-
Notifications
You must be signed in to change notification settings - Fork 175
/
Copy pathmodel.py
248 lines (148 loc) · 8.82 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
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
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
import tensorflow as tf
import warnings
warnings.filterwarnings("ignore")
import logging
logging.getLogger('tensorflow').disabled = True
import numpy as np
class DKLITE(object):
def __init__(self, input_dim, output_dim, num_hidden=50, num_layers =2, learning_rate=0.001,
reg_var=1.0,reg_rec=1.0):
self.num_layers = num_layers
self.output_dim = output_dim
self.num_hidden = num_hidden
self.input_dim = input_dim
self.size_z = num_hidden
self.ml_primal = {}
self.ker_inv = {}
self.params = {}
self.mean = {}
self.num = {}
''' Initialize parameter weight '''
self.params = self.initialize_weights()
self.mu = tf.reduce_mean(self.T)
self.Z_train = self.Encoder(self.X)
self.Z_test = self.Encoder(self.X_u)
self.loss_1 = tf.reduce_mean(tf.reduce_sum(tf.square(self.X - self.Decoder(self.Z_train)),axis=1))
Z_0 = tf.gather(self.Z_train, tf.where(self.T < 0.5)[:, 0])
Y_0 = tf.gather(self.Y, tf.where(self.T < 0.5)[:, 0])
Z_1 = tf.gather(self.Z_train, tf.where(self.T > 0.5)[:, 0])
Y_1 = tf.gather(self.Y, tf.where(self.T > 0.5)[:, 0])
mean_0 = tf.reduce_mean(Y_0)
mean_1 = tf.reduce_mean(Y_1)
Y_0 = (Y_0-mean_0)
Y_1 = (Y_1-mean_1)
self.GP_NN(Y_0, Z_0, 0)
self.GP_NN(Y_1, Z_1,1)
self.var_0 = tf.reduce_mean(tf.diag_part(tf.matmul(Z_1,tf.matmul(self.ker_inv['0'], tf.transpose(Z_1)))))
self.var_1 = tf.reduce_mean(tf.diag_part(tf.matmul(Z_0,tf.matmul(self.ker_inv['1'], tf.transpose(Z_0)))))
self.ele_var_0_tr = tf.diag_part(tf.matmul(self.Z_train,tf.matmul(self.ker_inv['0'], tf.transpose(self.Z_train))))
self.ele_var_1_tr = tf.diag_part(tf.matmul(self.Z_train,tf.matmul(self.ker_inv['1'], tf.transpose(self.Z_train))))
self.ele_var_0_te = tf.diag_part(tf.matmul(self.Z_test,tf.matmul(self.ker_inv['0'], tf.transpose(self.Z_test))))
self.ele_var_1_te = tf.diag_part(tf.matmul(self.Z_test,tf.matmul(self.ker_inv['1'], tf.transpose(self.Z_test))))
pred_tr_0 = tf.matmul(self.Z_train, self.mean['0']) + mean_0
pred_tr_1 = tf.matmul(self.Z_train, self.mean['1']) + mean_1
pred_te_0 = tf.matmul(self.Z_test, self.mean['0']) + mean_0
pred_te_1 = tf.matmul(self.Z_test, self.mean['1']) + mean_1
self.Y_train = tf.concat([pred_tr_0,pred_tr_1],axis=1)
self.Y_test = tf.concat([pred_te_0,pred_te_1],axis=1)
self.loss_0 = self.ml_primal['0']+ self.ml_primal['1']
self.prediction_loss = self.ml_primal['0']+ self.ml_primal['1'] + reg_var *(self.var_0 + self.var_1)+ reg_rec * self.loss_1
self.optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(self.prediction_loss)
self.sess = tf.Session()
self.sess.run(tf.global_variables_initializer())
def element_var(self, X, Y, T, X_u):
var_0_tr, var_1_tr,var_0_te, var_1_te = self.sess.run([self.ele_var_0_tr,self.ele_var_1_tr,self.ele_var_0_te,self.ele_var_1_te],
feed_dict={self.X: X, self.X_u: X_u, self.Y: Y, self.T: T})
return var_0_tr,var_1_tr,var_0_te,var_1_te
def embed(self, X, Y, T):
Z= self.sess.run(self.Z_train, feed_dict={self.X: X, self.Y: Y, self.T: T})
return Z
def fit(self, X, Y, T, num_iteration):
loss_list = []
for i in range(num_iteration):
loss, _ = self.sess.run([self.prediction_loss,self.optimizer], feed_dict={self.X: X, self.Y: Y, self.T: T})
loss_list.append(np.sum(loss))
diff_list = np.abs(np.diff(loss_list))
if i>50 and np.abs(np.mean(diff_list[-10:]) - np.mean(diff_list[-40:-10]) )< np.std(diff_list[-40:-10]):
break
def pred(self, X, Y, T, X_u):
Y_hat_train, Y_hat_test = self.sess.run([self.Y_train, self.Y_test], feed_dict={self.X: X, self.X_u: X_u, self.Y: Y, self.T: T})
return Y_hat_train, Y_hat_test
def destroy_graph(self):
tf.reset_default_graph()
def Encoder(self, X):
X_h =tf.nn.elu(tf.matmul(X, self.params['e_w_in']) + self.params['e_b_in'])
for layer_i in range( self.num_layers):
X_h = tf.nn.elu(tf.matmul(X_h, self.params['e_w_' + str(layer_i)])+self.params['e_b_' + str(layer_i)])
Z = tf.nn.elu(tf.matmul(X_h, self.params['e_w_' + str(self.num_layers)])+ self.params['e_b_' + str(self.num_layers)])
return Z
def Decoder(self,Z):
Z_pred = tf.nn.elu(tf.matmul(Z, self.params['d_w_in']) + self.params['d_b_in'])
for layer_i in range(self.num_layers):
Z_pred = tf.nn.elu(tf.matmul(Z_pred, self.params['d_w_' + str(layer_i)])+ self.params['d_b_' + str(layer_i)])
X_p = tf.matmul(Z_pred, self.params['d_w_' + str(self.num_layers)]+ self.params['d_b_' + str(self.num_layers)])
return X_p
def GP_NN(self, Y_f, Z_f,index):
beta = tf.ones([1,1],tf.float32)
lam = 1000*tf.ones([1,1],tf.float32)
r = beta / lam
self.DD = tf.shape(Z_f)[1]
phi_phi = tf.matmul(tf.transpose(Z_f), Z_f)
Ker = r * phi_phi + tf.eye(tf.shape(Z_f)[1], dtype=tf.float32)
L_matrix = tf.cholesky(Ker)
L_inv_reduce = tf.linalg.triangular_solve(L_matrix, rhs=tf.eye(self.DD, dtype=tf.float32))
L_y = tf.matmul(L_inv_reduce, tf.matmul(tf.transpose(Z_f), Y_f))
self.ker_inv[str(index)] = tf.matmul(tf.transpose(L_inv_reduce), L_inv_reduce) / lam
self.mean[str(index)] = r * tf.matmul(tf.transpose(L_inv_reduce), L_y)
term1 = - tf.reduce_mean(tf.square(L_y))
#term2 = tf.log(tf.linalg.diag_part(L_matrix)) / ((1-index)*tf.reduce_sum(1 - self.T) + (index)* tf.reduce_sum(self.T))
self.ml_primal[str(index)] = term1 #+ term2
def initialize_weights(self):
self.X = tf.placeholder(tf.float32, [None, self.input_dim])
self.X_u = tf.placeholder(tf.float32, [None, self.input_dim])
self.Y = tf.placeholder(tf.float32, [None, 1])
self.T = tf.placeholder(tf.float32, [None, 1])
all_weights = {}
''' Input layer of the encoder '''
name_wi = 'e_w_in'
all_weights[name_wi] = tf.get_variable(name=name_wi, shape=[self.input_dim, self.num_hidden], trainable=True)
name_bi = 'e_b_in'
all_weights[name_bi] = tf.get_variable(name=name_bi, shape=[self.num_hidden], trainable=True)
''' Hidden layer of the encoder '''
for layer_i in range(self.num_layers):
name_wi = 'e_w_' + str(layer_i)
all_weights[name_wi ] = tf.get_variable(name =name_wi, shape=[self.num_hidden,self.num_hidden], trainable=True)
name_bi = 'e_b_' + str(layer_i)
all_weights[name_bi] = tf.get_variable(name =name_bi, shape = [self.num_hidden], trainable=True)
''' Final layer of the encoder '''
name_wi = 'e_w_' + str(self.num_layers)
all_weights[name_wi] = tf.get_variable(name=name_wi, shape=[self.num_hidden, self.size_z], trainable=True)
name_bi = 'e_b_' + str(self.num_layers)
all_weights[name_bi] = tf.get_variable(name=name_bi, shape=[self.size_z], trainable=True)
name_wi = 'e_w_out_0'
all_weights[name_wi] = tf.get_variable(name=name_wi, shape=[self.size_z, self.output_dim], trainable=True)
name_bi = 'e_b_out_0'
all_weights[name_bi] = tf.get_variable(name=name_bi, shape=[self.output_dim], trainable=True)
name_wi = 'e_w_out_1'
all_weights[name_wi] = tf.get_variable(name=name_wi, shape=[self.size_z, self.output_dim], trainable=True)
name_bi = 'e_b_out_1'
all_weights[name_bi] = tf.get_variable(name=name_bi, shape=[self.output_dim], trainable=True)
''' Input layer of the decoder '''
name_wi = 'd_w_in'
all_weights[name_wi] = tf.get_variable(name=name_wi, shape=[self.size_z, self.num_hidden],trainable=True)
name_bi = 'd_b_in'
all_weights[name_bi] = tf.get_variable(name=name_bi, shape=[self.num_hidden],trainable=True)
''' Hidden layer of the decoder '''
for layer_i in range(self.num_layers):
name_wi = 'd_w_' + str(layer_i)
all_weights[name_wi ] = tf.get_variable(name =name_wi, shape=[self.num_hidden,self.num_hidden],trainable=True)
name_bi = 'd_b_' + str(layer_i)
all_weights[name_bi] = tf.get_variable(name =name_bi, shape = [self.num_hidden],trainable=True)
''' Final layer of the decoder '''
name_wi = 'd_w_' + str(self.num_layers)
all_weights[name_wi] = tf.get_variable(name=name_wi, shape=[self.num_hidden, self.input_dim],trainable=True)
name_bi = 'd_b_' + str(self.num_layers)
all_weights[name_bi] = tf.get_variable(name=name_bi, shape=[(self.input_dim)],trainable=True)
return all_weights