HAR-Pipeline/TRAINING_VARIABLES.py at master · hessenh/HAR-Pipeline · GitHub

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
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97

class VARS(object):
	# Used to format a list when creating cnn model name
	def format_list(list_input):
		string_list = ""
		for s in list_input:
			string_list +=str(s)+'_'
		return string_list


	''' Variables '''
	# Convertion of activities
	CONVERTION = {1:1, 2:2, 4:3, 5:4, 6:5, 7:6, 8:7, 10:8, 11:8, 13:9, 14:10}
	REMOVE_ACTIVITIES =  [0,3,9,11,16,12,15,17]
	ACTIVITY_NAMES_CONVERTION = {1:'WALKING',2:'RUNNING', 3:'STAIRS (UP)', 4:'STAIRS (DOWN)', 5:'STANDING', 6:'SITTING', 7:'LYING', 8:'BENDING', 9:'CYCLING (SITTING)', 10:'CYCLING (STANDING)'}
	ACTIVITIES = [0,1,2,3,4,5,6,7,8,9]
	NUMBER_OF_ACTIVITIES = 10

	# Amount of overlap used for training, testing and prediction phase. None equals zero overlap.
	# 20 equals 80% overlap. 40 = 60% overlap. Yeah, not intuitive, but...
	TESTING_OVERLAP = None
	TRAINING_OVERLAP = 20
	PREDICTING_OVERLAP = None
	# Length of data window. With sampling frequency as 100Hz, we chose 100, meaning 1.0 second windows.
	WINDOW_LENGTH = 100

	# Paths for training, testing and predicting data
	TRAINING_PATH = "DATA/TRAINING"
	TESTING_PATH = 'DATA/TESTING'
	PREDICTING_PATH = 'DATA/PREDICTING'

	# Name of sensors and label. This is how the system differentiate between the different files located in each subject folder.
	# The back sensor file must have a word (e.g. "BACK") separated by two underscores ("_"). This word can be changed by altering the variables bellow (e.g. "CHEST").
	SENSOR_1 = 'BACK'
	SENSOR_2 = 'THIGH'
	SENSORS = [SENSOR_1, SENSOR_2]
	LABEL = 'LAB'
	WINDOW_NAME_SENSOR = 'SENSORS'
	WINDOW_NAME_LABEL = 'LABEL'

	''' CNN SPECIFIC'''
	# Input size is window-length multiplied by the number of sensor axes.
	CNN_INPUT_SIZE = 600
	CNN_OUTPUT_SIZE = 10
	# Number of training iterations. In our report, we have used 20000.
	CNN_NUMBER_OF_ITERATIONS = 20000
	# Number of instanses used between changing training parameters
	CNN_BATCH_SIZE = 100
	# Length of kernel in horizontal orientation
	CNN_FILTER_X = 30
	# Length of kernel in vertical orientation
	CNN_FILTER_Y = 1
	# Used to resize the input
	CNN_RESIZE_Y = 6
	CNN_RESIZE_X = 100
	CNN_WINDOW = CNN_INPUT_SIZE / 6

	# Numer of kernels (features) in each layer. Do not change the first value (1).
	CNN_KERNEL_LIST = [1, 20, 40]
	CNN_NUMBER_OF_KERNELS = len(CNN_KERNEL_LIST)-1
	# Not so easy to grasp, but..
	CNN_CONNECTIONS_INN = CNN_RESIZE_Y * (CNN_RESIZE_X - (CNN_NUMBER_OF_KERNELS*CNN_FILTER_X) + CNN_NUMBER_OF_KERNELS) * CNN_KERNEL_LIST[-1]
	CNN_NEURAL_LIST = [CNN_CONNECTIONS_INN] + [1500] + [CNN_OUTPUT_SIZE]
	CNN_FILTER_TYPE = 'VALID'
	# This name is saved in all tensors throughout the model, meaning that this must be the same when creating and loading a model.
	CNN_MODEL_NAME = str(CNN_INPUT_SIZE) + '_' + "conv_" + format_list(CNN_KERNEL_LIST[1:]) + "neural_" +  format_list(CNN_NEURAL_LIST[1:-1]) + CNN_FILTER_TYPE + '_' + str(CNN_NUMBER_OF_ITERATIONS)
	CNN_MODEL_PATH = 'MODELS/' + CNN_MODEL_NAME


	''' VITERBI '''
	# This is the amount of data windows used to create the transition matrix used in the Viterbi.
	# Our recommended length is the number of windows for the first training subject.
	VITERBI_LENGTH_OF_TRANSITION_DATA = 19676

	# Number of iterations for the Baum Welch algorithm
	VITERBI_BAUM_WELCH_ITERATIONS = 5
	# Paths used for saving viterbi and cnn predictions
	VITERBI_PREDICTION_PATH_TRAINING = 'VITERBI/PREDICTION_TRAINING.csv'
	VITERBI_ACTUAL_PATH_TRAINING = 'VITERBI/ACTUAL_TRAINING.csv'
	VITERBI_PREDICTION_PATH_TESTING = 'VITERBI/PREDICTION_TESTING.csv'
	PREDICTION_RESULT_TESTING = 'VITERBI/PREDICTION_RESULT_TESTING.csv'
	VITERBI_RESULT_TESTING = 'RESULTS/RESULT_TESTING.csv'
	RESULT_TESTING_JSON = 'RESULTS/RESULT_TESTING_JSON.json'

	VITERBI_PREDICTION_PATH_PREDICTING = 'VITERBI/PREDICTION_PREDICTING.csv'
	VITERBI_RESULT_PREDICTING = 'RESULTS/RESULT_PREDICTING.csv'

	# The different states the Viterbi algorithm uses
	VITERBI_STATES = [1,2,3,4,5,6,7,8,9,10]

	# The vitebi algorithm saves and loads the transition matrix. This matrix is saves as a dictionary.
	VITERBI_TRANSITION_DICTIONARY_PATH = 'VITERBI/TRANSITION_DICTIONARY'

	# Convert the labels of the activities. Used for visualization
	VISUALIZATION_CONVERTION = {1:4, 2:10, 3:5, 4:6, 5:3, 6:2, 7:1, 8:9, 9:7, 10:8}