-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodel.py
185 lines (159 loc) · 11 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
import os
import sys
sys.path.append(os.path.dirname(os.getcwd()))
import tensorflow as tf
from bert import modeling
from bert import optimization
from crf import CRF
from module import multi_head_attention, GCN
class BertNer(object):
def __init__(self, config, is_training=True, num_train_step=None, num_warmup_step=None):
self.__bert_config_path = os.path.join(config["bert_model_path"], "bert_config.json")
self.__num_classes = config["num_classes"]
self.__learning_rate = config["learning_rate"]
self.__ner_layers = config["ner_layers"]
self.__k = config["top_k"]
self.picture_length = config["picture_length"]
self.picture_dimension = config["picture_dimension"]
self.__max_len = config["sequence_length"]
self.multi_heads = config["multi_heads"]
self.__random_base = config["random_base"]
self.__is_training = is_training
self.__num_train_step = num_train_step
self.__num_warmup_step = num_warmup_step
self.input_ids = tf.placeholder(dtype=tf.int32, shape=[None, None], name='input_ids')
self.input_masks = tf.placeholder(dtype=tf.int32, shape=[None, None], name='input_mask')
self.segment_ids = tf.placeholder(dtype=tf.int32, shape=[None, None], name='segment_ids')
self.label_ids = tf.placeholder(dtype=tf.int32, shape=[None, None], name="label_ids")
self.sequence_len = tf.placeholder(dtype=tf.int32, shape=[None], name="sequence_len")
self.picture_id = tf.placeholder(dtype=tf.float32, shape=[None, None, None], name='picture_id')
self.pt_simi_score_id = tf.placeholder(dtype=tf.float32, shape=[None, None, None], name='pt_simi_score_id')
self.pt_image_id = tf.placeholder(dtype=tf.float32, shape=[None, None, None, None], name='pt_image_id')
self.pp_simi_score_id = tf.placeholder(dtype=tf.float32, shape=[None, None, None], name='pp_simi_score_id')
self.pp_image_id = tf.placeholder(dtype=tf.float32, shape=[None, None, None, None], name='pp_image_id')
self.keep_prob = tf.placeholder(dtype=tf.float32, shape=None, name="keep_prob")
self.built_model()
self.init_saver()
def built_model(self):
bert_config = modeling.BertConfig.from_json_file(self.__bert_config_path)
model = modeling.BertModel(config=bert_config,
is_training=self.__is_training,
input_ids=self.input_ids,
input_mask=self.input_masks,
token_type_ids=self.segment_ids,
use_one_hot_embeddings=False)
output_layer = model.get_sequence_output()
embedding_dims = output_layer.shape[-1].value
output_picture_feature = self.picture_id
inputs_pic = tf.reshape(output_picture_feature, [-1, self.picture_length, self.picture_dimension])
pt_image_embedding = self.pt_image_id
inputs_pt_image = tf.reshape(pt_image_embedding, [-1, self.picture_length * self.__k, self.picture_dimension])
pp_image_embedding = self.pp_image_id
inputs_pp_image = tf.reshape(pp_image_embedding, [-1, self.picture_length * self.__k, self.picture_dimension])
pt_simi_score_matrix = self.pt_simi_score_id
input_pt_simi = tf.reshape(pt_simi_score_matrix, [-1, self.__k, self.__k])
pp_simi_score_matrix = self.pp_simi_score_id
input_pp_simi = tf.reshape(pp_simi_score_matrix, [-1, self.__k, self.__k])
inputs_pt_image = tf.reshape(inputs_pt_image, [-1, self.__k * self.picture_length, self.picture_dimension])
output_pt = GCN(inputs_pt_image, self.__k, input_pt_simi, self.picture_length, self.picture_dimension, self.__random_base, "pt")
output_pt_sen_img = tf.squeeze(tf.reduce_mean(output_pt, 1))
output_pt_sen_img = tf.reshape(output_pt_sen_img, [-1, self.picture_length, self.picture_dimension])
output_pt_sen_img = tf.reshape(output_pt_sen_img, [-1, self.picture_dimension // 2])
weight_image_to_text = tf.get_variable(name="image_to_text",shape=[self.picture_dimension // 2, embedding_dims],
initializer=tf.random_uniform_initializer(-self.__random_base, self.__random_base))
output_pt_sen_img = tf.matmul(output_pt_sen_img, weight_image_to_text)
output_pt_sen_img = tf.reshape(output_pt_sen_img, [-1, embedding_dims])
output_layer = tf.reshape(output_layer, [-1, embedding_dims])
gate_weight_pt_image = tf.get_variable(name="gate_weight_pt_image", shape=[embedding_dims, embedding_dims], initializer=tf.random_uniform_initializer(-self.__random_base, self.__random_base))
gate_pt = tf.sigmoid(tf.matmul(output_layer, gate_weight_pt_image) + tf.matmul(output_pt_sen_img, gate_weight_pt_image))
gate_pt = tf.reshape(gate_pt, [-1, self.__max_len, embedding_dims])
output_pt_sen_img = tf.reshape(output_pt_sen_img, [-1, self.__max_len, embedding_dims])
output_layer = tf.reshape(output_layer, [-1, self.__max_len, embedding_dims])
output_pt_sen_img = output_layer + gate_pt * output_pt_sen_img
inputs_pp_image = tf.reshape(inputs_pp_image, [-1, self.__k * self.picture_length, self.picture_dimension])
output_pp = GCN(inputs_pp_image, self.__k, input_pp_simi, self.picture_length, self.picture_dimension, self.__random_base, "pp")
output_pp_img_img = tf.squeeze(tf.reduce_mean(output_pp, 1))
output_pp_img_img = tf.reshape(output_pp_img_img, [-1, self.picture_dimension])
inputs_pic = tf.reshape(inputs_pic, [-1, self.picture_dimension])
gate_weight_pp_image = tf.get_variable(name="gate_weight_pp_image", shape=[self.picture_dimension, self.picture_dimension], initializer=tf.random_uniform_initializer(-self.__random_base, self.__random_base))
gate_pp = tf.sigmoid(tf.matmul(inputs_pic, gate_weight_pp_image) + tf.matmul(output_pp_img_img, gate_weight_pp_image))
gate_pp = tf.reshape(gate_pp, [-1, self.picture_length, self.picture_dimension])
output_pp_img_img = tf.reshape(output_pp_img_img, [-1, self.picture_length, self.picture_dimension])
inputs_pic = tf.reshape(inputs_pic, [-1, self.picture_length, self.picture_dimension])
output_pp_img_img = inputs_pic + gate_pp * output_pp_img_img
if self.__is_training:
output_pt_sen_img = tf.nn.dropout(output_pt_sen_img, keep_prob=0.9)
output_layer_sen = multi_head_attention(output_pt_sen_img, output_pt_sen_img, output_pt_sen_img, embedding_dims, self.multi_heads, scope="self_multihead_sen")
output_layer_imag = multi_head_attention(output_pp_img_img, output_pp_img_img, output_pp_img_img, self.picture_dimension, self.multi_heads, scope="self_multihead_imag")
output_layer_sen = multi_head_attention(output_layer_imag, output_layer_imag, output_layer_sen, embedding_dims, self.multi_heads, scope="cross_multihead")
ner_model = CRF(embedded_chars=output_layer_sen,
layers=self.__ner_layers,
keep_prob=self.keep_prob,
num_labels=self.__num_classes,
max_len=self.__max_len,
labels=self.label_ids,
sequence_lens=self.sequence_len,
is_training=self.__is_training)
self.loss, self.true_y, self.predictions = ner_model.construct_graph()
if self.__is_training:
with tf.name_scope('train_op'):
self.train_op = optimization.create_optimizer(
self.loss, self.__learning_rate, self.__num_train_step, self.__num_warmup_step, use_tpu=False)
def init_saver(self):
self.saver = tf.train.Saver(tf.global_variables())
def train(self, sess, batch, dropout_rate):
feed_dict = {self.input_ids: batch["input_ids"],
self.input_masks: batch["input_masks"],
self.segment_ids: batch["segment_ids"],
self.label_ids: batch["label_ids"],
self.sequence_len: batch["sequence_len"],
self.picture_id: batch["picture_id"],
self.pt_simi_score_id: batch["pt_simi_score_id"],
self.pt_image_id: batch["pt_image_id"],
self.pp_simi_score_id: batch["pp_simi_score_id"],
self.pp_image_id: batch["pp_image_id"],
self.keep_prob: dropout_rate}
_, loss, true_y, predictions = sess.run([self.train_op, self.loss, self.true_y, self.predictions],
feed_dict=feed_dict)
return loss, true_y, predictions
def eval(self, sess, batch):
feed_dict = {self.input_ids: batch["input_ids"],
self.input_masks: batch["input_masks"],
self.segment_ids: batch["segment_ids"],
self.label_ids: batch["label_ids"],
self.sequence_len: batch["sequence_len"],
self.picture_id: batch["picture_id"],
self.pt_simi_score_id: batch["pt_simi_score_id"],
self.pt_image_id: batch["pt_image_id"],
self.pp_simi_score_id: batch["pp_simi_score_id"],
self.pp_image_id: batch["pp_image_id"],
self.keep_prob: 1.0}
loss, true_y, predictions = sess.run([self.loss, self.true_y, self.predictions], feed_dict=feed_dict)
return loss, true_y, predictions
def infer(self, sess, batch):
feed_dict = {self.input_ids: batch["input_ids"],
self.input_masks: batch["input_masks"],
self.segment_ids: batch["segment_ids"],
self.sequence_len: batch["sequence_len"],
self.picture_id: batch["picture_id"],
self.pt_simi_score_id: batch["pt_simi_score_id"],
self.pt_image_id: batch["pt_image_id"],
self.pp_simi_score_id: batch["pp_simi_score_id"],
self.pp_image_id: batch["pp_image_id"],
self.keep_prob: 1.0}
predict = sess.run(self.predictions, feed_dict=feed_dict)
return predict
def infer_1(self, sess, batch):
feed_dict = {self.input_ids: batch["input_ids"],
self.input_masks: batch["input_masks"],
self.segment_ids: batch["segment_ids"],
self.label_ids: batch["label_ids"],
self.sequence_len: batch["sequence_len"],
self.picture_id: batch["picture_id"],
self.pt_simi_score_id: batch["pt_simi_score_id"],
self.pt_image_id: batch["pt_image_id"],
self.pp_simi_score_id: batch["pp_simi_score_id"],
self.pp_image_id: batch["pp_image_id"],
self.keep_prob: 1.0}
true_y, predict = sess.run([self.true_y, self.predictions], feed_dict=feed_dict)
return true_y, predict