-
Notifications
You must be signed in to change notification settings - Fork 27
/
Copy pathtransr_model.py
159 lines (143 loc) · 8.27 KB
/
transr_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
# coding:utf-8
import tensorflow as tf
import sys
import ctypes
import numpy as np
class TransR(object):
def __init__(self, entity_size, relation_size, hidden_size_e, hidden_size_r, margin=1.0, learning_rate=0.001,
l1_flag=True, model_name='transr_model', ckpt_path='./ckpt/transr/'):
self.entity_size = entity_size
self.relation_size = relation_size
self.hidden_size_e = hidden_size_e
self.hidden_size_r = hidden_size_r
self.margin = margin
self.learning_rate = learning_rate
self.l1_flag = l1_flag
self.model_name = model_name
self.ckpt_path = ckpt_path
# build graph
sys.stdout.write('\nBuilding Graph...')
tf.reset_default_graph()
# set inputs
# with tf.name_scope("read_inputs"):
self.pos_h = tf.placeholder(tf.int32, [None])
self.pos_t = tf.placeholder(tf.int32, [None])
self.pos_r = tf.placeholder(tf.int32, [None])
self.neg_h = tf.placeholder(tf.int32, [None])
self.neg_t = tf.placeholder(tf.int32, [None])
self.neg_r = tf.placeholder(tf.int32, [None])
with tf.name_scope("embedding"):
self.ent_embeddings = tf.get_variable(name="ent_embedding", shape=[self.entity_size, self.hidden_size_e],
initializer=tf.contrib.layers.xavier_initializer(uniform=False))
self.rel_embeddings = tf.get_variable(name="rel_embedding", shape=[self.relation_size, self.hidden_size_r],
initializer=tf.contrib.layers.xavier_initializer(uniform=False))
self.rel_matrix = tf.get_variable(name="rel_matrix", shape=[self.relation_size,
self.hidden_size_e * self.hidden_size_r],
initializer=tf.contrib.layers.xavier_initializer(uniform=False))
with tf.name_scope('lookup_embeddings'):
pos_h_e = tf.reshape(tf.nn.embedding_lookup(self.ent_embeddings, self.pos_h), [-1, self.hidden_size_e, 1])
pos_t_e = tf.reshape(tf.nn.embedding_lookup(self.ent_embeddings, self.pos_t), [-1, self.hidden_size_e, 1])
pos_r_e = tf.reshape(tf.nn.embedding_lookup(self.rel_embeddings, self.pos_r), [-1, self.hidden_size_r])
neg_h_e = tf.reshape(tf.nn.embedding_lookup(self.ent_embeddings, self.neg_h), [-1, self.hidden_size_e, 1])
neg_t_e = tf.reshape(tf.nn.embedding_lookup(self.ent_embeddings, self.neg_t), [-1, self.hidden_size_e, 1])
neg_r_e = tf.reshape(tf.nn.embedding_lookup(self.rel_embeddings, self.neg_r), [-1, self.hidden_size_r])
matrix = tf.reshape(tf.nn.embedding_lookup(self.rel_matrix, self.neg_r),
[-1, self.hidden_size_r, self.hidden_size_e])
pos_h_e = tf.reshape(tf.matmul(matrix, pos_h_e), [-1, self.hidden_size_r])
pos_t_e = tf.reshape(tf.matmul(matrix, pos_t_e), [-1, self.hidden_size_r])
neg_h_e = tf.reshape(tf.matmul(matrix, neg_h_e), [-1, self.hidden_size_r])
neg_t_e = tf.reshape(tf.matmul(matrix, neg_t_e), [-1, self.hidden_size_r])
if self.l1_flag:
pos = tf.reduce_sum(abs(pos_h_e + pos_r_e - pos_t_e), 1, keep_dims=True)
neg = tf.reduce_sum(abs(neg_h_e + neg_r_e - neg_t_e), 1, keep_dims=True)
self.predict = pos
else:
pos = tf.reduce_sum((pos_h_e + pos_r_e - pos_t_e) ** 2, 1, keep_dims=True)
neg = tf.reduce_sum((neg_h_e + neg_r_e - neg_t_e) ** 2, 1, keep_dims=True)
self.predict = pos
with tf.name_scope("output"):
self.loss = tf.reduce_sum(tf.maximum(pos - neg + margin, 0))
self.global_step = tf.Variable(0, name="global_step", trainable=False)
optimizer = tf.train.AdamOptimizer(self.learning_rate)
grads_and_vars = optimizer.compute_gradients(self.loss)
self.train_op = optimizer.apply_gradients(grads_and_vars, global_step=self.global_step)
sys.stdout.write('Done...\n')
def train(self, dataset, num_steps=3000, batch_size=100, sess=None):
saver = tf.train.Saver() # use to save the model
if sess is None:
sess = tf.Session()
sess.run(tf.global_variables_initializer())
ph = np.zeros(batch_size, dtype=np.int32)
pt = np.zeros(batch_size, dtype=np.int32)
pr = np.zeros(batch_size, dtype=np.int32)
nh = np.zeros(batch_size, dtype=np.int32)
nt = np.zeros(batch_size, dtype=np.int32)
nr = np.zeros(batch_size, dtype=np.int32)
ph_addr = ph.__array_interface__['data'][0]
pt_addr = pt.__array_interface__['data'][0]
pr_addr = pr.__array_interface__['data'][0]
nh_addr = nh.__array_interface__['data'][0]
nt_addr = nt.__array_interface__['data'][0]
nr_addr = nr.__array_interface__['data'][0]
dataset.getBatch.argtypes = [ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p,
ctypes.c_void_p, ctypes.c_void_p, ctypes.c_int]
nbatches = dataset.getTripleTotal() // batch_size
step = 0
sys.stdout.write('Training started...\n')
try:
for step in range(1, num_steps + 1):
res = 0
for batch in range(nbatches):
dataset.getBatch(ph_addr, pt_addr, pr_addr, nh_addr, nt_addr, nr_addr, batch_size)
feed_dict = {self.pos_h: ph, self.pos_t: pt, self.pos_r: pr, self.neg_h: nh, self.neg_t: nt,
self.neg_r: nr}
_, _, loss = sess.run([self.train_op, self.global_step, self.loss], feed_dict=feed_dict)
res += loss
print(' step %4d, loss: %8.4f' % (step, res / nbatches))
if step % 100 == 0: # save the model periodically
sys.stdout.write('Saving model at step %d... ' % step)
saver.save(sess, self.ckpt_path + self.model_name, global_step=step)
sys.stdout.write('Done...\n')
except KeyboardInterrupt:
sys.stdout.write('Interrupted by user at training step %d, saving model at this step.. ' % step)
saver.save(sess, self.ckpt_path + self.model_name, global_step=step)
sys.stdout.write('Done...\n')
return sess
def restore_last_session(self):
saver = tf.train.Saver()
sess = tf.Session() # create a session
ckpt = tf.train.get_checkpoint_state(self.ckpt_path) # get checkpoint state
if ckpt and ckpt.model_checkpoint_path: # restore session
saver.restore(sess, ckpt.model_checkpoint_path)
return sess
def test(self, testset, sess=None):
if sess is None:
print('restore model from last check point...')
sess = self.restore_last_session()
batch_size = testset.getEntityTotal()
ph = np.zeros(batch_size, dtype=np.int32)
pt = np.zeros(batch_size, dtype=np.int32)
pr = np.zeros(batch_size, dtype=np.int32)
ph_addr = ph.__array_interface__['data'][0]
pt_addr = pt.__array_interface__['data'][0]
pr_addr = pr.__array_interface__['data'][0]
testset.getHeadBatch.argtypes = [ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p]
testset.getTailBatch.argtypes = [ctypes.c_void_p, ctypes.c_void_p, ctypes.c_void_p]
testset.testHead.argtypes = [ctypes.c_void_p]
testset.testTail.argtypes = [ctypes.c_void_p]
num_steps = testset.getTestTotal()
for step in range(num_steps):
# test head
testset.getHeadBatch(ph_addr, pt_addr, pr_addr)
feed_dict = {self.pos_h: ph, self.pos_t: pt, self.pos_r: pr}
_, predict_head = sess.run([self.global_step, self.predict], feed_dict)
testset.testHead(predict_head.__array_interface__['data'][0])
# test tail
testset.getTailBatch(ph_addr, pt_addr, pr_addr)
feed_dict = {self.pos_h: ph, self.pos_t: pt, self.pos_r: pr}
_, predict_tail = sess.run([self.global_step, self.predict], feed_dict)
testset.testTail(predict_tail.__array_interface__['data'][0])
print(step)
if step % 50 == 0:
testset.test()
testset.test()