User:Chase-san/NewTech

From Robowiki
Jump to navigation Jump to search

Log page for my personal use in keeping track of scores for my new robot base, just finished my first test run.

Targeting Experiment (Mint Clover)

Mint Clover is at the moment a simple DC gun. But it has been awhile since I actually wrote a gun. I think I may have to rediscover some techniques and practices. In case you are wondering, my goal is to write a single (DC/GF) gun that is good at both random movement and anti-surfer strategy. While it may not be the best in both area. I don't like the extra complexity that a multi-gun system entails.

TCRM

Version Aspd Sprw Fhqw Yngw FlMn EASY Tron HTTC RnMB DlMc Grbb MED SnDT Cgrt Frtn WkOb RkMc HARD TOTAL Seasons
b0.9 91.67 95.42 95.26 97.26 92.94 94.51 87.21 88.49 87.11 87.39 81.84 86.41 73.84 82.87 79.26 81.65 81.19 79.76 86.89 40
b0.10 92.76 95.77 95.06 97.45 93.97 95.00 87.17 88.15 87.62 88.23 82.43 86.72 75.83 83.96 80.77 81.13 80.11 80.36 87.36 40
b0.11b 93.65 97.19 96.17 97.31 94.24 95.71 88.17 88.05 87.57 87.74 81.90 86.68 79.50 84.47 81.26 82.75 80.56 81.71 88.04 40
b0.12 94.96 96.90 96.20 97.26 93.44 95.75 87.90 87.05 85.94 87.80 83.50 86.44 79.74 85.01 80.99 82.84 79.53 81.62 87.94 40
b0.13 94.58 97.82 95.72 96.86 94.00 95.80 86.95 89.37 87.19 87.65 82.03 86.64 78.70 85.42 81.26 82.81 80.78 81.79 88.08 40
b0.14 94.71 97.62 95.41 97.35 93.24 95.66 88.18 88.54 87.81 88.69 82.38 87.12 79.63 84.91 82.06 81.30 81.26 81.83 88.21 40
b0.15 94.17 97.50 95.54 97.93 93.31 95.69 87.40 87.50 87.33 87.05 83.46 86.55 80.82 85.19 81.05 81.97 81.36 82.08 88.11 40
b0.19 93.82 97.70 96.08 96.79 93.82 95.64 87.24 87.98 87.03 88.58 83.78 86.92 81.05 85.27 80.92 82.23 80.87 82.07 88.21 104.1
Reference 91.26 97.96 97.28 98.86 99.56 96.98 86.12 87.06 91.10 85.93 82.95 86.63 77.03 82.91 81.61 85.73 80.01 81.46 88.36 40.0
Best Sane 95.93 99.18 98.75 98.86 95.96 97.74 91.55 89.98 94.11 89.80 90.90 91.27 91.33 86.82 82.97 88.48 84.61 86.84 91.95
Goal 88.5-90

TCAS

Version Sha Chk Cun Dia YP DC Phx DrGT CC Pear Dki VCS Total Seasons
b0.9 55.52 61.96 81.14 62.03 75.60 67.25 67.57 67.83 65.43 80.82 59.69 68.27 67.76 40
b0.10 56.26 60.29 81.82 59.84 76.07 66.86 69.10 65.16 63.75 80.53 57.54 67.22 67.04 40
b0.11b 58.45 61.19 82.68 61.82 75.79 67.99 67.84 66.10 66.00 81.76 57.34 67.81 67.90 40
b0.12 58.64 60.51 83.44 61.24 77.10 68.19 66.06 64.46 63.97 83.28 59.29 67.41 67.80 40
b0.13 58.45 61.70 82.34 61.62 75.51 67.92 67.94 65.25 65.96 82.10 58.83 68.02 67.97 40
b0.14 59.14 62.85 81.52 59.68 76.64 67.97 67.20 66.21 64.30 81.54 57.71 67.39 67.68 40
b0.15 60.77 63.51 82.96 60.19 76.27 68.74 67.81 64.99 66.16 80.69 57.18 67.37 68.05 40
b0.19 59.94 61.73 82.45 61.53 77.59 68.65 68.33 65.90 65.27 81.93 56.87 67.66 68.16 73.3
Reference 67.25 70.85 88.93 67.95 77.10 74.42 68.67 67.98 72.05 74.42 56.30 67.88 71.15 50.0
Best Sane 67.78 71.91 90.26 67.95 80.01 75.58 71.77 73.03 75.21 78.83 65.77 72.92 74.25
Goal 71.3-73

Movement Experiment (FlashStep)

Evolved version of Kakeru, packing every piece of experimental, theoretical and awesome scrap of code I can conceive. Doing everything you 'should' do in a top level surfer. I am considering even making a GT version. Bullet Shadows. Nth Wave Surfing. Dynamic Nth Wave Weighting. Nth back tick Heat Wave Surfing. Statistical Dynamic Distancing Control. Statistical KNN Neighbor consideration and discard. Enemy Movement Pattern Consideration in Surfing. Smart movement path selection.

Basically when (if) I finish, it will be one completely unique robot, and hopefully more then powerful enough to give DrussGTs movement a serious run for it's money. But half of it is experimental and the other half only theory.

Movement Experiment (Kakeru)

MC2K6 Fast (35 round)

Version WSCBotA WSCBotB WSCBotC WSC APMC CassiusClay FloodHT Shadow ADAPT Total Seasons Comment
MC54 99.69 88.88 86.27 91.62 30.53 31.60 53.82 26.70 37.38 53.17 60.0

MC2K7 Fast (35 rounds)

Version HOF SPL GRB Sub1 WAY (Sub2) GR3 RKM Sub3 ASC CC CHK Sub4 Total Seasons Comment
MC54 99.02 87.32 92.06 92.80 68.73 70.70 81.34 76.02 37.12 42.48 42.68 40.76 69.58 74.0
MC55 99.19 86.80 92.54 92.84 69.80 70.51 81.66 76.09 35.46 44.14 44.68 41.43 70.04 74.0
MC57 99.71 87.62 93.19 93.51 69.80 71.81 81.90 76.86 37.06 43.10 44.70 41.62 70.45 80.0

MC54

  • Modified the risk calculation to something else that could support bullet shadowing

MC55

  • Bullet Shadowing Experiment A

MC57

  • Awesome :)

Todo. Go back to 54 and tune to be better vs simple targeting again.


Kakeru Base

Code for Kakeru

Movement

MC2K6 Fast (35 round)

Version WSCBotA WSCBotB WSCBotC WSC APMC CassiusClay FloodHT Shadow ADAPT Total Seasons Comment
MC40 99.55 87.84 84.61 90.66 16.77 17.03 43.07 8.57 22.89 43.44 20.0
MC42 99.49 77.42 88.01 88.31 28.75 14.99 60.73 12.49 29.41 48.82 20.0
MC43 99.54 73.20 84.00 85.58 29.06 15.90 59.90 12.82 29.54 48.06 20.0
MC44 99.95 90.23 89.03 93.07 30.52 27.64 57.26 21.67 35.52 53.04 20.0
MC45 99.98 90.73 90.03 93.58 27.56 27.22 51.13 23.55 33.96 51.70 30.0
MC46 99.95 92.07 91.60 94.54 29.14 25.57 54.94 23.51 34.67 52.78 30.0
MC47 99.96 93.40 92.40 95.25 29.69 26.77 55.94 24.11 35.61 53.52 60.0
MC48 99.97 93.48 93.64 95.70 30.52 26.17 55.95 24.67 35.60 53.94 60.0
MC49 99.98 93.35 93.41 95.58 29.87 26.41 55.18 24.92 35.50 53.65 60.0

MC2K7 Fast (35 rounds)

Version HOF SPL GRB Sub1 WAY (Sub2) GR3 RKM Sub3 ASC CC CHK Sub4 Total Seasons Comment
MC40 99.25 83.38 83.61 88.75 62.54 75.10 76.41 75.75 27.32 36.25 33.24 32.27 64.83 20.0
MC42 99.68 86.47 86.67 90.94 65.35 74.73 77.36 76.05 30.70 37.11 33.15 33.65 66.50 20.0
MC43 99.72 86.29 85.40 90.47 65.23 75.32 78.50 76.91 30.56 36.94 36.38 34.63 66.81 20.0
MC44 99.78 87.12 90.95 92.62 66.53 71.77 82.01 76.89 34.52 41.12 38.62 38.09 68.53 20.0
MC45 99.77 86.97 91.83 92.85 68.03 71.90 82.04 76.97 35.13 44.04 42.60 40.59 69.61 30.0
MC46 99.71 87.22 91.75 92.89 68.54 71.71 82.06 76.89 33.46 41.21 40.98 38.55 69.22 30.0
MC47 99.73 87.77 92.43 93.31 68.05 72.25 82.09 77.17 34.56 42.03 42.47 39.68 69.56 80.0
MC48 99.69 87.65 92.31 93.22 68.15 72.11 82.44 77.28 35.58 42.39 43.66 40.54 69.80 60.0
MC49 99.74 87.71 93.51 93.65 69.08 72.26 81.65 76.95 36.57 40.71 42.83 40.04 69.93 80.0
MC57 99.71 87.62 93.19 93.51 69.80 71.81 81.90 76.86 37.06 43.10 44.70 41.62 70.45 80.0


The Fish Tests

Barracuda Challenge Head-on Targeting. Used for Dive Protection testing.

DevilFISH Challenge Linear Targeting. Fights Close. Used for close range combat testing.

Version Barracuda DevilFISH Total Comment
MC40 100.00 99.54 99.77 4.0 seasons
MC42 100.00 99.72 99.86 4.0 seasons
MC43 99.33 99.50 99.41 4.0 seasons
MC46 100.00 99.72 99.86 4.0 seasons

Same as above, but fast over only 35 rounds. A slightly more useful metric as it sorts out how well it does with learning to deal with these movements quickly, rather then over a period of time.

Version Barracuda DevilFISH Total Comment
MC40 100.00 91.22 95.61 20.0 seasons
MC42 100.00 95.69 97.85 20.0 seasons
MC43 100.00 91.42 95.71 20.0 seasons
MC44 100.00 93.08 96.54 20.0 seasons
MC46 100.00 95.29 97.65 30.0 seasons


Version Notes

Version Base Notes
MC0 N/A Simple 16 GF list w/ older dropped; hard coded HOT dodging; Single Wave only; Simple start distancer; No wall smoothing
MC1 MC0 Modified to prefer to orbit clockwise rather then stop condition
MC2 MC0 Added danger if closer then 200 to enemy
MC3 MC1 Added danger if closer then 200 to enemy
MC4 MC2 Added standard wall smoothing
MC5 MC3 Added standard wall smoothing
MC6 MC0 Added standard wall smoothing. Unlisted MC8 is MC2K7 edition
MC7 MC1 Added standard wall smoothing. Unlisted MC9 is MC2K7 edition
MC10 MC6 Changed start of round distancing; Holds onto wave longer before removing from log (shouldn't effect it)
MC11 MC10 Switched from per tick danger to full intersection range danger
MC12 MC11 Added danger based on nearness of gfs that did not intersect. Expecting huge drop vs linear/circular
MC13 MC12 Removed hardcoded head on dodging and added it to factor list instead
MC14 MC13 Removed out of field danger increase (should have removed this when I added wall smoothing)
MC15 MC14 Changed ETA > from -2 to -1 for determining wave soonest to hit. Will stop surfing waves 1 turn quicker. Not perfect, but will not be an issue later when PI wave removal is added. See MC10 (this is why it wouldn't effect it)
MC16 MC15 Changed to surf a second wave, adding danger/4.0 using danger check from second wave
MC17 MC14 Removed MC12 change (just to see)
MC18 MC14 Changed to use PI wave removal (once not intersecting after intersecting, it is removed); Uses the lowest ETA even if negative. See MC15
MC19 MC18 Two wave surfing; Added movements maxVelocity rule into drive for use in prediction.
Plan from here is to get the HOT up to near 100% vs WSCBotA, HOF and Barracuda. Before adding a multiple wave surfing if possible. Probably dive protection of the normal sort, if I cannot get a alternate method to work.
MC20 MC19 Removed the two wave surfing.
MC21 MC20 Increased max buffer to 32. (check)
MC22 MC20 Surfs imaginary waves based from gun heat (2 ticks). Inspired by RougeDC/DrussGT descriptions of this technique. Hopefully I did it right, and it will help resolve close range issues.
MC23 MC22 Added very low weight distance danger.
MC24 MC22 Defer to dodge bullets that do more damage if both would hit within a short period. (this might be the winner)
MC25 MC24 Changed preferred distance from 400 to 500.
MC28 MC25 Disabled Heat Waves, added DC segmenting on advancing and lateral velocity. Using Rednaxela's KdTree. Using nearest 16 only; unweighted.
MC29 MC28 Re-enabled Heat Waves.
MC30 MC29 Added collision danger.
MC31 MC29 Normalized segment data, folded lateral velocity (absolute).
MC32 MC31 Fixed bugs with data normalization.
MC33 MC32 Removed node limit in KdTree. (whoops :)
MC34 MC33 Added dive protected based on suggestion from Voidious.
MC35 MC34 Added entry danger weighting, based on time, importance, and similarity to current situation. (no segment specific weights yet)
MC36 MC35
  • Added limiter so it will never surf a wave that couldn't possibly hit it next turn
  • Removed limiter on look ahead in danger check
  • Changed danger check to start from current time, instead next turn
  • Reduced quarter turn time to account for heat waves.
MC37 MC36 Calculate danger from nearest edge to factor with static value for intersecting. Rather then additional danger for intersection.
MC38 MC36 Refactoring
  • Added complex 'avoid getting jammed in the ****ing corner' code.
    • Spent a few hours debugging this, had to force my robot to emulate getting jammed in the corner at the start of the round to debug it to my satisfaction.
  • Fixed bug in determining angle from enemy for waves (in both normal and heat waves)
  • Weighting has been removed (not readded yet)
MC39 MC38 Readded weighting; added weight based on scan similarity; Did kernel calculation on all waves based on distance from scan etc.
MC40 MC39 Fixed a bug with the weighting (maybe?)
MC41 MC40 Don't ask.
MC42 MC39 Fixed a bug with bulletHandler; Weighted lateral velocity 6 times more then advancing.
MC43 MC42 Removed segment weighting (important not to d these till the base is done, to get a good idea of improvements)
MC44 MC43
  • Added second wave surfing
  • Added less aggressive distancing (to be revised further later on)
  • Fixed heat wave bug
  • Fixed no-wave behavior bug
MC45 MC44
  • Changed to doing 30 seasons (getting to the point where spotting boosts in score with only 20 seasons is getting harder)
  • Added a smarter distancer function
  • Added some naïvely tweaked and weighted segments
  • Fixed a bug in the anti-corner movement
MC46 MC45
  • Tweaked the segments and weighting.
  • Fixed a bug in forward wall segment
MC47 MC46 Changed second wave weighting. (Etna 1.3's version)
MC48 MC47 Switched out wall smoothing algorithm; performance optimizations
MC49 MC48 Fixed a bug in the orbit center.

Version Chart
http://file.csdgn.org/robocode/kakeru-version-chart-9.jpg