-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmodel.py
91 lines (86 loc) · 6.15 KB
/
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
import tensorflow as tf
from tensorflow.keras.models import Sequential, Model
from tensorflow.keras import regularizers
from tensorflow.keras.layers import Flatten, Dense, Dropout, SpatialDropout1D, LSTM, Conv1D, MaxPooling1D, GlobalAveragePooling1D, BatchNormalization, GlobalMaxPooling1D, Bidirectional, LSTM, GRU, Input
from keras.layers.merge import concatenate
# n_timesteps=180
# n_features=69
n_timesteps=8
n_features=17
#['AU01_r_std', 'AU02_r_std', 'AU04_r_std', 'AU05_r_std', 'AU06_r_std', 'AU07_r_std', 'AU09_r_std', 'AU10_r_std', 'AU12_r_std', 'AU14_r_std', 'AU15_r_std', 'AU17_r_std', 'AU20_r_std', 'AU23_r_std', 'AU25_r_std', 'AU26_r_std', 'AU45_r_std', 'AU01_r_variation', 'AU02_r_variation', 'AU04_r_variation', 'AU05_r_variation', 'AU06_r_variation', 'AU07_r_variation', 'AU09_r_variation', 'AU10_r_variation', 'AU12_r_variation', 'AU14_r_variation', 'AU15_r_variation', 'AU17_r_variation', 'AU20_r_variation', 'AU23_r_variation', 'AU25_r_variation', 'AU26_r_variation', 'AU45_r_variation', 'AU01_r_max', 'AU02_r_max', 'AU04_r_max', 'AU05_r_max', 'AU06_r_max', 'AU07_r_max', 'AU09_r_max', 'AU10_r_max', 'AU12_r_max', 'AU14_r_max', 'AU15_r_max', 'AU17_r_max', 'AU20_r_max', 'AU23_r_max', 'AU25_r_max', 'AU26_r_max', 'AU45_r_max', 'AU01_c_frequency', 'AU02_c_frequency', 'AU04_c_frequency', 'AU05_c_frequency', 'AU06_c_frequency', 'AU07_c_frequency', 'AU09_c_frequency', 'AU10_c_frequency', 'AU12_c_frequency', 'AU14_c_frequency', 'AU15_c_frequency', 'AU17_c_frequency', 'AU20_c_frequency', 'AU23_c_frequency', 'AU25_c_frequency', 'AU26_c_frequency', 'AU28_c_frequency', 'AU45_c_frequency']
def Conv1D_model(n_timesteps, n_features, n_outputs=1):
model = Sequential()
model.add(Conv1D(64, 7, input_shape=(n_timesteps, n_features), activation='relu', padding='same', name='con1'))
model.add(MaxPooling1D(pool_size=2, padding='same', strides=2, name='mp1'))
model.add(BatchNormalization(name='bn1'))
model.add(Conv1D(128, 7, padding='same', activation='relu', name='con2'))
model.add(MaxPooling1D(pool_size=2, padding='same', strides=2, name='mp2'))
model.add(BatchNormalization(name='bn2'))
model.add(Conv1D(256, 7, padding='same', activation='relu', name='con3'))
model.add(MaxPooling1D(pool_size=2, strides=2, padding='same', name='mp3'))
# model.add(Conv1D(2, 5, padding='same', activation='relu'))
# model.add(LSTM(32))
model.add(GlobalAveragePooling1D(name='gap'))
# model.add(GlobalMaxPooling1D())
# model.add(Flatten())
model.add(Dense(512, activation='relu', name='fc'))
model.add(Dropout(0.5, name='drop'))
model.add(Dense(n_outputs, activation='sigmoid', kernel_regularizer=regularizers.l2(0.05), name='output'))
# model.compile(optimizer=tf.keras.optimizers.Adam(lr=1e-3), loss='binary_crossentropy', metrics=['mse'])
return model
def ConcatCNN_model(n_timesteps, n_features, n_outputs=1):
inputs = Input(shape=(n_timesteps, n_features))
# 卷积层和池化层，设置卷积核大小分别为3,4,5
cnn1 = Conv1D(256, 3, padding='same', strides=1, activation='relu', name='con1')(inputs)
cnn1 = MaxPooling1D(pool_size=8, name='mp1')(cnn1)
cnn2 = Conv1D(256, 4, padding='same', strides=1, activation='relu', name='con2')(inputs)
cnn2 = MaxPooling1D(pool_size=4, name='mp2')(cnn2)
cnn3 = Conv1D(256, 5, padding='same', strides=1, activation='relu', name='con3')(inputs)
cnn3 = MaxPooling1D(pool_size=2, name='mp3')(cnn3)
# 合并三个模型的输出向量
cnn = concatenate([cnn1, cnn2, cnn3], axis=1)
flat = Flatten(name='flat')(cnn)
#加上dropout防止过拟合
drop = Dropout(0.5, name='drop')(flat)
main_output = Dense(n_outputs, activation='sigmoid', name='output')(drop)
model = Model(inputs=inputs, outputs=main_output)
return model
def ConcatCNN_SAMM_model(n_timesteps, n_features, n_outputs=1):
inputs = Input(shape=(n_timesteps, n_features))
# 卷积层和池化层，设置卷积核大小分别为3,4,5
cnn1 = Conv1D(256, 5, padding='same', strides=2, activation='relu', name='con1')(inputs)
cnn1 = MaxPooling1D(pool_size=8, name='mp1')(cnn1)
cnn2 = Conv1D(256, 7, padding='same', strides=2, activation='relu', name='con2')(inputs)
cnn2 = MaxPooling1D(pool_size=4, name='mp2')(cnn2)
cnn3 = Conv1D(256, 9, padding='same', strides=2, activation='relu', name='con3')(inputs)
cnn3 = MaxPooling1D(pool_size=2, name='mp3')(cnn3)
# 合并三个模型的输出向量
cnn = concatenate([cnn1, cnn2, cnn3], axis=1)
flat = Flatten(name='flat')(cnn)
#加上dropout防止过拟合
drop = Dropout(0.5, name='drop')(flat)
main_output = Dense(n_outputs, activation='sigmoid', name='output')(drop)
model = Model(inputs=inputs, outputs=main_output)
return model
def LSTM_model(n_timesteps, n_features, n_outputs=1, n_gru_units=128, n_dense_units=512):
model = Sequential()
model.add(Bidirectional(LSTM(n_gru_units, return_sequences=True, name='lstm1'), input_shape=(n_timesteps, n_features)))
model.add(Bidirectional(LSTM(n_gru_units, return_sequences=True, name='lstm2')))
model.add(GlobalAveragePooling1D(name='gap'))
model.add(BatchNormalization(name='bn'))
model.add(Dense(n_dense_units, activation='relu', name='fc'))
model.add(Dropout(0.5, name='dropout'))
model.add(Dense(n_outputs, activation='sigmoid', kernel_regularizer=regularizers.l2(0.05), name='output'))
# model.compile(optimizer=tf.keras.optimizers.Adam(lr=1e-3), loss='binary_crossentropy', metrics=['mse'])
return model
def GRU_model(n_timesteps, n_features, n_outputs=1, n_gru_units=128, n_dense_units=512):
model = Sequential()
model.add(Bidirectional(GRU(n_gru_units, return_sequences=True, name='gru1'), input_shape=(n_timesteps, n_features)))
model.add(Bidirectional(GRU(n_gru_units, return_sequences=True), name='gru2'))
model.add(GlobalAveragePooling1D(name='gap'))
model.add(BatchNormalization(name='bn'))
model.add(Dense(n_dense_units, activation='relu', name='fc'))
model.add(Dropout(0.5))
model.add(Dense(n_outputs, activation='sigmoid', kernel_regularizer=regularizers.l2(0.05), name='output'))
# model.compile(optimizer=tf.keras.optimizers.Adam(lr=1e-3), loss='binary_crossentropy', metrics=['mse'])
return model