Difference between revisions of "User:Starrynte/DCLSHResearch"

From Robowiki
Jump to navigation Jump to search
(0001)
m (e)
 
(25 intermediate revisions by the same user not shown)
Line 1: Line 1:
−
Experimenting with a new way of doing [[Dynamic Clustering]]: Locality Sensitive Hashing, when I have time I'll try to explain it. This will be the basis of the gun for [[Cannon]], so it will have to fit in a micro.
+
Experimenting with a new way of doing [[Dynamic Clustering]]: Locality Sensitive Hashing. Basically, the idea with LSH is that you reduce the data point down to 1 or 2 dimensions (for now, we'll say 1). Then, you place the reduced data point into bins. The idea is that data points that are close together will fall into the same bin. Then, when you search, you just get the bin of the query and search through that.
−
When I have time, I'll try to run a Targeting Challenge (from [[RoboResearch]] :) )
+
 
−
(I will probably have time tomorrow)
+
This will be the basis of the gun for [[Cannon]], so it will have to fit in a micro.
  
 
==Versions==
 
==Versions==
−
* '''0000:''' Distance and Lateral Velocity, no weighting, 32 LSH bins
+
* '''0000:''' ''BUG! Resets stats every round'' Distance and Lateral Velocity, no weighting, 32 LSH bins, euclidean distance
−
* '''0001:''' Realized I could use arrays instead of a Hashtable, along with a few more codesize shrinks saved 94 bytes (not doing major codesize shrinks yet). Also a few bugfixes
+
* '''0001:''' ''BUG! Resets stats every round'' Realized I could use arrays instead of a Hashtable, saved codesize. Also a few bugfixes
 +
* '''0002:''' ''BUG! Resets stats every round'' 0001 with bugfixes
 +
* '''0003:''' ''BUG! Resets stats every round'' 0002 with a gunheat segment
 +
* '''0004:''' ''BUG! Resets stats every round'' 0002 with segments: round number, lateral velocity, distance from center, distance from me; experimenting with weightings still (codesize 660 without shrinking yet)
 +
* '''0005:''' ''BUG! Resets stats every round'' Shrink of 0004, removing distance-from-me segment (codesize 615)
 +
* '''0006:''' Resetting stats bug fixed. Segments lateral velocity, distance from center, distance from me; distance is then weighted by recentness (codesize 625)
 +
* '''0007:''' Shrink of 0006, STILL playing with weightings (amazing how a difference of 0.3 can cause as much as 1 extra point in challenge) (codesize 607)
 +
* '''0008:''' ''BUG! ArrayIndexOutOfBounds: negative'' Added lateral acceleration segment. All segments unweighted (codesize 650)
 +
* '''0009:''' Wasn't sure why the bug was there, quickly hacked a fix (will look more into it when I have time) (codesize 650)
 +
==Challenges==
 +
===[[Targeting Challenge 2K7|Targeting Challenge 2K7 (Fast Learning)]]===
 +
{| border="1" cellpadding="3" style="border-collapse: collapse; font-size: 85%; color: black;"
 +
!Version
 +
!CC
 +
!RMX
 +
!SHA
 +
!WS
 +
!WOE
 +
!Surf
 +
!DM
 +
!FT
 +
!GG
 +
!RMC
 +
!WLO
 +
!No Surf
 +
!Total
 +
!Seasons
 +
!Comments
 +
|-
 +
| 0002 || 57.93 || 63.25 || 59.36 || 56.67 || 63.37 || '''60.12''' || 75.48 || 71.52 || 60.87 || 68.59 || 67.59 || '''68.81''' || '''64.46''' || 30.0 seasons || buggy
 +
|-
 +
| 0003 || 57.34 || 64.69 || 60.48 || 58.62 || 65.41 || '''61.31''' || 73.49 || 69.88 || 59.93 || 68.06 || 64.79 || '''67.23''' || '''64.27''' || 50.0 seasons || buggy
 +
|-
 +
| 0004 || 60.94 || 68.58 || 62.27 || 67.62 || 69.27 || '''65.73''' || 79.81 || 73.62 || 68.46 || 72.92 || 71.15 || '''73.19''' || '''69.46''' || 50.0 seasons || buggy
 +
|-
 +
| 0006 || 64.28 || 72.04 || 66.45 || 68.62 || 70.71 || '''68.42''' || 78.09 || 74.07 || 68.46 || 72.83 || 72.68 || '''73.23''' || '''70.82''' || 50.0 seasons ||
 +
|-
 +
| 0007 || 64.79 || 70.67 || 65.55 || 69.97 || 70.09 || '''68.21''' || 79.02 || 74.81 || 67.60 || 72.67 || 71.84 || '''73.19''' || '''70.70''' || 50.0 seasons ||
 +
|-
 +
| 0009 || 66.23 || 72.87 || 63.75 || 70.05 || 72.57 || '''69.09''' || 79.38 || 73.51 || 72.00 || 74.34 || 71.69 || '''74.19''' || '''71.64''' || 10.0 seasons || more to be run
 +
|}
 +
===[[Targeting Challenge RM]]===
 +
{| border="1" cellpadding="3" style="border-collapse: collapse; font-size: 85%; color: black;"
 +
!Version
 +
!Aspd
 +
!Sprw
 +
!Fhqw
 +
!Yngw
 +
!FlMn
 +
!EASY
 +
!Tron
 +
!HTTC
 +
!RnMB
 +
!DlMc
 +
!Grbb
 +
!MEDIUM
 +
!SnDT
 +
!Cgrt
 +
!Frtn
 +
!WkOb
 +
!RkMc
 +
!HARD
 +
!TOTAL
 +
!Seasons
 +
!Comments
 +
|-
 +
| 0002 || 75.61 || 82.73 || 76.67 || 88.58 || 76.78 || '''80.08''' || 76.80 || 74.37 || 71.28 || 75.45 || 61.37 || '''71.85''' || 57.67 || 56.86 || 70.94 || 67.49 || 69.50 || '''64.49''' || '''72.14''' || 50.0 seasons || buggy
 +
|-
 +
| 0003 || 75.67 || 80.30 || 74.07 || 89.54 || 76.36 || '''79.19''' || 75.77 || 73.83 || 72.95 || 73.49 || 59.93 || '''71.19''' || 56.61 || 57.23 || 69.88 || 64.79 || 68.06 || '''63.31''' || '''71.23''' || 50.0 seasons || buggy
 +
|-
 +
| 0004 || 75.98 || 82.85 || 84.83 || 92.72 || 82.58 || '''83.79''' || 77.74 || 77.94 || 76.64 || 79.81 || 68.46 || '''76.12''' || 63.52 || 64.25 || 73.62 || 71.15 || 72.92 || '''69.09''' || '''76.33''' || 50.0 seasons || buggy
 +
|-
 +
| 0006 || 77.27 || 84.23 || 85.05 || 92.90 || 80.99 || '''84.09''' || 75.92 || 75.06 || 79.09 || 78.09 || 68.46 || '''75.33''' || 62.09 || 63.06 || 74.07 || 72.68 || 72.83 || '''68.95''' || '''76.12''' || 50.0 seasons ||
 +
|-
 +
| 0007 || 77.11 || 83.77 || 84.05 || 91.71 || 81.79 || '''83.69''' || 77.11 || 74.09 || 79.41 || 79.02 || 67.60 || '''75.45''' || 61.63 || 63.34 || 74.81 || 71.84 || 72.67 || '''68.86''' || '''76.00''' || 50.0 seasons ||
 +
|}

Latest revision as of 06:33, 30 September 2009

Experimenting with a new way of doing Dynamic Clustering: Locality Sensitive Hashing. Basically, the idea with LSH is that you reduce the data point down to 1 or 2 dimensions (for now, we'll say 1). Then, you place the reduced data point into bins. The idea is that data points that are close together will fall into the same bin. Then, when you search, you just get the bin of the query and search through that.

This will be the basis of the gun for Cannon, so it will have to fit in a micro.

Versions

  • 0000: BUG! Resets stats every round Distance and Lateral Velocity, no weighting, 32 LSH bins, euclidean distance
  • 0001: BUG! Resets stats every round Realized I could use arrays instead of a Hashtable, saved codesize. Also a few bugfixes
  • 0002: BUG! Resets stats every round 0001 with bugfixes
  • 0003: BUG! Resets stats every round 0002 with a gunheat segment
  • 0004: BUG! Resets stats every round 0002 with segments: round number, lateral velocity, distance from center, distance from me; experimenting with weightings still (codesize 660 without shrinking yet)
  • 0005: BUG! Resets stats every round Shrink of 0004, removing distance-from-me segment (codesize 615)
  • 0006: Resetting stats bug fixed. Segments lateral velocity, distance from center, distance from me; distance is then weighted by recentness (codesize 625)
  • 0007: Shrink of 0006, STILL playing with weightings (amazing how a difference of 0.3 can cause as much as 1 extra point in challenge) (codesize 607)
  • 0008: BUG! ArrayIndexOutOfBounds: negative Added lateral acceleration segment. All segments unweighted (codesize 650)
  • 0009: Wasn't sure why the bug was there, quickly hacked a fix (will look more into it when I have time) (codesize 650)

Challenges

Targeting Challenge 2K7 (Fast Learning)

Version CC RMX SHA WS WOE Surf DM FT GG RMC WLO No Surf Total Seasons Comments
0002 57.93 63.25 59.36 56.67 63.37 60.12 75.48 71.52 60.87 68.59 67.59 68.81 64.46 30.0 seasons buggy
0003 57.34 64.69 60.48 58.62 65.41 61.31 73.49 69.88 59.93 68.06 64.79 67.23 64.27 50.0 seasons buggy
0004 60.94 68.58 62.27 67.62 69.27 65.73 79.81 73.62 68.46 72.92 71.15 73.19 69.46 50.0 seasons buggy
0006 64.28 72.04 66.45 68.62 70.71 68.42 78.09 74.07 68.46 72.83 72.68 73.23 70.82 50.0 seasons
0007 64.79 70.67 65.55 69.97 70.09 68.21 79.02 74.81 67.60 72.67 71.84 73.19 70.70 50.0 seasons
0009 66.23 72.87 63.75 70.05 72.57 69.09 79.38 73.51 72.00 74.34 71.69 74.19 71.64 10.0 seasons more to be run

Targeting Challenge RM

Version Aspd Sprw Fhqw Yngw FlMn EASY Tron HTTC RnMB DlMc Grbb MEDIUM SnDT Cgrt Frtn WkOb RkMc HARD TOTAL Seasons Comments
0002 75.61 82.73 76.67 88.58 76.78 80.08 76.80 74.37 71.28 75.45 61.37 71.85 57.67 56.86 70.94 67.49 69.50 64.49 72.14 50.0 seasons buggy
0003 75.67 80.30 74.07 89.54 76.36 79.19 75.77 73.83 72.95 73.49 59.93 71.19 56.61 57.23 69.88 64.79 68.06 63.31 71.23 50.0 seasons buggy
0004 75.98 82.85 84.83 92.72 82.58 83.79 77.74 77.94 76.64 79.81 68.46 76.12 63.52 64.25 73.62 71.15 72.92 69.09 76.33 50.0 seasons buggy
0006 77.27 84.23 85.05 92.90 80.99 84.09 75.92 75.06 79.09 78.09 68.46 75.33 62.09 63.06 74.07 72.68 72.83 68.95 76.12 50.0 seasons
0007 77.11 83.77 84.05 91.71 81.79 83.69 77.11 74.09 79.41 79.02 67.60 75.45 61.63 63.34 74.81 71.84 72.67 68.86 76.00 50.0 seasons
Retrieved from "http://robowiki.net/w/index.php?title=User:Starrynte/DCLSHResearch&oldid=12662"