Archive of July 2012

adaptive training #3

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
012345678910111213141516171819202122232425
avg0.12991110.15207480.20727720.31870840.39584120.64687490.60108480.76554070.92835640.95608071.270251.2402941.0785981.2572121.0924871.3274671.088351.1549031.3703621.1982141.2119511.1785781.2537440.95138751.3470260.8872428
max0.61878870.76564751.1996931.0606281.6245532.0148892.002721.5329191.7799932.9245082.0194351.8904252.370942.0550641.8125472.1251032.0440511.911442.0505212.573581.4987481.4173371.9088071.2616831.808071.39744
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

adaptive training #2

run length:48
012345678910111213141516171819202122232425262728293031323334353637383940414243444546
avg0.113330.12169220.14104080.22551690.34389470.53086220.70928660.74734130.83809930.83153670.92424280.83931471.1031981.0540691.1116831.0516291.0280541.0436360.96778351.015111.0450511.1380881.0021890.87519880.94239440.90267321.0629760.91720461.3347261.0799310.70328121.3776991.2364361.7413621.0817631.3413660.89425670.92669591.0798421.0677050.75573451.3546180.93386291.0700211.2887321.1961181.037447
max0.48422240.73180870.78723031.3143611.1677711.0261971.4225641.5387461.4745011.2834531.4234181.3491161.6149291.3924651.46851.2743141.2034641.2288271.081251.1712781.1069391.3598231.1064740.92189871.063581.0071611.3766670.99576691.4478711.1733470.82806491.5154071.3694681.8393451.1613621.3774320.94413690.95007731.1096591.0922070.77086321.388190.94599271.0938741.3221041.240411.065003
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

adaptive training

run length:35
0123456789101112131415161718192021222324252627282930313233
avg0.10547280.11858760.12581310.1305870.15065980.21223190.45814970.85867041.201991.1702190.97634921.1764810.9386680.95721230.9022071.1338121.0987550.98921851.54691.3848551.1561850.88165311.1553131.3702980.90868941.0918010.92115311.0643591.4455520.92722041.2198711.2816231.3450691.290509
max0.17875080.34220440.48802440.48697691.0478151.2885451.5048311.275061.438191.7967961.3673741.2294141.0677251.2654681.2304311.2160181.1790891.1734581.6813611.4966241.2256761.2055231.2308031.4367371.252531.1742990.98011091.2070271.7110451.160891.4354491.3458951.409911.366372
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

overhead run #4

and a run w/another hidden layer
run length:33
012345678910111213141516171819202122232425262728293031
avg0.11048140.1368160.19085030.32848490.66843531.2827341.3022611.2063011.3588411.1262561.3977361.2466041.3233510.89090271.1224781.1082331.6312261.2917641.2202251.215471.2995221.2957941.3604451.0066651.0052651.2855061.7236880.98484360.94554330.93132651.1489891.272345
max0.28713411.2145761.0672691.1039581.3546961.4754171.3662021.2345411.3906231.1732981.4265181.7649771.3961520.91080251.1449391.1335091.905031.3205141.2461031.2393811.4987611.3394911.463061.1353131.0306481.3285661.7724881.1661320.97462161.0904311.1798141.307164
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

Overhead run #3

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
01234567891011121314151617181920212223242526
avg0.11118350.11751510.11617130.12565720.13688070.1563310.16561640.20839510.26791690.37956520.49561230.65011810.5612270.7630230.6263450.6848750.91963410.75143650.88850591.0652710.96150930.8098810.7446751.0010240.8382290.89157540.8473446
max0.24504590.27551090.22391210.42355140.48726720.48368050.91547671.0488010.80019580.85897091.1265221.1067790.94524681.0103830.79434391.0487691.185260.94576761.1496091.2348391.2027670.98070281.0266311.0796510.87430930.93146660.9513353
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