forked from ziyujia/ECML-PKDD_MMCNN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathMMCNN_model.py
283 lines (251 loc) · 12 KB
/
MMCNN_model.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
270
271
272
273
274
275
276
277
278
279
280
281
282
283
from keras.layers import Add
from keras.layers.convolutional import Convolution1D, MaxPooling1D, ZeroPadding1D, AveragePooling1D
from keras.layers.core import Dense, Activation, Flatten
from keras.layers import Input
from keras.layers import Lambda
from keras.layers import Reshape
from keras.layers import AveragePooling1D
from keras.layers import GlobalAveragePooling1D
from keras.layers import Dense
from keras.layers import multiply
import tensorflow as tf
from keras import layers
from keras import Input
from keras.models import Model
from keras import regularizers
from keras.layers import BatchNormalization
from keras import metrics
from keras import losses
from keras.utils import to_categorical
from tensorflow.keras.callbacks import EarlyStopping
from keras.layers import merge
from sklearn.metrics import confusion_matrix
from sklearn.metrics import f1_score
from keras import optimizers
class MMCNN_model():
def __init__(self,
channels=3,
samples=1000):
self.channels = 3
self.samples = 1000
self.activation = 'elu'
self.learning_rate = 0.0001
self.dropout = 0.8
self.inception_filters = [16, 16, 16, 16]
# the parameter of the first part :EEG Inception block
self.inception_kernel_length = [[5, 10, 15, 10],
[40, 45, 50, 100],
[60, 65, 70, 100],
[80, 85, 90, 100],
[160, 180, 200, 180], ]
self.inception_stride = [2, 4, 4, 4, 16]
self.first_maxpooling_size = 4
self.first_maxpooling_stride = 4
# the parameter of the second part :Residual block
self.res_block_filters = [16, 16, 16]
self.res_block_kernel_stride = [8, 7, 7, 7, 6]
# the parameter of the third part :SE block
self.se_block_kernel_stride = 16
self.se_ratio = 8
self.second_maxpooling_size = [4, 3, 3, 3, 2]
self.second_maxpooling_stride = [4, 3, 3, 3, 2]
self.model = self.build_model(self.channels, self.samples)
adam = optimizers.Adam(lr=self.learning_rate)
self.model.compile(loss=losses.binary_crossentropy,
optimizer=adam,
metrics=['mae', metrics.binary_accuracy])
def build_model(self, channels, samples):
output_conns = []
input_tensor = Input(shape=(samples, channels))
'''
EIN-a
'''
x = self.inception_block(input_tensor,
self.inception_filters,
self.inception_kernel_length[0],
self.inception_stride[0],
self.activation)
x = layers.MaxPooling1D(pool_size=self.first_maxpooling_size,
strides=self.first_maxpooling_stride,
padding='same')(x)
x = BatchNormalization()(x)
x = layers.core.Dropout(self.dropout)(x)
x = self.conv_block(x,
self.res_block_filters,
self.res_block_kernel_stride[0],
activation=self.activation)
x = self.squeeze_excitation_layer(x,
self.se_block_kernel_stride,
self.activation,
ratio=self.se_ratio)
x = layers.MaxPooling1D(pool_size=self.second_maxpooling_size[0],
strides=self.second_maxpooling_stride[0],
padding='same')(x)
x = layers.core.Flatten()(x)
output_conns.append(x)
'''
EIN-b
'''
y1 = self.inception_block(input_tensor,
self.inception_filters,
self.inception_kernel_length[1],
self.inception_stride[1],
self.activation)
y1 = layers.MaxPooling1D(pool_size=self.first_maxpooling_size,
strides=self.first_maxpooling_stride,
padding='same')(y1)
y1 = BatchNormalization()(y1)
y1 = layers.core.Dropout(self.dropout)(y1)
y1 = self.conv_block(y1,
self.res_block_filters,
self.res_block_kernel_stride[1],
self.activation)
y1 = self.squeeze_excitation_layer(y1,
self.se_block_kernel_stride,
self.activation,
ratio=self.se_ratio)
y1 = layers.MaxPooling1D(pool_size=self.second_maxpooling_size[1],
strides=self.second_maxpooling_stride[1],
padding='same')(y1)
y1 = layers.core.Flatten()(y1)
output_conns.append(y1)
'''
EIN-c
'''
y2 = self.inception_block(input_tensor,
self.inception_filters,
self.inception_kernel_length[2],
self.inception_stride[2],
self.activation)
y2 = layers.MaxPooling1D(pool_size=self.first_maxpooling_size,
strides=self.first_maxpooling_stride,
padding='same')(y2)
y2 = BatchNormalization()(y2)
y2 = layers.core.Dropout(self.dropout)(y2)
y2 = self.conv_block(y2,
self.res_block_filters,
self.res_block_kernel_stride[2],
self.activation)
y2 = self.squeeze_excitation_layer(y2,
self.se_block_kernel_stride,
self.activation,
ratio=self.se_ratio)
y2 = layers.MaxPooling1D(pool_size=self.second_maxpooling_size[2],
strides=self.second_maxpooling_stride[2],
padding='same')(y2)
y2 = layers.core.Flatten()(y2)
output_conns.append(y2)
'''
EIN-d
'''
y3 = self.inception_block(input_tensor,
self.inception_filters,
self.inception_kernel_length[3],
self.inception_stride[3],
self.activation)
y3 = layers.MaxPooling1D(pool_size=self.first_maxpooling_size,
strides=self.first_maxpooling_stride,
padding='same')(y3)
y3 = BatchNormalization()(y3)
y3 = layers.core.Dropout(self.dropout)(y3)
y3 = self.conv_block(y3,
self.res_block_filters,
self.res_block_kernel_stride[3],
self.activation)
y3 = self.squeeze_excitation_layer(y3,
self.se_block_kernel_stride,
self.activation,
ratio=self.se_ratio)
y3 = layers.MaxPooling1D(pool_size=self.second_maxpooling_size[3],
strides=self.second_maxpooling_stride[3],
padding='same')(y3)
y3 = layers.core.Flatten()(y3)
output_conns.append(y3)
'''
EIN-e
'''
z = self.inception_block(input_tensor,
self.inception_filters,
self.inception_kernel_length[4],
self.inception_stride[4],
self.activation)
z = layers.MaxPooling1D(pool_size=self.first_maxpooling_size,
strides=self.first_maxpooling_stride,
padding='same')(z)
z = BatchNormalization()(z)
z = layers.core.Dropout(self.dropout)(z)
z = self.conv_block(z,
self.res_block_filters,
self.res_block_kernel_stride[4],
self.activation)
z = self.squeeze_excitation_layer(z,
self.se_block_kernel_stride,
self.activation,
ratio=self.se_ratio)
z = layers.MaxPooling1D(pool_size=self.second_maxpooling_size[4],
strides=self.second_maxpooling_stride[4],
padding='same')(z)
z = layers.core.Flatten()(z)
output_conns.append(z)
output_conns = layers.Concatenate(axis=-1)(output_conns)
output_conns = layers.core.Dropout(self.dropout)(output_conns)
output_tensor = layers.Dense(2, activation='sigmoid')(output_conns)
# output_tensor = layers.Dense(4,activation = 'softmax')(output_conns)
model = Model(input_tensor, output_tensor)
return model
'''
se_block
'''
def squeeze_excitation_layer(self, x, out_dim, activation, ratio=8):
squeeze = GlobalAveragePooling1D()(x)
excitation = Dense(units=out_dim // ratio)(squeeze)
excitation = Activation(activation)(excitation)
excitation = Dense(units=out_dim, activation='sigmoid')(excitation)
excitation = Reshape((1, out_dim))(excitation)
scale = multiply([x, excitation])
return scale
'''
res_block
'''
def conv_block(self, x, nb_filter, length, activation):
k1, k2, k3 = nb_filter
out = layers.Conv1D(k1, length, strides=1, padding='same', kernel_regularizer=regularizers.l2(0.002))(x)
out = BatchNormalization()(out)
out = Activation(activation)(out)
out = layers.Conv1D(k2, length, strides=1, padding='same', kernel_regularizer=regularizers.l2(0.002))(out)
out = BatchNormalization()(out)
out = Activation(activation)(out)
out = layers.Conv1D(k3, length, strides=1, padding='same', kernel_regularizer=regularizers.l2(0.002))(out)
out = BatchNormalization()(out)
x = layers.Conv1D(k3, 1, strides=1, padding='same')(x)
x = BatchNormalization()(x)
out = Add()([out, x])
out = Activation(activation)(out)
out = layers.core.Dropout(self.dropout)(out)
return out
'''
inception_block
'''
def inception_block(self, x, ince_filter, ince_length, stride, activation):
k1, k2, k3, k4 = ince_filter
l1, l2, l3, l4 = ince_length
inception = []
x1 = layers.Conv1D(k1, l1, strides=stride, padding='same', kernel_regularizer=regularizers.l2(0.01))(x)
x1 = BatchNormalization()(x1)
x1 = Activation(activation)(x1)
inception.append(x1)
x2 = layers.Conv1D(k2, l2, strides=stride, padding='same', kernel_regularizer=regularizers.l2(0.01))(x)
x2 = BatchNormalization()(x2)
x2 = Activation(activation)(x2)
inception.append(x2)
x3 = layers.Conv1D(k3, l3, strides=stride, padding='same', kernel_regularizer=regularizers.l2(0.01))(x)
x3 = BatchNormalization()(x3)
x3 = Activation(activation)(x3)
inception.append(x3)
x4 = layers.MaxPooling1D(pool_size=l4, strides=stride, padding='same')(x)
x4 = layers.Conv1D(k4, 1, strides=1, padding='same')(x4)
x4 = BatchNormalization()(x4)
x4 = Activation(activation)(x4)
inception.append(x4)
v1 = layers.Concatenate(axis=-1)(inception)
return v1