This is a supporting page to our paper 
Shapelet Ensemble for Multi-dimensional Time Series

 

Supporting Material

We have a VS C++ code and executable files. Please email me for the passwords.

  • The colorful paper is here.
  • Supplementary document is here.
  • VS C++ source and sample data sets to run all experiments is here. (ask for password: mcetin@unm.edu)

 

Spreadsheet of Experimental Results 

We have compiled results of all the experiments in a spreadsheet that shows the accuracy comparison and execution time (All time measurements are in seconds) to the state-of-the-art.

 

 

 

 

 

mc^2

State-of-the-art algorithm

Original algorithm

Data Sets

Classes

Train

Test

TC Length

Execution Time

Accuracy

Accuracy(voting)

StDev

Execution Time

Accuracy

StDev

Execution Time

Accuracy

StDev

Beef                 

5

30

30

470

18.48

54.17

60.00

6.74

106.27

55.83

4.82

242.29

56.67

0.00

CBF                  

3

30

900

128

3.56

93.12

95.55

1.70

8.35

94.12

1.25

66.86

88.56

0.00

ChlorineConcentration

3

467

3840

166

183.53

57.83

58.95

1.38

620.79

57.81

1.22

36402.28

61.85

0.00

CinC_ECG_torso       

4

40

1380

1639

190.83

56.46

57.17

4.19

1053.30

58.18

2.58

2149.95

69.86

0.00

Coffee               

2

28

28

286

3.16

90.71

96.43

2.43

15.55

91.96

3.04

621.91

96.43

0.00

Cricket_X            

12

390

390

300

1122.33

43.71

55.13

3.01

4941.33

45.08

4.28

Not completed

Cricket_Y            

12

390

390

300

1399.22

47.65

58.21

3.08

4596.29

49.95

3.49

Not completed

Cricket_Z            

12

390

390

300

1035.46

40.95

59.23

2.99

3798.16

41.85

2.56

Not completed

DiatomSizeReduction  

4

16

306

345

1.97

92.45

93.46

1.59

14.34

90.21

2.46

184.34

80.07

0.00

ECG2                  

2

92

89

750

51.09

100.00

100.00

0.00

268.23

100.00

0.00

Not completed

ECG3                 

3

92

89

750

177.78

100.00

100.00

0.00

688.22

100.00

0.00

Not completed

ECGFiveDays          

2

23

861

136

1.95

99.82

99.42

0.40

4.22

99.61

0.23

47.64

99.42

0.00

FaceAll              

14

560

1690

131

377.21

62.21

68.22

1.31

1051.71

63.57

0.92

16255.48

65.86

0.00

FaceFour              

4

24

88

350

15.23

90.85

94.32

2.91

86.19

91.19

1.60

561.18

48.86

0.00

FacesUCR             

14

200

2050

131

91.82

68.19

84.39

3.71

249.75

68.68

1.96

2528.52

66.24

0.00

fish                  

7

175

175

463

63.60

80.51

83.43

2.28

415.76

80.91

1.37

11153.03

77.71

0.00

Gun_Point            

2

50

150

150

2.62

93.33

93.33

0.30

8.07

93.77

1.64

266.10

89.33

0.00

Haptics               

5

155

308

1092

214.63

35.13

43.18

2.36

1464.48

35.18

3.69

Not completed

InlineSkate          

7

100

550

1882

1556.70

25.16

29.27

2.41

4251.60

23.57

2.81

Not completed

ItalyPowerDemand     

2

67

1029

24

0.25

90.27

92.23

3.07

0.25

91.64

2.21

4.92

93.59

0.00

Lighting2            

2

60

61

637

201.67

55.98

55.45

4.77

683.85

55.66

5.22

5297.60

42.62

0.00

Lighting7            

7

70

73

319

82.56

55.89

68.49

5.12

292.87

57.05

3.65

8619.35

54.79

0.00

MALLAT               

8

55

2345

1024

47.61

87.83

93.22

5.08

384.64

88.86

2.76

1254.91

65.63

0.00

MedicalImages        

10

381

760

99

68.71

57.06

62.24

2.06

153.10

60.86

1.67

19325.20

58.68

0.00

MoteStrain           

2

20

1252

84

0.53

79.69

79.79

0.46

1.16

79.27

0.92

6.87

83.23

0.00

OSULeaf              

6

200

242

427

163.16

66.03

77.28

3.03

1563.46

66.69

3.24

14186.53

68.60

0.00

OliveOil             

4

30

30

570

5.93

70.33

76.67

3.40

43.20

71.50

4.65

502.27

83.33

0.00

Sony                 

2

601

20

70

31.91

93.50

95.00

2.86

53.62

92.75

3.43

Not completed

SonyAIBORobotSurface 

2

20

601

70

0.57

68.55

96.34

0.00

1.04

69.82

5.69

4.56

86.02

0.00

SonyAIBORobotSurfaceII

2

27

953

65

0.74

77.84

85.41

4.03

1.26

79.21

1.15

9.76

84.58

0.00

StarLightCurves      

3

1000

8236

1024

1344.15

89.93

91.88

1.00

10068.04

90.43

1.00

Not completed

SwedishLeaf          

15

500

625

128

141.50

77.26

86.08

2.50

404.17

77.70

2.18

11953.61

81.28

0.00

Symbols              

6

25

995

398

5.01

83.07

88.94

3.90

43.31

84.63

2.66

894.26

64.32

0.00

synthetic_control    

6

300

300

60

36.66

91.52

94.33

1.52

56.05

92.65

1.63

3667.43

47.00

0.00

Trace                

4

100

100

275

62.30

99.60

100.00

0.60

184.53

99.60

0.82

4626.86

100.00

0.00

TwoLeadECG           

2

23

1139

82

0.42

91.90

92.45

2.27

0.91

92.74

1.03

14.29

85.60

0.00

Two_Patterns         

4

1000

4000

128

739.34

90.17

89.85

5.19

1980.66

92.39

1.37

65783.11

53.90

0.00

uWaveGestureLibrary_X

8

896

3582

315

921.62

63.96

73.45

1.23

5460.62

66.35

1.12

Not completed

uWaveGestureLibrary_Y

8

896

3582

315

1148.54

56.63

63.79

1.60

5934.02

57.36

1.22

Not completed

uWaveGestureLibrary_Z

8

896

3582

315

1128.92

61.28

65.80

1.81

5929.55

62.82

1.63

Not completed

wafer                

2

1000

6000

152

171.93

99.67

99.75

0.25

348.93

99.51

0.34

34653.13

99.88

0.00

yoga

2

300

3000

426

131.31

68.88

71.97

1.94

923.36

69.99

1.78

11388.99

74.00

0.00

fMRI Singe Task

2

50

25

145

19.89

67.60

76.00

8.50

125.31

62.00

6.68

Not completed

fMRI Multi Task

2

50

25

411

28.72

66.32

72.00

10.61

239

65.80

7.62

Not completed

MEG Rest

2

60

31

14000

13991.20

60.50

60.00

5.42

43280.08

63.00

6.00

Not completed

 

(a) Accuracy comparison between our algorithm and the current state-of-the-art algorithm (b) Accuracy comparison when we use voting based ensemble (c) Execution time comparison between our algorithm and the current state of the algorithm.

 

(a) Accuracy comparison between our algorithm and the original algorithm (b) Accuracy comparison when we use voting based ensemble (c) Execution time comparison between our algorithm and the original algorithm.

 

Datasets

For our experiments, we use 43 datasets to compare the efficiency of our algorithm to the state-of-the-art algorithm. We attempted tests on all the 44 datasets that we have for original algorithm; however, we abandoned the experiments (14 data sets) in which the original algorithm had not finished after 18 hours. Each row in every file is a time series and the first value of a row is the class label.

 

Sensory Gating task for functional MRI (fMRI)

 

We used the time series of sensory motor network (putamen) component. There are 50(SP=23, HC=27) time series for training data sets and 25 (SP=9, HC=16) time series for testing data sets. The length of each time series is 145.

 

The decision tree from concatenated signals is shown in the top. Three trees from the three sessions are shown in the bottom. Shapelets and nodes in the trees have matching colors. Label 0 represents con-trols and 1 represents patients. Signals in red are from patients and in blue are from healthy subjects.

 

Multi-modal sensory integration task for functional MRI (fMRI)

 

We used the time series of sensory motor network (putamen) component. There are 50(SP=23, HC=27) time series for training data sets and 25 (SP=9, HC=16) time series for testing data sets. The length of each time series is 411.

 

The decision tree from concatenated signals is shown in the top. Trees from the six sessions are shown in the bottom. Shapelets and nodes in the trees have matching colors. Label 0 represents con-trols and 1 represents patients. Signals in red are from patients and in blue are from healthy subjects.

 

Fetal Electrocardiogram

 

The training set contains a balanced (45/44/46) mix of 135 time series of three abdominal channels. The testing set also contains a balanced (42/50/42) mix of time series. The length of each time series is 750.

 

 

(a) Training set of FECG data for all classes. (b) Shapelets are shown in green. First shapelet classifies the sesnsor-1, Second shapelet classifies the sesnsor-3.

 

This page is created by - Mustafa S Cetin