-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathmodel.py
197 lines (170 loc) · 10 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
import tensorflow as tf
import json, pickle
from datetime import datetime
class adict(dict):
''' Attribute dictionary - a convenience data structure, similar to SimpleNamespace in python 3.3
One can use attributes to read/write dictionary content.
'''
def __init__(self, *av, **kav):
dict.__init__(self, *av, **kav)
self.__dict__ = self
def conv2d(input_, output_dim, k_h, k_w, name="conv2d"):
with tf.variable_scope(name):
w = tf.get_variable('w', [k_h, k_w, input_.get_shape()[-1], output_dim])
b = tf.get_variable('b', [output_dim])
return tf.nn.conv2d(input_, w, strides=[1, 1, 1, 1], padding='VALID') + b
class RNN_CNNs():
def _add_placeholder(self):
self.sentence = tf.placeholder(tf.int32, [None, self.config["sentence_length"]],
name="sentence")
self.word_list = tf.placeholder(tf.int32, [None, self.config["sentence_length"], self.config["word_length"]],
name="word_list")
#self.word_addition = tf.placeholder(tf.float32, [None, self.config["sentence_length"], 5],
# name="word_addition")
#self.char_addition = tf.placeholder(tf.float32,
# [None, self.config["sentence_length"], self.config["word_length"], 4],
# name="char_addition")
self.sent_len = tf.placeholder(tf.int32, [None], name="sent_len")
self.labels = tf.placeholder(tf.int32, [None, self.config["sentence_length"]], name ="labels")
def _add_embdding(self):
word_initializer = tf.constant(self.word_embed,
dtype=tf.float32)
char_initializer = tf.constant(self.char_embded,
dtype=tf.float32)
with tf.variable_scope("embeded_layer"):
word_embded = tf.get_variable("word_embeded", initializer=word_initializer)
char_embded = tf.get_variable("char_embeded", initializer=char_initializer)
#char_add_embded = tf.get_variable("char_add_embded",shape=[4])
self.word_vectors = tf.nn.embedding_lookup(word_embded, self.sentence)
self.char_vectors = tf.nn.embedding_lookup(char_embded, self.word_list)
#self.char_vectors = tf.concat(3, (self.char_vectors, self.char_addition))
#self.word_vectors = tf.concat(2, (self.word_vectors, self.word_addition))
#def _add_cnns_layer(self):
# filter_shape = [self.config["filter_size"], self.config["char_embded_size"] + 4,
# 1, self.config["char_feature_size"]]
# self.W_cnns = tf.get_variable("W_cnns", filter_shape, tf.float32)
# self.b_cnns = tf.get_variable("b_cnns", [self.config["char_feature_size"]],tf.float32)
# tf.add_to_collection("loss", tf.nn.l2_loss(self.W_cnns) + tf.nn.l2_loss(self.b_cnns))
#def _run_cnn(self, input):
# input = tf.expand_dims(input, -1)
# conv = tf.nn.conv2d(input, self.W_cnns, strides=[1,1,1,1], padding="VALID")
# h = tf.nn.relu(tf.nn.bias_add(conv, self.b_cnns))
# pooled = tf.nn.max_pool(h,
# ksize=[1, self.config["word_length"] - self.config["filter_size"] + 1, 1, 1],
# strides=[1, 1, 1, 1], padding="VALID")
# return tf.squeeze(pooled,[1])
@staticmethod
def conv2d(input_, output_dim, k_h, k_w, name="conv2d"):
with tf.variable_scope(name):
w = tf.get_variable('w', [k_h, k_w, input_.get_shape()[-1], output_dim])
b = tf.get_variable('b', [output_dim])
return tf.nn.conv2d(input_, w, strides=[1, 1, 1, 1], padding='VALID') + b
def _tdnn(self, input_, scope="tdnn"):
input_ = tf.expand_dims(input_, 1)
reduced_length = self.config["word_length"] - self.config["filter_size"] + 1
conv = self.conv2d(input_,
self.config["char_feature_size"], 1,
self.config["filter_size"],
name="kernel")
pool = tf.nn.max_pool(tf.tanh(conv), [1, 1, reduced_length, 1], [1, 1, 1, 1], 'VALID')
return pool
def _add_model(self):
# word_tensor = []
# char_level_inputs = [tf.squeeze(x, [1]) for x in tf.split(1, self.config["sentence_length"], self.char_vectors)]
# self._add_cnns_layer()
# for x in char_level_inputs:
# word_tensor.append(self._run_cnn(x))
# word_tensor = tf.concat(1, word_tensor)
char_level_inputs = tf.reshape(self.char_vectors,
[-1, self.config["word_length"],self.config["char_embded_size"]])
input_cnn = self._tdnn(char_level_inputs)
word_tensor = tf.reshape(input_cnn,
[-1, self.config["sentence_length"], self.config["char_feature_size"]]
)
input = tf.concat(2, (self.word_vectors, word_tensor))
cell = tf.nn.rnn_cell.LSTMCell(num_units=self.config["rnn_hidden"], state_is_tuple=True)
cell = tf.nn.rnn_cell.DropoutWrapper(cell, output_keep_prob=self.config["rnn_dropout"])
(output_fw, output_bw), _ = tf.nn.bidirectional_dynamic_rnn(cell, cell,
tf.transpose(input, perm=[1,0,2]),
dtype=tf.float32, sequence_length=self.sent_len,
time_major=True)
#output_fw = tf.reshape(tf.transpose(tf.pack(output_fw), perm=[1, 0, 2]), [-1, self.config["rnn_hidden"]])
#output_bw = tf.reshape(tf.transpose(tf.pack(output_bw), perm=[1, 0, 2]), [-1, self.config["rnn_hidden"]])
#W_fw = tf.get_variable('W_fw', [self.config["rnn_hidden"], self.config["num_class"]], tf.float32)
#b_fw = tf.get_variable('b_fw', [self.config["num_class"]], tf.float32)
#W_bw = tf.get_variable('W_bw', [self.config["rnn_hidden"], self.config["num_class"]], tf.float32)
#b_bw = tf.get_variable('b_bw', [self.config["num_class"]], tf.float32)
output = tf.concat(2, (output_fw, output_bw))
output = tf.reshape(tf.transpose(tf.pack(output), perm=[1, 0, 2]), [-1, 2 * self.config["rnn_hidden"]])
W = tf.get_variable('W_fw', [2 * self.config["rnn_hidden"], self.config["num_class"]], tf.float32)
b = tf.get_variable('b_fw', [self.config["num_class"]], tf.float32)
predict = tf.nn.softmax(tf.matmul(output, W) + b)
#predict_fw = tf.nn.softmax(tf.matmul(output_fw, W_fw) + b_fw)
#predict_bw = tf.nn.softmax(tf.matmul(output_bw, W_bw) + b_bw)
#tf.add_to_collection("loss", tf.nn.l2_loss(W_fw) + tf.nn.l2_loss(b_fw) +\
# tf.nn.l2_loss(W_bw) + tf.nn.l2_loss(b_bw))
self.prediction = tf.reshape(predict,
[-1, self.config["sentence_length"], self.config["num_class"]])
#self.loss = self._cost()
log_likehood, self.transition_params = tf.contrib.crf.crf_log_likelihood(
self.prediction, self.labels, self.sent_len)
self.loss = tf.reduce_mean(-log_likehood, name="loss")
optimizer = tf.train.AdamOptimizer(0.003)
tvars = tf.trainable_variables()
grads, _ = tf.clip_by_global_norm(tf.gradients(self.loss, tvars), 10)
self.train_op = optimizer.apply_gradients(zip(grads, tvars))
#self.loss = tf.nn.softmax_cross_entropy_with_logits(self.prediction, self.labels)
#self.loss = tf.reduce_sum(self.loss, reduction_indices=1) / tf.cast(self.sent_len, tf.float32)
#self.loss = tf.reduce_mean(self.loss)
#self.label_predict = tf.argmax(self.prediction, axis=2)
#optimizer = tf.train.AdamOptimizer(self.config["learning_rate"])
#self.train_op = optimizer.minimize(self.loss)
def _cost(self):
ce = self.labels * tf.log(self.prediction)
ce = -tf.reduce_sum(ce, reduction_indices=2)
mask = tf.sign(tf.reduce_max(tf.abs(self.labels), reduction_indices=2))
ce *= mask
ce = tf.reduce_sum(ce, reduction_indices=1)
ce /= tf.cast(self.sent_len, tf.float32)
return tf.reduce_min(ce)
def _build_graph(self):
self._add_placeholder()
self._add_embdding()
self._add_model()
def __init__(self, config, word_embded, char_embded):
self.word_embed = word_embded
self.char_embded = char_embded
self.config = config
self._build_graph()
init = tf.global_variables_initializer()
self.sess = tf.Session()
self.sess.run(init)
self.saver = tf.train.Saver()
def save_model(self):
now = datetime.utcnow().strftime("%Y%m%d%H%M%S")
save_path=saver.save(sess,"tmp/model-%s.ckpt" %now)
def partial_fit(self, sentence, word_list, sent_len, labels ):
feed_dict = {self.sentence : sentence,
self.word_list : word_list,
self.sent_len: sent_len,
self.labels: labels
}
cost, opt = self.sess.run((self.loss, self.train_op), feed_dict= feed_dict)
return cost
def calc_total_cost(self, sentence, word_list,sent_len, labels ):
feed_dict = {self.sentence : sentence,
self.word_list : word_list,
self.sent_len: sent_len,
self.labels: labels
}
return self.sess.run(self.loss, feed_dict = feed_dict)
def transform(self, sentence, word_list, sent_len):
feed_dict = {self.sentence : sentence,
self.word_list : word_list,
self.sent_len: sent_len
}
logits, transition_params = self.sess.run([self.prediction, self.transition_params],
feed_dict=feed_dict)
viterbi_sequence, viterbi_score = tf.contrib.crf.viterbi_decode(
logits, transition_params)
return viterbi_sequence