-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathmatrix_vae.py
188 lines (144 loc) · 8.17 KB
/
matrix_vae.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
from __future__ import division, print_function
import numpy as np
import tensorflow as tf
def weight_variable(shape, name):
#xavier initialisation
in_dim = shape[0]
xavier_stddev = 1. / tf.sqrt(in_dim / 2.)
return tf.Variable(tf.random_normal(shape=shape, stddev=xavier_stddev), name=name)
def bias_variable(shape, name):
b_init = tf.constant_initializer(0.)
return tf.get_variable(name, shape, initializer=b_init)
class VAEMF(object):
def __init__(self, sess, num_user, num_item,
hidden_encoder_dim=216, hidden_decoder_dim=216, latent_dim=24,
learning_rate=0.002, batch_size=64, reg_param=0,
user_embed_dim=216, item_embed_dim=216, activate_fn=tf.tanh, vae=True):
if reg_param < 0 or reg_param > 1:
raise ValueError("regularization parameter must be in [0,1]")
self.sess = sess
self.num_user = num_user
self.num_item = num_item
self.hidden_encoder_dim = hidden_encoder_dim
self.hidden_decoder_dim = hidden_decoder_dim
self.latent_dim = latent_dim
self.learning_rate = learning_rate
self.batch_size = batch_size
self.reg_param = reg_param
self.user_embed_dim = user_embed_dim
self.item_embed_dim = item_embed_dim
self.activate_fn = activate_fn
self.vae = vae
self.build_model()
def build_model(self):
self.l2_loss = tf.constant(0.0)
self.user = tf.placeholder("float", shape=[None, self.num_item])
self.valid_rating = tf.placeholder("float", shape=[None, self.num_item])
self.test_rating = tf.placeholder("float", shape=[None, self.num_item])
self.W_encoder_input_hidden_user = weight_variable(
[self.num_item, self.hidden_encoder_dim], 'W_encoder_input_hidden_user')
self.b_encoder_input_hidden_user = bias_variable(
[self.hidden_encoder_dim], 'b_encoder_input_hidden_user')
self.l2_loss += tf.nn.l2_loss(self.W_encoder_input_hidden_user)
# Hidden layer encoder
self.hidden_encoder_user = self.activate_fn(tf.matmul(
self.user, self.W_encoder_input_hidden_user) + self.b_encoder_input_hidden_user)
self.W_encoder_hidden_mu_user = weight_variable(
[self.hidden_encoder_dim, self.latent_dim], 'W_encoder_hidden_mu_user')
self.b_encoder_hidden_mu_user = bias_variable(
[self.latent_dim], 'b_encoder_hidden_mu_user')
self.l2_loss += tf.nn.l2_loss(self.W_encoder_hidden_mu_user)
# Mu encoder
self.mu_encoder_user = tf.matmul(
self.hidden_encoder_user, self.W_encoder_hidden_mu_user) + self.b_encoder_hidden_mu_user
self.W_encoder_hidden_logvar_user = weight_variable(
[self.hidden_encoder_dim, self.latent_dim], 'W_encoder_hidden_logvar_user')
self.b_encoder_hidden_logvar_user = bias_variable(
[self.latent_dim], 'b_encoder_hidden_logvar_user')
self.l2_loss += tf.nn.l2_loss(self.W_encoder_hidden_logvar_user)
if self.vae:
# Sigma encoder
self.logvar_encoder_user = tf.matmul(
self.hidden_encoder_user, self.W_encoder_hidden_logvar_user) + self.b_encoder_hidden_logvar_user
# Sample epsilon
self.epsilon_user = tf.random_normal(
tf.shape(self.logvar_encoder_user), name='epsilon_user')
# Sample latent variable
self.std_encoder_user = tf.exp(0.5 * self.logvar_encoder_user)
self.z_user = self.mu_encoder_user + \
tf.multiply(self.std_encoder_user, self.epsilon_user)
else:
self.z_user = self.mu_encoder_user
# decoding network
self.W_decoder_z_hidden_user = weight_variable(
[self.latent_dim, self.hidden_decoder_dim], 'W_decoder_z_hidden_user')
self.b_decoder_z_hidden_user = bias_variable(
[self.hidden_decoder_dim], 'b_decoder_z_hidden_user')
self.l2_loss += tf.nn.l2_loss(self.W_decoder_z_hidden_user)
# Hidden layer decoder
self.hidden_decoder_user = self.activate_fn(tf.matmul(
self.z_user, self.W_decoder_z_hidden_user) + self.b_decoder_z_hidden_user)
self.W_decoder_hidden_reconstruction_user = weight_variable(
[self.hidden_decoder_dim, self.num_item], 'W_decoder_hidden_reconstruction_user')
self.b_decoder_hidden_reconstruction_user = bias_variable(
[self.num_item], 'b_decoder_hidden_reconstruction_user')
self.l2_loss += tf.nn.l2_loss(
self.W_decoder_hidden_reconstruction_user)
self.reconstructed_user = tf.matmul(
self.hidden_decoder_user, self.W_decoder_hidden_reconstruction_user) + self.b_decoder_hidden_reconstruction_user
weight = tf.not_equal(self.user, tf.constant(0, dtype=tf.float32))
self.MSE = tf.losses.mean_squared_error(self.user, self.reconstructed_user, weight)
self.MAE = tf.losses.absolute_difference(self.user, self.reconstructed_user, weight)
if self.vae:
# KL divergence between prior and variational distributions
self.KLD = -0.5 * tf.reduce_sum(1 + self.logvar_encoder_user - tf.pow(
self.mu_encoder_user, 2) - tf.exp(self.logvar_encoder_user), reduction_indices=1)
self.loss = tf.reduce_mean(self.KLD + self.MSE)
else:
self.loss = tf.reduce_mean(self.MSE)
self.regularized_loss = self.loss + self.reg_param * self.l2_loss
valid_weight = tf.not_equal(self.valid_rating, tf.constant(0, dtype=tf.float32))
test_weight = tf.not_equal(self.test_rating, tf.constant(0, dtype=tf.float32))
self.valid_RMSE = tf.sqrt(tf.losses.mean_squared_error(self.valid_rating, self.reconstructed_user, valid_weight))
self.test_RMSE = tf.sqrt(tf.losses.mean_squared_error(self.test_rating, self.reconstructed_user, test_weight))
tf.summary.scalar("MSE", self.MSE)
tf.summary.scalar("MAE", self.MAE)
tf.summary.scalar("Loss", self.loss)
tf.summary.scalar("valid-RMSE", self.valid_RMSE)
tf.summary.scalar("test-RMSE", self.test_RMSE)
tf.summary.scalar("Reg-Loss", self.regularized_loss)
self.train_step = tf.train.AdamOptimizer(
self.learning_rate).minimize(self.regularized_loss)
# add op for merging summary
self.summary_op = tf.summary.merge_all()
# add Saver ops
self.saver = tf.train.Saver()
def train_test_validation(self, M, train_idx, test_idx, valid_idx, n_steps=100000, result_path='result/'):
nonzero_user_idx = M.nonzero()[0]
nonzero_item_idx = M.nonzero()[1]
trainM = np.zeros(M.shape)
trainM[nonzero_user_idx[train_idx], nonzero_item_idx[train_idx]] = M[nonzero_user_idx[train_idx], nonzero_item_idx[train_idx]]
validM = np.zeros(M.shape)
validM[nonzero_user_idx[valid_idx], nonzero_item_idx[valid_idx]] = M[nonzero_user_idx[valid_idx], nonzero_item_idx[valid_idx]]
testM = np.zeros(M.shape)
testM[nonzero_user_idx[test_idx], nonzero_item_idx[test_idx]] = M[nonzero_user_idx[test_idx], nonzero_item_idx[test_idx]]
for i in range(self.num_user):
if np.sum(trainM[i]) == 0:
testM[i] = 0
validM[i] = 0
train_writer = tf.summary.FileWriter(
result_path + '/train', graph=self.sess.graph)
best_val_rmse = np.inf
best_test_rmse = 0
self.sess.run(tf.global_variables_initializer())
for step in range(1, n_steps):
feed_dict = {self.user: trainM, self.valid_rating:validM, self.test_rating:testM}
_, mse, mae, valid_rmse, test_rmse, summary_str = self.sess.run(
[self.train_step, self.MSE, self.MAE, self.valid_RMSE, self.test_RMSE, self.summary_op], feed_dict=feed_dict)
train_writer.add_summary(summary_str, step)
print("Iter {0} Train RMSE:{1}, Valid RMSE:{2}, Test RMSE:{3}".format(step, np.sqrt(mse), valid_rmse, test_rmse))
if best_val_rmse > valid_rmse:
best_val_rmse = valid_rmse
best_test_rmse = test_rmse
self.saver.save(self.sess, result_path + "/model.ckpt")
return best_test_rmse