-
Notifications
You must be signed in to change notification settings - Fork 3
/
hierachical_laa.py
245 lines (218 loc) · 11.1 KB
/
hierachical_laa.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
import tensorflow as tf
import numpy as np
import deep_laa_support as dls
import sys
# read data
# filename = "web_processed_data_feature_2"
# filename = "age_data_3_category"
filename = "bluebird_data"
# filename = "flower_data"
data_all = np.load(filename+'.npz')
user_labels = data_all['user_labels']
label_mask = data_all['label_mask']
true_labels = data_all['true_labels']
category_size = data_all['category_num']
source_num = data_all['source_num']
n_samples, _ = np.shape(true_labels)
input_size = source_num * category_size
batch_size = n_samples
# define x
x = tf.placeholder(dtype=tf.float32, shape=(batch_size, input_size))
mask = tf.placeholder(dtype=tf.float32, shape=(batch_size, input_size))
# define source-wise template
source_wise_template = np.zeros((input_size, input_size), dtype=np.float32)
for i in range(input_size):
source_wise_template[i*category_size:(i+1)*category_size, i*category_size:(i+1)*category_size] = 1
# define constraint template
constraint_template_encoder = np.matlib.repmat(np.eye(category_size), source_num, 1)
constraint_template_decoder = np.matlib.repmat(np.eye(category_size), 1, source_num)
# mv_y
mv_y = dls.get_majority_y(user_labels, category_size)
# define constant_y
constant_y = dls.get_constant_y(batch_size, category_size)
# x -> y1 (confusion matrix)
with tf.variable_scope('encoder_x_y1'):
# network
weights = tf.Variable(
tf.truncated_normal(shape=(input_size, category_size), mean=0.0, stddev=0.01), name='w_encoder')
biases = tf.Variable(
tf.zeros(shape=([category_size]), dtype=tf.float32), name='b_encoder')
y1 = tf.nn.softmax(
tf.add(tf.matmul(x, weights), biases))
# loss
loss_w_encoder_l2 = tf.nn.l2_loss(weights - constraint_template_encoder)
loss_b_encoder_l2 = tf.nn.l2_loss(biases)
y1_encoder_constraint_strength = tf.placeholder(dtype=tf.float32)
loss_y1_encoder = y1_encoder_constraint_strength/source_num/category_size/category_size * (loss_w_encoder_l2 + loss_b_encoder_l2)
# x -> y2 (weighted majority voting)
with tf.variable_scope('encoder_x_y2'):
# network
weights = tf.Variable(
tf.truncated_normal(shape=(input_size, category_size), mean=0.0, stddev=0.01), name='w_encoder')
biases = tf.Variable(
tf.zeros(shape=([category_size]), dtype=tf.float32), name='b_encoder')
y2 = tf.nn.softmax(
tf.add(tf.matmul(x, weights), biases))
# loss
loss_w_encoder_l1 = tf.reduce_sum(tf.abs(weights))
loss_b_encoder_l1 = tf.reduce_sum(tf.abs(biases))
y2_encoder_constraint_strength = tf.placeholder(dtype=tf.float32)
loss_y2_encoder = y2_encoder_constraint_strength/source_num/category_size/category_size * (loss_w_encoder_l1 + loss_b_encoder_l1)
# y1, y2 -> y
with tf.variable_scope('encoder_y1_y2_y'):
# network
weights_y1 = tf.Variable(
tf.truncated_normal(shape=(category_size, category_size), mean=0.0, stddev=0.01), name='w_encoder')
weights_y2 = tf.Variable(
tf.truncated_normal(shape=(category_size, category_size), mean=0.0, stddev=0.01), name='w_encoder')
biases = tf.Variable(
tf.zeros(shape=([category_size]), dtype=tf.float32), name='b_encoder')
y = tf.nn.softmax(
tf.add(tf.matmul(y1, weights_y1) + tf.matmul(y2, weights_y2), biases))
# loss
y_prior = tf.placeholder(dtype=tf.float32, shape=(batch_size, category_size))
loss_y_kl = tf.reduce_mean(tf.reduce_sum(tf.mul(y, tf.log(1e-10 + y)) - tf.mul(y, tf.log(1e-10 + y_prior)), reduction_indices=1))
y_kl_strength = tf.placeholder(dtype=tf.float32)
loss_w1_encoder_l2 = tf.nn.l2_loss(weights_y1 - np.eye(category_size))
loss_w2_encoder_l2 = tf.nn.l2_loss(weights_y2 - np.eye(category_size))
loss_b_encoder_l2 = tf.nn.l2_loss(biases)
y_encoder_constraint_strength = tf.placeholder(dtype=tf.float32)
loss_y = y_kl_strength * loss_y_kl + \
y_encoder_constraint_strength/category_size/category_size/2 * (loss_w1_encoder_l2 + loss_w2_encoder_l2 + loss_b_encoder_l2)
# y -> y1
with tf.variable_scope('decoder_y_y1'):
# network
weights = tf.Variable(
tf.truncated_normal(shape=(category_size, category_size), mean=0.0, stddev=0.01), name='w_decoder')
biases = tf.Variable(
tf.zeros(shape=([category_size]), dtype=tf.float32), name='b_decoder')
y1 = tf.nn.softmax(
tf.add(tf.matmul(y, weights), biases))
# loss
loss_w_decoder_l2 = tf.nn.l2_loss(weights - np.eye(category_size))
loss_b_encoder_l2 = tf.nn.l2_loss(biases)
y1_decoder_constraint_strength = tf.placeholder(dtype=tf.float32)
loss_y1_decoder = y1_decoder_constraint_strength/category_size/category_size * (loss_w_decoder_l2 + loss_b_encoder_l2)
# y -> y2
with tf.variable_scope('decoder_y_y2'):
# network
weights = tf.Variable(
tf.truncated_normal(shape=(category_size, category_size), mean=0.0, stddev=0.01), name='w_decoder')
biases = tf.Variable(
tf.zeros(shape=([category_size]), dtype=tf.float32), name='b_decoder')
y2 = tf.nn.softmax(
tf.add(tf.matmul(y, weights), biases))
# loss
loss_w_decoder_l2 = tf.nn.l2_loss(weights - np.eye(category_size))
loss_b_encoder_l2 = tf.nn.l2_loss(biases)
y2_decoder_constraint_strength = tf.placeholder(dtype=tf.float32)
loss_y2_decoder = y2_decoder_constraint_strength/category_size/category_size * (loss_w_decoder_l2 + loss_b_encoder_l2)
# y1 -> x1
with tf.variable_scope('decoder_y1_x1'):
# network
weights = tf.Variable(
tf.truncated_normal(shape=(category_size, input_size), mean=0.0, stddev=.01), name='w_recons')
biases = tf.Variable(
tf.zeros(shape=([input_size]), dtype=tf.float32), name='b_recons')
x_reconstr_tmp = tf.add(tf.matmul(y1, weights), biases)
x_reconstr_1 = tf.div(tf.exp(x_reconstr_tmp), tf.matmul(tf.exp(x_reconstr_tmp), source_wise_template))
#loss
_tmp_cross_entropy = - tf.mul(x, tf.log(1e-10 + x_reconstr_1))
# divide label numbers !!
loss_x1_cross_entropy = tf.reduce_mean(tf.reduce_sum(tf.mul(mask, _tmp_cross_entropy), reduction_indices=1))
loss_w_decoder_l2 = tf.nn.l2_loss(weights - constraint_template_decoder)
loss_b_decoder_l2 = tf.nn.l2_loss(biases)
x1_decoder_constraint_strength = tf.placeholder(dtype=tf.float32)
loss_x1_decoder = x1_decoder_constraint_strength/source_num/category_size/category_size * (loss_w_decoder_l2 + loss_b_decoder_l2)
# y2 -> x2
with tf.variable_scope('decoder_y2_x2'):
# network
weights = tf.Variable(
tf.truncated_normal(shape=(category_size, input_size), mean=0.0, stddev=.01), name='w_recons')
biases = tf.Variable(
tf.zeros(shape=([input_size]), dtype=tf.float32), name='b_recons')
x_reconstr_tmp = tf.add(tf.matmul(y2, weights), biases)
x_reconstr_2 = tf.div(tf.exp(x_reconstr_tmp), tf.matmul(tf.exp(x_reconstr_tmp), source_wise_template))
# loss
_tmp_cross_entropy = - tf.mul(x, tf.log(1e-10 + x_reconstr_2))
# divide label numbers !!
loss_x2_cross_entropy = tf.reduce_mean(tf.reduce_sum(tf.mul(mask, _tmp_cross_entropy), reduction_indices=1))
loss_w_decoder_l1 = tf.reduce_sum(tf.abs(weights))
loss_b_decoder_l1 = tf.reduce_sum(tf.abs(biases))
x2_decoder_constraint_strength = tf.placeholder(dtype=tf.float32)
loss_x2_decoder = x2_decoder_constraint_strength/source_num/category_size/category_size * (loss_w_decoder_l1 + loss_b_decoder_l1)
# loss
loss_x_cross_entropy = loss_x1_cross_entropy + loss_x2_cross_entropy
loss_overall = loss_x_cross_entropy \
+ loss_y1_encoder + loss_y2_encoder \
+ loss_y \
+ loss_y1_decoder + loss_y2_decoder \
+ loss_x1_decoder + loss_x2_decoder
y_target = tf.placeholder(dtype=tf.float32, shape=(batch_size, category_size))
_tmp_y_cross_entropy = - tf.mul(y_target, tf.log(1e-10 + y))
loss_pre_train = tf.reduce_mean(tf.reduce_sum(_tmp_y_cross_entropy, reduction_indices=1))
# optimizer
learning_rate = 0.02
optimizer_autoencoder = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss_overall)
optimizer_pre_train = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss_pre_train)
# evaluate with true labels
y_label = tf.placeholder(dtype=tf.int64, shape=(batch_size, 1))
inferred_category = tf.reshape(tf.argmax(y, 1), (batch_size, 1))
hit_num = tf.reduce_sum(tf.to_int32(tf.equal(inferred_category, y_label)))
# session
with tf.Session() as sess:
tf.initialize_all_variables().run()
# pre-train
epochs = 50
total_batches = int(n_samples / batch_size)
for epoch in xrange(epochs):
total_hit = 0
for batch in xrange(total_batches):
batch_x, batch_mask, batch_y_label, batch_majority_y = user_labels, label_mask, true_labels, mv_y
# x -> y, update classifier
_, batch_y, batch_hit_num = sess.run(
[optimizer_pre_train, y, hit_num],
feed_dict={x:batch_x, mask:batch_mask, y_label:batch_y_label, y_prior:mv_y, y_target:mv_y,
y1_encoder_constraint_strength:0.001,
y2_encoder_constraint_strength:0.001,
y_kl_strength:0.0001,
y_encoder_constraint_strength:0.001})
total_hit += batch_hit_num
print "epoch: {0} accuracy: {1}".format(epoch, float(total_hit)/n_samples)
epochs = 300
total_batches = int(n_samples / batch_size)
for epoch in xrange(epochs):
total_cost = 0.0
total_hit = 0
for batch in xrange(total_batches):
batch_x, batch_mask, batch_y_label, batch_majority_y = user_labels, label_mask, true_labels, mv_y
# x -> y, update classifier
_, batch_y, batch_hit_num = sess.run(
[optimizer_autoencoder, y, hit_num],
feed_dict={x:batch_x, mask:batch_mask, y_label:batch_y_label, y_prior:mv_y,
y1_encoder_constraint_strength:0.1,
y2_encoder_constraint_strength:0.1,
y_kl_strength:0.001,
y_encoder_constraint_strength:0.1,
y1_decoder_constraint_strength:0.1,
y2_decoder_constraint_strength:0.1,
x1_decoder_constraint_strength:0.1,
x2_decoder_constraint_strength:0.1})
total_hit += batch_hit_num
print "epoch: {0} accuracy: {1}".format(epoch, float(total_hit)/n_samples)
if epoch == epochs-1:
debug_y1, debug_y2, _ = sess.run(
[y1, y2, y],
feed_dict={x:batch_x, mask:batch_mask, y_label:batch_y_label, y_prior:mv_y,
y1_encoder_constraint_strength:0.1,
y2_encoder_constraint_strength:0.1,
y_kl_strength:0.001,
y_encoder_constraint_strength:0.1,
y1_decoder_constraint_strength:0.1,
y2_decoder_constraint_strength:0.1,
x1_decoder_constraint_strength:0.1,
x2_decoder_constraint_strength:0.1})
# print debug_z
print "y1:", debug_y1
print "y2:", debug_y2
print "Done!"