-
Notifications
You must be signed in to change notification settings - Fork 4
/
training.py
121 lines (103 loc) · 5.24 KB
/
training.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
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from keras import backend
def rmse(y_true, y_pred):
return backend.sqrt(backend.mean(backend.square(y_pred - y_true), axis=-1))
#======read data===================================================
Y=np.load(r"Total_Y.npy")
graph_conv_filter=preprocess_adj_tensor_with_identity(matrix) #matrix:edge weight of graph.
#we utilize cosine similarity to bulid the relationship between each sensors.
#X_train to X_train5 means 6 past steps data for all the sensors, respectively.
X_train=final[0:train_size]
X_test=final[train_size:]
X_train1=final1[0:train_size]
X_test1=final1[train_size:]
X_train2=final2[0:train_size]
X_test2=final2[train_size:]
X_train3=final3[0:train_size]
X_test3=final3[train_size:]
X_train4=final4[0:train_size]
X_test4=final4[train_size:]
X_train5=final5[0:train_size]
X_test5=final5[train_size:]
y_train=Y[0:train_size]
y_test=Y[train_size:]
cnnX_train=finalcnn[0:train_size]
cnnX_test=finalcnn[train_size:]
MX_train=graph_conv_filter[0:train_size]
MX_test=graph_conv_filter[train_size:]
tX_train=finalcnn[0:train_size]
tX_test=finalcnn[train_size:]
#===========train model================================
import keras
from keras.models import Input, Model, Sequential
from keras.layers import Dense, Activation, Dropout, Lambda,Flatten,LSTM,AveragePooling1D,GRU,Conv2D,AveragePooling2D,GRU,Multiply
import keras.backend as K
from sklearn.utils import shuffle
past_steps=6
number_of_sensors=26
num_filters = 2
graph_conv_filters_input = Input(shape=(number_of_sensors*2, number_of_sensors))
tinput=Input(shape=(past_steps,number_of_sensors))
t=timeattention(number_of_sensors)(tinput)
t=Flatten()(t)
#X_input1 to X_input6 means that there are 6 past_steps,respectively.
X_input1 = Input(shape=(number_of_sensors, 1))
X_input2 = Input(shape=(number_of_sensors, 1))
X_input3 = Input(shape=(number_of_sensors, 1))
X_input4 = Input(shape=(number_of_sensors, 1))
X_input5 = Input(shape=(number_of_sensors, 1))
X_input6 = Input(shape=(number_of_sensors, 1))
output1 = MultiGraphCNN(1, num_filters)([X_input1, graph_conv_filters_input])
output1 = MultiGraphCNN(1, num_filters)([output1, graph_conv_filters_input])
output1 = MultiGraphCNN(1, num_filters)([output1, graph_conv_filters_input])
output1=Flatten()(output1)
output2 = MultiGraphCNN(1, num_filters)([X_input2, graph_conv_filters_input])
output2 = MultiGraphCNN(1, num_filters)([output2, graph_conv_filters_input])
output2 = MultiGraphCNN(1, num_filters)([output2, graph_conv_filters_input])
output2=Flatten()(output2)
output3 = MultiGraphCNN(1, num_filters)([X_input3, graph_conv_filters_input])
output3 = MultiGraphCNN(1, num_filters)([output3, graph_conv_filters_input])
output3 = MultiGraphCNN(1, num_filters)([output3, graph_conv_filters_input])
output3=Flatten()(output3)
output4 = MultiGraphCNN(1, num_filters)([X_input4, graph_conv_filters_input])
output4 = MultiGraphCNN(1, num_filters)([output4, graph_conv_filters_input])
output4 = MultiGraphCNN(1, num_filters)([output4, graph_conv_filters_input])
output4=Flatten()(output4)
output5 = MultiGraphCNN(1, num_filters)([X_input5, graph_conv_filters_input])
output5 = MultiGraphCNN(1, num_filters)([output5, graph_conv_filters_input])
output5 = MultiGraphCNN(1, num_filters)([output5, graph_conv_filters_input])
output5=Flatten()(output5)
output6 = MultiGraphCNN(1, num_filters)([X_input6, graph_conv_filters_input])
output6 = MultiGraphCNN(1, num_filters)([output6, graph_conv_filters_input])
output6 = MultiGraphCNN(1, num_filters)([output6, graph_conv_filters_input])
output6=Flatten()(output6)
con=keras.layers.concatenate([output1,output2,output3,output4,output5,output6])
con2=keras.layers.Reshape((past_steps,number_of_sensors))(con)
con2=keras.layers.Reshape((past_steps,number_of_sensors,1))(con2)
#cnn
cnnoutput=Conv2D(1,3,1)(con2)
maxpooling=AveragePooling2D((4,1))(cnnoutput)
cnnoutput=Flatten()(maxpooling)
cnninput=Input(shape=(past_steps,number_of_sensors))
#LSTM
output=LSTM(number_of_sensors)(cnninput)
output1=Dense(number_of_sensors)(output)
#Learning attention weight for adjustment
output=keras.layers.average([output1,cnnoutput])
output=Multiply()([output,t])
nb_epochs = 50
batch_size = 32
model1 = Model(inputs=[X_input1,X_input2,X_input3,X_input4,X_input5,X_input6,graph_conv_filters_input,cnninput,tinput], outputs=output)
model1.summary()
Adam =keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)
model1.compile(loss='mse', optimizer=Adam, metrics=['mae',rmse])
history=model1.fit([np.array(X_train),np.array(X_train1),np.array(X_train2)
,np.array(X_train3),np.array(X_train4),np.array(X_train5),np.array(MX_train),np.array(cnnX_train),np.array(tX_train)],np.array(y_train),
batch_size=batch_size, validation_split=0.1,
epochs=nb_epochs, shuffle=True, verbose=1)
score=model1.evaluate([np.array(X_test),np.array(X_test1),np.array(X_test2),np.array(X_test3)
,np.array(X_test4),np.array(X_test5),np.array(MX_test),np.array(cnnX_test),np.array(tX_test)],np.array(y_test))
predict1=model1.predict([np.array(X_test),np.array(X_test1),np.array(X_test2),np.array(X_test3)
,np.array(X_test4),np.array(X_test5),np.array(MX_test),np.array(cnnX_test),np.array(tX_test)])