-
Notifications
You must be signed in to change notification settings - Fork 45
/
Copy pathmain.py
executable file
·171 lines (148 loc) · 7.23 KB
/
main.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
import os
import tensorflow as tf
import numpy as np
import ujson as json
from tqdm import tqdm
import sys
from model import Hybrid as Model
from util import get_record_parser, convert_tokens, evaluate, get_batch_dataset, get_dataset
def train(config):
print(dict(config.__dict__['__flags']))
print()
print(sys.stdout.flush())
os.environ["CUDA_VISIBLE_DEVICES"] = config.choose_gpu
with open(config.word_emb_file, "r") as fh:
word_mat = np.asarray(json.load(fh), dtype=np.float32)
with open(config.char_emb_file, "r") as fh:
char_mat = np.asarray(json.load(fh), dtype=np.float32)
with open(config.train_eval_file, "r") as fh:
train_eval_file = json.load(fh)
with open(config.dev_eval_file, "r") as fh:
dev_eval_file = json.load(fh)
with open(config.dev_meta, "r") as fh:
meta = json.load(fh)
dev_total = meta["total"]
print("Building model......")
parser = get_record_parser(config)
train_dataset = get_batch_dataset(config.train_record_file, parser, config)
dev_dataset = get_dataset(config.dev_record_file, parser, config)
handle = tf.placeholder(tf.string, shape=[])
iterator = tf.data.Iterator.from_string_handle(
handle, train_dataset.output_types, train_dataset.output_shapes)
train_iterator = train_dataset.make_one_shot_iterator()
dev_iterator = dev_dataset.make_one_shot_iterator()
model = Model(config, iterator, word_mat, char_mat)
sess_config = tf.ConfigProto(allow_soft_placement=True)
sess_config.gpu_options.allow_growth = True
sess_config.gpu_options.per_process_gpu_memory_fraction = config.gpu_memory_fraction
loss_save = 100.0
patience = 0
lr = config.init_lr
with tf.Session(config=sess_config) as sess:
writer = tf.summary.FileWriter(config.log_dir)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
train_handle = sess.run(train_iterator.string_handle())
dev_handle = sess.run(dev_iterator.string_handle())
sess.run(tf.assign(model.is_train, tf.constant(True, dtype=tf.bool)))
sess.run(tf.assign(model.lr, tf.constant(lr, dtype=tf.float32)))
for _ in tqdm(range(1, config.num_steps + 1)):
global_step = sess.run(model.global_step) + 1
loss, train_op = sess.run([model.loss, model.train_op],
feed_dict={handle:train_handle})
if global_step % config.period == 0:
loss_sum = tf.Summary(value=[tf.Summary.Value(
tag="model/loss", simple_value=loss), ])
writer.add_summary(loss_sum, global_step)
if global_step % config.checkpoint == 0:
sess.run(tf.assign(model.is_train,
tf.constant(False, dtype=tf.bool)))
_, summ = evaluate_batch(
model, config.val_num_batches, train_eval_file, sess, "train", handle, train_handle)
for s in summ:
writer.add_summary(s, global_step)
metrics, summ = evaluate_batch(
model, dev_total // config.batch_size + 1, dev_eval_file, sess, "dev", handle, dev_handle)
sess.run(tf.assign(model.is_train,
tf.constant(True, dtype=tf.bool)))
dev_loss = metrics["loss"]
if dev_loss < loss_save:
loss_save = dev_loss
patience = 0
else:
patience += 1
if patience >= config.patience:
lr /= 2.0
loss_save = dev_loss
patience = 0
sess.run(tf.assign(model.lr, tf.constant(lr, dtype=tf.float32)))
for s in summ:
writer.add_summary(s, global_step)
writer.flush()
filename = os.path.join(config.save_dir,
"model_{}.ckpt".format(global_step))
saver.save(sess, filename)
def evaluate_batch(model, num_batches, eval_file, sess, data_type, handle, str_handle):
answer_dict = {}
losses = []
for _ in tqdm(range(1, num_batches + 1)):
qa_id, loss, yp1, yp2 = sess.run(
[model.qa_id, model.loss, model.yp1, model.yp2], feed_dict={handle:str_handle})
answer_dict_, _ = convert_tokens(
eval_file, qa_id.tolist(), yp1.tolist(), yp2.tolist())
answer_dict.update(answer_dict_)
losses.append(loss)
loss = np.mean(losses)
metrics = evaluate(eval_file, answer_dict)
metrics["loss"] = loss
loss_sum = tf.Summary(value=[tf.Summary.Value(
tag="{}/loss".format(data_type), simple_value=loss), ])
f1_sum = tf.Summary(value=[tf.Summary.Value(
tag="{}/f1".format(data_type), simple_value=metrics["f1"]), ])
em_sum = tf.Summary(value=[tf.Summary.Value(
tag="{}/em".format(data_type), simple_value=metrics["exact_match"]), ])
print("[{}] loss:{}, f1:{}, em:{}".format(data_type, loss, metrics["f1"], metrics["exact_match"]))
return metrics, [loss_sum, f1_sum, em_sum]
def test(config):
os.environ["CUDA_VISIBLE_DEVICES"] = config.choose_gpu
with open(config.word_emb_file, "r") as fh:
word_mat = np.array(json.load(fh), dtype=np.float32)
with open(config.char_emb_file, "r") as fh:
char_mat = np.array(json.load(fh), dtype=np.float32)
with open(config.test_eval_file, "r") as fh:
eval_file = json.load(fh)
with open(config.test_meta, "r") as fh:
meta = json.load(fh)
total = meta["total"]
print("Loading model...")
test_batch = get_dataset(config.test_record_file, get_record_parser(
config, is_test=True), config).make_one_shot_iterator()
model = Model(config, test_batch, word_mat, char_mat, trainable=False)
sess_config = tf.ConfigProto(allow_soft_placement=True)
sess_config.gpu_options.allow_growth = True
try:
sess_config.gpu_options.per_process_gpu_memory_fraction = config.gpu_memory_fraction
except:
sess_config.gpu_options.per_process_gpu_memory_fraction = 0.5
with tf.Session(config=sess_config) as sess:
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
saver.restore(sess, tf.train.latest_checkpoint(config.save_dir))
sess.run(tf.assign(model.is_train, tf.constant(False, dtype=tf.bool)))
losses = []
answer_dict = {}
remapped_dict = {}
for step in tqdm(range(total // config.batch_size + 1)):
qa_id, loss, yp1, yp2 = sess.run(
[model.qa_id, model.loss, model.yp1, model.yp2])
answer_dict_, remapped_dict_ = convert_tokens(
eval_file, qa_id.tolist(), yp1.tolist(), yp2.tolist())
answer_dict.update(answer_dict_)
remapped_dict.update(remapped_dict_)
losses.append(loss)
loss = np.mean(losses)
metrics = evaluate(eval_file, answer_dict)
with open(config.answer_file, "w") as fh:
json.dump(remapped_dict, fh)
print("Exact Match: {}, F1: {}".format(
metrics['exact_match'], metrics['f1']))