-
Notifications
You must be signed in to change notification settings - Fork 17
/
augmentation.py
269 lines (236 loc) · 12.1 KB
/
augmentation.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
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
import os
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
import tensorflow as tf
import numpy as np
from dataset import masked_timeseries_dataset, varying_context_timeseries_dataset
from tensorflow.keras.utils import timeseries_dataset_from_array
def scaling(x, sigma=1.1):
'''
Apply same factor of scaling for each dimension in whole window batch
:param X:
:return:
'''
# https://arxiv.org/pdf/1706.00527.pdf
factor = tf.random.normal(mean=2, stddev=sigma, shape=(x.shape[0], 1, x.shape[2]))
factor = tf.tile(factor,[1,x.shape[1],1])
factor = tf.cast(factor,tf.float64)
return tf.math.multiply(x,factor)
def jittering(x, sigma=0.8):
return x + tf.random.normal(mean=0, stddev=sigma, shape=x.shape, dtype=tf.float64)
class SingleWindow():
'''
Windowing and extract receptive area.
'''
def __init__(self, window_length, receptive_length):
self.window_length = window_length
self.receptive_length = receptive_length
self.num_overlap = 1
def sample_first_regions(self, y, label_length, n, seed):
np.random.seed(seed=0)
# np.random.seed(seed=seed)
num_class = len(np.unique(y))
mask = np.zeros_like(y)
y_short = y[self.window_length // 2 + 1:-self.window_length // 2 - 1]
# generate classwise first timestamp list for each segment where first timestamp + length does not cover boundary
first_timestamp_list = []
for i in range(num_class):
y_short_class_ind = np.where(y_short == i)[0]
boundaries = []
segment_len = []
length = 0
for ind, (prev_ind, curr_ind) in enumerate(zip(y_short_class_ind, y_short_class_ind[1:])):
length += 1
if curr_ind != prev_ind + 1:
boundaries.append(curr_ind)
segment_len.append(length)
length = 0
segment_len.append(length + 1) # append last segment length
boundaries.insert(0, y_short_class_ind[0]) # append first segment start position
# make start ts from boundary start to boundary end (= start + length)
boundaries = np.array(boundaries)
ts_list = []
# for ind_j, j in enumerate(boundaries):
for ind_j, j in enumerate(boundaries[np.where(np.array(segment_len) > label_length)[0]]):
ts_list += list(range(j, j + segment_len[ind_j] - label_length, label_length))
# print(f"ts_list:{ts_list}")
if len(ts_list) < n:
print(f"Not enough segments, ts_list length:{len(ts_list)}, number of sampled seg:{n}")
ts_list = np.array(ts_list) + self.window_length // 2
spld_indice = np.random.choice(ts_list, size=n, replace=False).tolist()
first_timestamp_list += spld_indice
center_timestamps = []
for ts in first_timestamp_list:
mask[ts:ts + label_length] = 1
center_timestamps.append(ts + label_length//2)
np.random.seed()
return mask, center_timestamps
def sample_first_timestamp(self, y, n, seed):
'''
Args:
x: input time series
y: timestamp labels
n: number of timestamps for each class
Returns:
mask vector for timestamp-wise classification. If mask=1, label exists and back propagation occurs at the
timestamp.
'''
if n > 0:
num_class = len(np.unique(y))
mask = np.zeros_like(y)
y_short = y[self.window_length // 2 + 1:-self.window_length // 2 - 1]
np.random.seed(seed=0) # fix sampled timestamp for each dataset. change 0 to seed when randomizing initial labels.
for i in range(num_class):
indice = np.where(y_short==i)[0]
# print(indice,y,y_short)
if len(indice) < n:
print(f"number of timestamp label for class {i} is less than the number of required labels {n}")
spld_indice = np.random.choice(indice,size=n,replace=False)
spld_indice += self.window_length//2
# print(spld_indice, y[spld_indice])
mask[spld_indice]=1
np.random.seed()
else:
mask = np.ones_like(y)
print("full labels for each class are used")
return mask
def dataloader(self, X_mask_y, batch_size, mask=[], center_timestamps=[]):
if len(mask) < 1:
dataset = timeseries_dataset_from_array(data=X_mask_y, targets=None,
sequence_length=self.window_length,
sequence_stride=self.receptive_length, # self.overlap_length
start_index=self.receptive_length // 2,
end_index=X_mask_y.shape[0]-self.receptive_length // 2,
shuffle=True, batch_size=batch_size)
else:
dataset = masked_timeseries_dataset(data=X_mask_y, targets=None, mask=mask,
sequence_length=self.window_length, sequence_stride=self.receptive_length,
shuffle=True, batch_size=batch_size, center_timestamps=center_timestamps)
return dataset
def extract_overlap(self, output):
start = self.window_length//2-self.receptive_length//2
end = self.window_length//2+self.receptive_length//2
output = tf.gather(output, tf.range(start, end), axis=1)
return output
class Overlap():
def __init__(self, window_length, overlap_length, start_position=None, end_position=None):
self.window_length = window_length
self.overlap_length = overlap_length
if start_position == None:
self.start_position = window_length-overlap_length
else:
self.start_position = start_position
if end_position == None:
self.end_position = window_length
else:
self.end_position = end_position
self.total_length = self.window_length * 2 - self.overlap_length
self.num_overlap = 2
def sample_first_regions(self, y, label_length, n, seed):
np.random.seed(seed=0)
# np.random.seed(seed=seed)
num_class = len(np.unique(y))
mask = np.zeros_like(y)
y_short = y[self.total_length // 2 + 1:-self.total_length]
print(y.shape, np.unique(y))
first_timestamp_list = []
for i in range(num_class):
y_short_class_ind = np.where(y_short == i)[0]
boundaries = []
segment_len = []
length = 0
for ind, (prev_ind, curr_ind) in enumerate(zip(y_short_class_ind, y_short_class_ind[1:])):
length += 1
if curr_ind != prev_ind + 1:
boundaries.append(curr_ind)
segment_len.append(length)
length = 0
segment_len.append(length + 1) # append last segment length
boundaries.insert(0, y_short_class_ind[0]) # append first segment start position
# make start ts from boundary start to boundary end (= start + length)
boundaries = np.array(boundaries)
ts_list = []
for ind_j, j in enumerate(boundaries[np.where(np.array(segment_len) > label_length)[0]]):
ts_list += list(range(j, j + segment_len[ind_j] - label_length, label_length))
# print(f"ts_list:{ts_list}")
if len(ts_list) < n:
raise ValueError("Not enough segments")
ts_list = np.array(ts_list) + self.total_length // 2
spld_indice = np.random.choice(ts_list, size=n, replace=False).tolist()
first_timestamp_list += spld_indice
center_timestamps = []
for ts in first_timestamp_list:
mask[ts:ts + label_length] = 1
center_timestamps.append(ts + label_length//2)
np.random.seed()
return mask, center_timestamps
def sample_first_timestamp(self, y, n, seed):
'''
Args:
x: input time series
y: timestamp labels
n: number of timestamps for each class
Returns:
mask vector for timestamp-wise classification. If mask=1, label exists and back propagation occurs at the
timestamp.
'''
if n > 0:
num_class = len(np.unique(y))
mask = np.zeros_like(y)
y_short = y[self.total_length // 2:-self.total_length // 2]
np.random.seed(seed=0) # fix sampled timestamp for each dataset. change 0 to seed when randomizing initial labels.
for i in range(num_class):
indice = np.where(y_short==i)[0]
# print(indice,y,y_short)
if len(indice) < n:
print(f"number of timestamp label for class {i} is less than the number of required labels {n}")
spld_indice = np.random.choice(indice,size=n,replace=False)
spld_indice += self.total_length//2
# print(spld_indice, y[spld_indice])
mask[spld_indice]=1
np.random.seed()
else:
mask = np.ones_like(y)
print("full labels for each class are used")
return mask
def dataloader(self, X_mask_y, batch_size, num_iter, mask=[], center_timestamps=[]):
if len(mask) < 1:
dataset = varying_context_timeseries_dataset(data=X_mask_y,
targets=None,
overlap_length=self.overlap_length,
sequence_length=self.total_length, # maximum window length = overlap_length + 2 * context_length
batch_size=batch_size,
iterations=num_iter)
else:
dataset = masked_timeseries_dataset(data=X_mask_y, targets=None, mask=mask,
sequence_length=self.total_length, sequence_stride=self.overlap_length,
shuffle=True, batch_size=batch_size, center_timestamps=center_timestamps)
return dataset
def windowing(self, batch):
# print(self.window_length, self.overlap_length, self.total_length)
left = tf.gather(batch, tf.range(self.window_length), axis=1)
right = tf.gather(batch, tf.range(self.window_length-self.overlap_length, self.total_length), axis=1)
return tf.concat([left, right], axis=0)
def extract_overlap(self, output):
output_left, output_right = tf.split(output, 2, axis=0)
output_left = tf.gather(output_left, tf.range(self.window_length-self.overlap_length, self.window_length), axis=1)
output_right = tf.gather(output_right, tf.range(0,self.overlap_length), axis=1)
output = tf.concat([output_left, output_right], axis=0)
return output
def extract_rest(self, output):
output_left, output_right = tf.split(output, 2, axis=0)
output_left = tf.gather(output_left, tf.range(0, self.window_length-self.overlap_length), axis=1)
output_right = tf.gather(output_right, tf.range(self.overlap_length,self.window_length), axis=1)
return output_left, output_right
if __name__ == "__main__":
batch = 2
dim = 3
window_length = 8
overlap_length = 2
total_length = window_length*2-overlap_length
data = tf.range(batch*total_length*dim)
data = tf.reshape(data,(batch,total_length,dim))
print(data)
o = OverlapN(window_length,overlap_length)
print(o.windowing(data))
print(o.extract_overlap(o.windowing(data)))