Archive for March, 2017

Mar
09

convolutional neural network for 3d object

3D Convolutional Neural Networks for Human Action Recognition

http://machinelearning.wustl.edu/mlpapers/paper_files/icml2010_JiXYY10.pdf

VoxNet: A 3D Convolutional Neural Network for Real-Time Object Recognition

https://www.ri.cmu.edu/pub_files/2015/9/voxnet_maturana_scherer_iros15.pdf

Question Website:

use tensorflow

http://stackoverflow.com/questions/33630840/convolutional-neural-networks-and-3d-images

use keras

https://github.com/fchollet/keras/issues/1359

Mar
08

credit flawed detection kaggle dataset

  1. base line: use all the attributes, tree algorithm, cross-validation parameter 3: 
      true 0 true 1 class precision
    pred. 0 284253 132 99.95%
    pred. 1 62 360 85.31%
    class recall 99.98% 73.17%  
    PerformanceVector:
    accuracy: 99.93% +/- 0.01% (mikro: 99.93%)
    ConfusionMatrix:
    True:	0	1
    0:	284253	132
    1:	62	360
    precision: 85.68% +/- 6.02% (mikro: 85.31%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284253	132
    1:	62	360
    recall: 73.14% +/- 6.06% (mikro: 73.17%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284253	132
    1:	62	360
    AUC (optimistic): 0.965 +/- 0.028 (mikro: 0.965) (positive class: 1)
    AUC: 0.858 +/- 0.035 (mikro: 0.858) (positive class: 1)
    AUC (pessimistic): 0.752 +/- 0.057 (mikro: 0.752) (positive class: 1)
    
    
    Use random forest tree:
    PerformanceVector:
    accuracy: 99.87% +/- 0.01% (mikro: 99.87%)
    ConfusionMatrix:
    True:	0	1
    0:	284298	342
    1:	17	150
    precision: 93.21% +/- 7.43% (mikro: 89.82%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284298	342
    1:	17	150
    recall: 30.47% +/- 12.44% (mikro: 30.49%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284298	342
    1:	17	150
    AUC (optimistic): 0.992 +/- 0.010 (mikro: 0.992) (positive class: 1)
    AUC: 0.852 +/- 0.039 (mikro: 0.852) (positive class: 1)
    AUC (pessimistic): 0.713 +/- 0.081 (mikro: 0.713) (positive class: 1)
  2. use selected attributes: time, v1,v2,v3, amount, class

    PerformanceVector

    PerformanceVector:
    accuracy: 99.84% +/- 0.01% (mikro: 99.84%)
    ConfusionMatrix:
    True:	0	1
    0:	284310	456
    1:	5	36
    precision: 93.00% +/- 11.40% (mikro: 87.80%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284310	456
    1:	5	36
    recall: 7.34% +/- 4.55% (mikro: 7.32%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284310	456
    1:	5	36
    AUC (optimistic): 0.998 +/- 0.006 (mikro: 0.998) (positive class: 1)
    AUC: 0.537 +/- 0.024 (mikro: 0.537) (positive class: 1)
    AUC (pessimistic): 0.075 +/- 0.044 (mikro: 0.075) (positive class: 1)
  3. select v1,v2,v3

    PerformanceVector

    PerformanceVector:
    accuracy: 99.85% +/- 0.01% (mikro: 99.85%)
    ConfusionMatrix:
    True:	0	1
    0:	284295	408
    1:	20	84
    precision: 81.93% +/- 12.69% (mikro: 80.77%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284295	408
    1:	20	84
    recall: 17.08% +/- 5.20% (mikro: 17.07%) (positive class: 1)
    ConfusionMatrix:
    True:	0	1
    0:	284295	408
    1:	20	84
    AUC (optimistic): 0.992 +/- 0.010 (mikro: 0.992) (positive class: 1)
    AUC: 0.586 +/- 0.026 (mikro: 0.586) (positive class: 1)
    AUC (pessimistic): 0.181 +/- 0.055 (mikro: 0.181) (positive class: 1)