This repository has been archived by the owner on Jul 21, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathFeature.py
195 lines (173 loc) · 8.3 KB
/
Feature.py
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
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
import pandas
import random
import numpy
import glob
import itertools
from scipy import signal
from MatFile import *
import os
import sklearn.preprocessing
import sklearn.decomposition
class Feature:
def __init__(self, setting):
self.setting = setting
self.result_x = None
self.result_y = None
def groupIntoBands(self, fft_data, fft_frequency, band_num):
bands = None
if band_num == 5:
bands = [0.5, 4, 8, 15, 30, 128]
if band_num == 8:
bands = [0.1, 4, 8, 12, 30, 50, 70, 100, 180]
frequency_bands = numpy.digitize(fft_frequency, bands)
if fft_data.ndim > 1:
channels = fft_data.shape[0]
result = []
for i in xrange(channels):
data_frame = pandas.DataFrame({"fft":fft_data[i], "band":frequence_bands})
data_frame = data_frame.groupby("band").mean()
result.append(data_frame.fft[1: -1])
return result
data_frame = pandas.DataFrame({"fft":fft_data, "band": frequency_bands})
data_frame = data_frame.groupby("band").mean()
return data_frame.fft[1: -1]
def fft(self, data_x, data_y, band_num = 8, sampling_rate = 400, window_length = 30, stride = 30):
#data_x's shape is matFileNumber * channels * matdata
channels = data_x.shape[1]
data_length = data_x.shape[2] / sampling_rate
steps = (data_length - window_length) / stride + 1
new_array = numpy.zeros((data_x.shape[0], channels, int(band_num + 1), int(steps)))
size = data_x.shape[0]
for i in range(size):
for j in range(channels):
for frame_index, window_index in enumerate(range(0, int(data_length - window_length + 1), stride)):
data = data_x[i, j, window_index * sampling_rate:(window_index + window_length) * sampling_rate]
fft_data = numpy.log10(numpy.absolute(numpy.fft.rfft(data)))
fft_frequency = numpy.fft.rfftfreq(n = data.shape[-1], d = 1.0 / sampling_rate)
new_array[i, j, :band_num, frame_index] = self.groupIntoBands(fft_data, fft_frequency, band_num = 8)
new_array[i, j, -1, frame_index] = numpy.std(data)
self.result_x = new_array
self.result_y = data_y
return new_array, data_y
def saveToDisk(self, feature_name, name, is_train):
assert self.result_x is None or self.result_y is None
save_path = self.setting.processedDataPath + feature_name
if is_train:
os.makedirs(os.path.join(save_path, self.setting.name), exist_ok=True)
numpy.save(os.path.join(save_path, self.setting.name, str(name) + "_trainX"), self.result_x)
numpy.save(os.path.join(save_path, self.setting.name, str(name) + "_trainY"), self.result_y)
else:
os.makedirs(os.path.join(save_path, self.setting.name), exist_ok=True)
numpy.save(os.path.join(save_path, self.setting.name, str(name) + "_testX"), self.result_x)
numpy.save(os.path.join(save_path, self.setting.name, str(name) + "_testY"), self.result_y)
def loadFromDisk(self, feature_name, is_train):
save_path = self.setting.processedDataPath + feature_name
files = None
if is_train:
files = glob.glob(os.path.join(save_path, self.setting.name, "*trainX.npy"))
else:
files = glob.glob(os.path.join(save_path, self.setting.name, "*testX.npy"))
files = sorted(files)
trainX = None
trainY = None
testX = None
testY = None
if is_train:
trainX = numpy.load(files[0])
file_name = files[0].replace("trainX", "trainY")
trainY = numpy.load(file_name)
else:
testX = numpy.load(files[0])
file_name = files[0].replace("testX", "testY")
testY = numpy.load(file_name)
for f in files[1:]:
if is_train:
tmp = numpy.load(f)
trainX = numpy.concatenate((trainX, numpy.load(f)), axis = 0)
f = f.replace("trainX", "trainY")
trainY = numpy.concatenate((trainY, numpy.load(f)), axis = 0)
else:
testX = numpy.concatenate((testX, numpy.load(f)), axis = 0)
f = f.replace("testX", "testY")
testY = numpy.concatenate((testY, numpy.load(f)), axis = 0)
if is_train:
self.result_x = trainX
self.result_y = trainY
return trainX, trainY
else:
self.result_x = testX
self.result_y = testY
return testX, testY
def overlap(self, x, y):
shape_x = x.shape
shape_y = y.shape
zero_indics = numpy.where(y == 0)
one_indics = numpy.where(y == 1)
zero_indics = numpy.array(zero_indics[0]).tolist()
one_indics = numpy.array(one_indics[0]).tolist()
first_part = int(shape_x[3] / 2)
numpy.newaxis
tmp_x = numpy.concatenate((x[zero_indics[0], :, :, :first_part], x[zero_indics[1], :, :, first_part:]), axis = 2)
tmp_x = tmp_x.reshape(1, tmp_x.shape[0], tmp_x.shape[1], tmp_x.shape[2])
tmp_y = []
tmp_y.append(0)
for i in range(1, len(zero_indics) - 1):
if (i + 1) % 6 != 0:
tmp = numpy.concatenate((x[zero_indics[i], :, :, :first_part], x[zero_indics[i + 1], :, :, first_part:]), axis = 2)
tmp = tmp.reshape(1, tmp.shape[0], tmp.shape[1], tmp.shape[2])
tmp_x = numpy.concatenate((tmp_x, tmp), axis = 0)
tmp_y.append(0)
for i in range(len(one_indics) - 1):
if(i + 1) % 6 != 0:
tmp = numpy.concatenate((x[one_indics[i], :, :, :first_part], x[one_indics[i + 1], :, :, first_part:]), axis = 2)
tmp = tmp.reshape(1, tmp.shape[0], tmp.shape[1], tmp.shape[2])
tmp_x = numpy.concatenate((tmp_x, tmp), axis = 0)
tmp_y.append(1)
x = numpy.concatenate((x, tmp_x), axis = 0)
y = numpy.concatenate((y, numpy.asarray(tmp_y)), axis = 0)
return x, y
def pca(self, data_x, data_y, band_num = 8, sampling_rate = 400, window_length = 30, stride = 30):
#data_x's shape is mat_file_number * channels * mat_data
channels = data_x.shape[1]
data_length = data_x.shape[2] / sampling_rate
steps = (data_length - window_length) / stride + 1
new_array = numpy.zeros((data_x.shape[0], channels, channels, int(steps)))
pca = sklearn.decomposition.PCA()
size = data_x.shape[0]
for i in range(size):
for frame_index, window_index, in enumerate(range(0, int(data_length - window_length + 1), stride)):
data = data_x[i, :, window_index * sampling_rate:(window_index + window_length) * sampling_rate]
pca_data = pca.fit_transform(data)
pca_data = numpy.log10(numpy.absolute(pca_data))
new_array[i, :, :, frame_index] = pca_data
self.result_x = new_array
self.result_y = data_y
return self.result_x, self.result_y
def shuffle(self,dataX1, dataX2, dataY):
resultArray = zip(dataX1, dataX2, dataY)
random.shuffle(resultArray)
tempX1, tempX2, tempY = zip(*resultArray)
tempX1 = numpy.array(tempX1, dtype="float32")
tempX2 = numpy.array(tempX2, dtype="float32")
tempY = numpy.array(tempY, dtype="int8")
return tempX1, tempX2, tempY
def scaleAcrossTime(self, data, scalers = None):
sample_num = data.shape[0]
channel_num = data.shape[1]
bin_num = data.shape[2]
time_step_num = data.shape[3]
if scalers is None:
scalers = [None] * channel_num
for i in range(channel_num):
dataI = numpy.transpose(data[:, i, :, :], axes = (0, 2, 1))
dataI = dataI.reshape((sample_num * time_step_num, bin_num))
if scalers[i] is None:
scalers[i] = sklearn.preprocessing.StandardScaler()
scalers[i].fit(dataI)
dataI = scalers[i].transform(dataI)
dataI = dataI.reshape((sample_num, time_step_num, bin_num))
dataI = numpy.transpose(dataI, axes = (0, 2, 1))
data[:, i, :, :] = dataI
return data, scalers
if __name__ == "__main__":
pass