points
    • Tweaked evasion angle and (enemy)wave impact distance
  • June 14 2004 - Version 2.0.1 released - 1989 points
    • More precise future position prediction
    • Less special cases in the WallSmoothing (now down to one special case)
  • May 26 2004 - Version 1.9.9.4.3 released - ? points
    • Removed the last gun change since I obviously couldn't afford the effects of the new movement wave management.
    • 45 bins for movement and gun waves
    • Goes easier on the gun wall segmentation
    • More precise factor-to-bin-and-back translation for the gun waves.
  • May 26 2004 - Version 1.9.9.4.2 released - 1963 points
    • Slightly different EnemyWave handling to save some 20 bytes for:
    • Gun tweak from Kawigi's suggestion. (Let's hope the enemy wave management change doesn't bite me...)
  • May 23 2004 - Version 1.9.9.4.1 released - 2008 points
    • 1.9.9.4 movement and 1.9.9.8.1 gun tweak (Same gun as used in SilverFist and .p versions of Shadow and RaikoMX.)
  • (Deferred) - Version 2.0 released (will be) - ? points
    • Finally a true WaveSurfer. No wall bouncing ever, ever. Zero bytes left though...
  • May 23 2004 - Version 1.9.9.8.2 released - 1977 points
    • Back to gun wall segmentof 1.9.9.4 (checking if my "improved" wall smoothing algo might be losing me points.)
  • May 22 2004 - Version 1.9.9.8.1 released - 1986 points
    • Gun wall segment tweak in between 1.9.9.4 and 1.9.9.7
    • Wall smoothing change. No more wall avoidance problems I hope.
  • May 22 2004 - Version 1.9.9.7 released - 1997 points (Looks pretty funny, doesn't it?)
    • Gun wall segmentation tweak.
  • May 21 2004 - Version 1.9.9.4 released - 2001 points (At least after 699 battles, I missed what it was at 500)
    • Gun segmentation tweaks.
  • May 21 2004 - Version 1.9.9.1 released - 2004 points !!!!!!!!!!!!!! Did it. Did it. Did it. YAHHHHHHHHHHHHH!!!!!
  • May 21 2004 - Version 1.9.9 released - 1985 points
    • Trying without the targeting bug om 1.9.7
  • May 20 2004 - Version 1.9.7b.6 released - 1990 points
    • Reinstalled the bug of 1.9.7 and tweaked it.
  • May 19 2004 - Version 1.9.7b released - 1986 points
    • Targeting bug fix of 1.9.7.
  • May 18 2004 - Version 1.9.7.6 released - 1976 points
    • Should be functionally equivalent to 1.9.7. Only slightly refactored to save me 19 bytes.
    • Like 1.9.7 but with a combined last_velocity and wall segment dimension.
  • May 18 2004 - Version 1.9.7.5 released - 1969 points (Either I introduced a but in the segmentation or I just can't afford the slower learning...)
    • Like 1.9.7 but with a combined last_velocity and wall segment dimension.
  • May 17 2004 - Version 1.9.7.1 released - 1989 points
    • A little steeper in evasion
    • Fires power 1.8 all of the time
  • May 17 2004 - Version 1.9.8 released - 1975 points
    • Lots of tweaks and changes everywhere... I might revisit them later since some of them probably was better than others.
  • May 17 2004 - Version 1.9.7 released - 1992 points (Strongest yet, and it seems to have taken a firm hold on the minibot ranking #1 slot.)
    • Recalled surfing tuning of 1.9.6
    • Tuned wall smoothing
    • Back to using calculated max_escape_angle
  • May 16 2004 - Version 1.9.6 released - 1968 points
    • Tuned surfing reverse and forward prediction
    • Back to 1.9.4.3b segmentation
  • May 16 2004 - Version 1.9.4.5b released - 1972 points
    • Fixed bug with fast learning weight. Obviously some bugs should be left in. =)
  • May 16 2004 - Version 1.9.4.5 released - 1979 points
    • Segmenting on both last_velocity and wall (and distance and velocity too of course).
    • Has a bug where fast learning (unsegmented) stats atre weighted too high.
  • May 15 2004 - Version 1.9.5 released - 1980 points
    • Applied some simple edge protection. Not a good idea it seems.
  • May 15 2004 - Version 1.9.4.4 released - 1987 points
    • Replaced last_velocity segment with wall segment
    • Not caluclating max_escape_angle, just using 1.0. (Saves bytes, and shouldn't make a difference anyway.)
  • May 15 2004 - Version 1.9.4.3b released - 1990 points
    • Replaced movement wall segmentation with last_last_velocity and last_velocity
    • Weight fast-learning stats by 1/num_segment_combinations like ABC has suggested.
  • May 15 2004 - Version 1.9.4.2 released - 1986 points
    • Back to good ol' gun segmentation
    • Shortened blind-mans-stick to 0.3 radians
  • May 14 2004 - Version 1.9.4 released - 1963 points
    • Even more distance segment bins
    • Fixed surfing bug
  • May 14 2004 - Version 1.9.3.4 released - 1966 points
    • One more distance segmentation bin and fewer of most other segmentation bins. Meant to speed up learning.
  • May 14 2004 - Version 1.9.3.3 released - 1977 points
    • Wall-bouncing at close ranges. Suggested by Axe.
  • May 13 2004 - Version 1.9.3.2 released - 1958 points
    • Using a rough estimate of the length of my WallSmoothing blind-mans-stick instead of a calculated one. Saves bytes and makes for a less stubborn behaviour when fighting at point blank range.
  • May 12 2004 - Version 1.9.3.1 released - 1923 points
    • Yet another try with pure surfing. (That is, no wall bouncing and such).
  • May 11 2004 - Version 1.6.4.30b released - 1982 points
    • Small gun tweak from 1.6.4.4
  • May 7 2004 - Version 1.6.4.17 released - 1940 points
    • 37 enemy wave stat bins
    • Smoothing with Math.pow(distance, 1.5)
    • Edge protection with Axe's idea of an inverted smoothed flat curve. (Adaption of ABC's implementation as posted on the SilverSurfer page.)

  • May 7 2004 - Version 1.6.4.16 released - ? points
    • 37 enemy wave stat bins
    • Smoothing with Math.pow(distance, 1.5) (like SilverSurfer, though with much fewer bins)
  • May 7 2004 - Version 1.6.4.15 released - 1929 points
    • 17 enemy wave stat bins
    • Gravitational smoothing
    • (Yeah, like Shadow)

  • May 7 2004 - Version 1.6.4.14 released - 1959 points
    • 15 enemy wave stat bins
    • No smoothing
  • May 7 2004 - Version 1.6.4.13 released - 1955 points
    • 9 enemy wave stat bins
    • No smoothing
  • May 6 2004 - Version 1.6.4.12b released - 1972 points
    • 47 enemy wave stat bins
  • May 6 2004 - Version 1.6.4.11 released - 1972 points
    • No smoothing on gun stats
    • Back to 27 stat bins
  • May 6 2004 - Version 1.6.4.10 released - 1973 points
    • Same as 1.6.4.9 but with:
      • Edge smoothing protection
      • Smoothed gun stats
      • 47 stat bins (instead of 27) for both gun and movement
  • May 5 2004 - Version 1.9 released - 1914 points
    • Movement:
      • Linear bin smoothing
      • Edge smoothing protection
      • Iterated EnemyWave impact points
      • Absolutely no wall bouncing! Pure WaveSurfing this time.
    • Gun:
      • Bin smoothing (with edge protection)
  • May 5 2004 - Version 1.6.4.9 released - 1973 points
    • Like 1.6.4.4 with linear distance smoothing. (Saves some bytes!)
  • May 5 2004 - Version 1.6.4.8 released - 1967 points
    • Like 1.6.4.4 with gravitational smoothing.
  • May 4 2004 - Version 1.7.7 released - 1951 points
    • Movemenet: Fast learning stats again, with low weights. And gravitational bin smoothing by suggestions from ABC
  • May 4 2004 - Version 1.6.4.7 released - 1952 points
    • Now weighting the stats based on enemy wave distance. Still no smoothing.
  • May 4 2004 - Version 1.6.4.6 released - 1944 points (or there about, I recalled this version before it got 500 battles under its belt).
    • Trying without smoothing the EnemyWave stats (not even weighting it for distance).
  • May 4 2004 - Version 1.7.6 released - 1947 points
    • 1.6.4.4 With fast-learning applied. (Or, I tried to apply it. I think I might be doing it the wrong way.)
  • May 3 2004 - Version 1.6.4.5b released - 1975 points
    • 1.6.4.4 with saved gun-data on some 30 opponents (couldn't fit more since the gun is heavily segmented). Obviously this data didn't make any difference at all.
  • May 2 2004 - Version 1.6.4.4 released - 1975 points
    • 1.6.4 with 39 bytes free and some really minor tunings.
  • May 1 2004 - Version 1.6.4.2 released - 1951 points
    • Lower granularity in the time-since-deccel and wall gun segments.
  • May 1 2004 - Version 1.6.4.1 released - 1954 points
    • Back to the 1.6.4 track. Now with a gun tuning added.
  • April 30 2004 - Version 1.8.1 released - 1958 points
    • Trying to predict the EnemyWave impact locations via iteration rather than the crude estimate I had before. Should make things much more accurate near walls. Though I don't know how to account for decceleration time this way yet...
  • April 30 2004 - Version 1.6.7 released - 1973 points
    • This is the same as 1.6.4 except I try to protect the edge smoothing a tad more.
  • April 29 2004 - Version 1.7.2 released - 1962 points
    • Since I'm desperate I decided to try with smoothed gun GFs as well. It only costs me 1 byte anyway so any rating points gained will be quite cheap. (Even if it is very unlikely I'll gain any points at all.)
  • April 29 2004 - Version 1.7.1 released - 1967 points
    • With both fast-learning mode and edge-smoothing protection activated.
    • Strong but doesn't collect as much points as 1.6.4. Now it's something like a cross road for me where to take P...
  • April 29 2004 - Version 1.6.6 released - 1963 points
    • Back to using only lightly segmented EnemyWaves
    • Trying without dealing with the edge smoothing problem. (If this doesn't give less point I'd say there is no "edge smoothing problem".)
      • -10 points. I guess there really exist such a beast as the "edge-smoothing problem". Since I removed edge-smoothing protection in 1.7 this probably means that 1.7.1 (with fast-learning mode and edge-smoothing protection) should give me 1980+ points. I just must find 10 or so bytes somewhere...
  • April 28 2004 - Version 1.7 released - 1969 points
    • Now uses unsegmented EnemyWaves when there's no data in the segmented one.
    • Starved on bytes agains. Only 2 left!
  • April 28 2004 - Version 1.6.5 released - 1973 points
    • Gun tuning (narrowed down the GF +/- 1 boundaries).
  • April 28 2004 - Version 1.6.4 released - 1976 points
    • Back to 1.5.6.2 EnemyWave segmentations (distance and fire power)
    • Gun wall segment tweak.
  • April 27 2004 - Version 1.6.3 released - 1959 points
    • EnemyWave segmentation on distance, bullet power, velocity. Rough segments with few buckets.
    • Doesn't win or lose one single rating point compared to unsegmented waves...
  • April 27 2004 - Version 1.6.2 released - 1959 points
    • Trashed the new gun segmentation. Makes me lose bigger against iiley's bots. But with that segmentation I obviously lost some against the rest making the end sum -4 or so points.
    • Now trying totally unsegmented EnemyWaves.
  • April 26 2004 - Version 1.6 released - 1959 points
    • Threw in yet another segmentation for the gun. One which I don't think many WaveSurfers are using. Now I have exactly zero bytes left in this bot. Let's hope this segmentation doesn't work. =)
  • April 26 2004 - Version 1.5.6.2 released - 1963 points
    • Gun tweaks
  • April 22 2004 - Version 1.5.6 released - 1932 points
    • Removed wall segmentation from the EnemyWaves
    • Increased the granularity of wall segmentation in the gun Waves.
  • April 22 2004 - Version 1.5.4.1 released - 1930 points
    • Fixed bug with array index out of bounds at extreme distances. Thanks to Axe for noticing that something was this wrong!
  • April 21 2004 - Version 1.5.4 released - 1926 points
    • A minor code restructure. Less code. Maybe less bugs...
  • Mars 23 2004 - Version 1.5.3 released - 1918 points
    • Another tuning.
  • Mars 23 2004 - Version 1.5.2 released - 1855 points - recalled
    • Tuning, tuning.
  • Mars 21 2004 - Version 1.5.1 released - 1925 points
    • Evades a bit steeper at close ranges.
  • Mars 21 2004 - Version 1.5 released - 1923 points
    • Just some minor tweaks.
    • Should have started versioning 1.5.x with 1.4.5 actually.
  • Mars 19 2004 - Version 1.4.6 released - 1921 points
    • Yayalt. What happens if I switch off the flattener?
  • Mars 19 2004 - Version 1.4.5 released - 1908 points
    • Back to sloppy prediction.
    • Fixed a bug with measuring delta bearing from the wrong origin.
  • Mars 17 2004 - Version 1.4.4 released - 1890 points
    • Back to 1.4.2 tuning.
    • I bit less steep in evasion and instead generally backs away a little faster.
  • Mars 17 2004 - Version 1.4.3 released - 1846 points
    • Trying to make fewer untested assumptions...
  • Mars 16 2004 - Version 1.4.2 released - 1874 points
    • Tuned reverse and forward sensor placement.
  • Mars 16 2004 - Version 1.4.1 released - 1893 points
    • Same as 1.4 but without the flattener
    • Fixed a bug with using the wrong distance measure for movement.
  • Mars 16 2004 - Version 1.4 released - 1870 points
    • Fixed a bug with using the wrong distance measure for movement.
    • Switched on the flattener again
    • Tweaked bin smoothing and distance to the reverse / forward sensors.
  • Mars 13 2004 - Version 1.3.6 released - 1893 points
    • I don't expect it to work. It's YAYALT
    • Purely adaptive movement. (Removed the flattener.) This is beacause Jamougha told me to try get that to work first. Probably better against weak bots but worse against strong this way.
    • No future position prediction. Now I just look ahead some 50 pixels and look backwards 2 pixels and see where I have least impact from a previous hit. Massive bin smoothing as per suggestion from rozu.
  • Mars 11 2004 - Version 1.3.3 released - 1882 points
    • Back to bin smoothing. But now linearly instead of using the square root of the distances.
    • Some strategy tweaks too.
  • Mars 11 2004 - Version 1.3.2 released - 1877 points
    • Less of a GF1 spike by not bin smoothing the EnemyWaves. Doesn't perform on par with 1.3.1 though...
  • Mars 10 2004 - Version 1.3.1 released - 1904 points
    • OK. Big GF1 spike but I want to test the surfing board anyway. Surfing every wave now again.
  • Mars 10 2004 - Version 1.3.1 near release
    • RobocodeGLV014 has helped me big time to nail down seven (7!) bugs in my WaveSurfing.
    • Struggling witha GF1 spike at the moment. But I might release it with this flaw. This gives me some room for improvement if nothing else. =)
  • Mars 5 2004 - Version 1.3 released - 1651 points
  • Mars 4 2004 - Version 1.3 under development
    • Now trying with a Shadow style WaveCutting instead. The problem with overusing the positive GFs is much worse. I must really do something about my movement prediction technique. But I'm a bit out of ideas on how to do this any good with the CodeSize limit... I also have some problems to sort out with my CustomEvent driven EnemyWaves. See EventManagement for more info this...
  • Mars 4 2004 - Version 1.2 released
    • Fixed major bug when I was surfing an arbritratuy bullet-wave instead of the closest. Still overuse positive GFs a lot. I think this has to do with my forward prediction not being accurate enough.
  • Mars 3 2004 - Version 1.11b released
    • Some bug fixes and tweaks
  • Mars 3 2004 - Version 1.1 released - 1744 points
    • Now featuring WaveSurfing (even if it's not very good yet...)