it's amazing what changing the targets can do. now they're spread evenly from the starting position (uniform distance) and they seem to do much better, rather than finding a local maximum (a single target that's easier to get to than the others).
run length:27 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 |
---|
avg | 0.1299111 | 0.1520748 | 0.2072772 | 0.3187084 | 0.3958412 | 0.6468749 | 0.6010848 | 0.7655407 | 0.9283564 | 0.9560807 | 1.27025 | 1.240294 | 1.078598 | 1.257212 | 1.092487 | 1.327467 | 1.08835 | 1.154903 | 1.370362 | 1.198214 | 1.211951 | 1.178578 | 1.253744 | 0.9513875 | 1.347026 | 0.8872428 |
---|
max | 0.6187887 | 0.7656475 | 1.199693 | 1.060628 | 1.624553 | 2.014889 | 2.00272 | 1.532919 | 1.779993 | 2.924508 | 2.019435 | 1.890425 | 2.37094 | 2.055064 | 1.812547 | 2.125103 | 2.044051 | 1.91144 | 2.050521 | 2.57358 | 1.498748 | 1.417337 | 1.908807 | 1.261683 | 1.80807 | 1.39744 |
---|
- Population Size
- 100
- Run Length
- 20
- Do Training
- False
- Do Hybrid Training
- True
- Do Competitive Run
- False
- Do Adaptive Training
- True
- Inputs
- hasTarget,targetLeft,targetRight,targetNorth,targetSouth,dirToTarget,distToTarget,wall,sensor0,sensor1,sensor2
- Outputs
- left,right,up,down,run,sensordir0,sensordir1,sensordir2
- Hidden Layers
- 8 12 12 12 8
- Back Propogation
- True
- Learning rate
- 0.5
- Momentum
- 0.1
- Growth rate
- 0.5
and a run w/another hidden layer
run length:33 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 |
---|
avg | 0.1104814 | 0.136816 | 0.1908503 | 0.3284849 | 0.6684353 | 1.282734 | 1.302261 | 1.206301 | 1.358841 | 1.126256 | 1.397736 | 1.246604 | 1.323351 | 0.8909027 | 1.122478 | 1.108233 | 1.631226 | 1.291764 | 1.220225 | 1.21547 | 1.299522 | 1.295794 | 1.360445 | 1.006665 | 1.005265 | 1.285506 | 1.723688 | 0.9848436 | 0.9455433 | 0.9313265 | 1.148989 | 1.272345 |
---|
max | 0.2871341 | 1.214576 | 1.067269 | 1.103958 | 1.354696 | 1.475417 | 1.366202 | 1.234541 | 1.390623 | 1.173298 | 1.426518 | 1.764977 | 1.396152 | 0.9108025 | 1.144939 | 1.133509 | 1.90503 | 1.320514 | 1.246103 | 1.239381 | 1.498761 | 1.339491 | 1.46306 | 1.135313 | 1.030648 | 1.328566 | 1.772488 | 1.166132 | 0.9746216 | 1.090431 | 1.179814 | 1.307164 |
---|
- Population Size
- 100
- Run Length
- 20
- Do Training
- False
- Do Hybrid Training
- True
- Do Competitive Run
- False
- Inputs
- hasTarget,targetLeft,targetRight,targetNorth,targetSouth,dirToTarget,distToTarget,wall,sensor0,sensor1,sensor2
- Outputs
- left,right,up,down,run,sensordir0,sensordir1,sensordir2
- Hidden Layers
- 8 12 12 12 8
- Back Propogation
- True
- Learning rate
- 0.5
- Momentum
- 0.1
- Growth rate
- 0.5
Revisiting this whole neural net thing. Did another, longer run after tweaking a bunch of things (mostly removing cruft from the side-view test, like jumping and "grounded" and such).
run length:28 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 |
---|
avg | 0.1111835 | 0.1175151 | 0.1161713 | 0.1256572 | 0.1368807 | 0.156331 | 0.1656164 | 0.2083951 | 0.2679169 | 0.3795652 | 0.4956123 | 0.6501181 | 0.561227 | 0.763023 | 0.626345 | 0.684875 | 0.9196341 | 0.7514365 | 0.8885059 | 1.065271 | 0.9615093 | 0.809881 | 0.744675 | 1.001024 | 0.838229 | 0.8915754 | 0.8473446 |
---|
max | 0.2450459 | 0.2755109 | 0.2239121 | 0.4235514 | 0.4872672 | 0.4836805 | 0.9154767 | 1.048801 | 0.8001958 | 0.8589709 | 1.126522 | 1.106779 | 0.9452468 | 1.010383 | 0.7943439 | 1.048769 | 1.18526 | 0.9457676 | 1.149609 | 1.234839 | 1.202767 | 0.9807028 | 1.026631 | 1.079651 | 0.8743093 | 0.9314666 | 0.9513353 |
---|
- Population Size
- 100
- Run Length
- 20
- Do Training
- False
- Do Hybrid Training
- True
- Do Competitive Run
- False
- Inputs
- hasTarget,targetLeft,targetRight,targetNorth,targetSouth,dirToTarget,distToTarget,wall,sensor0,sensor1,sensor2
- Outputs
- left,right,up,down,run,sensordir0,sensordir1,sensordir2
- Hidden Layers
- 8 12 12 8
- Back Propogation
- True
- Learning rate
- 0.5
- Momentum
- 0.1
- Growth rate
- 0.5