-
Notifications
You must be signed in to change notification settings - Fork 198
/
adversarial_abblstm.py
181 lines (149 loc) · 7.52 KB
/
adversarial_abblstm.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
from tensorflow.contrib.rnn import BasicLSTMCell
from tensorflow.python.ops.rnn import bidirectional_dynamic_rnn as bi_rnn
import time
from utils.prepare_data import *
from utils.model_helper import *
def scale_l2(x, norm_length):
# shape(x) = (batch, num_timesteps, d)
# Divide x by max(abs(x)) for a numerically stable L2 norm.
# 2norm(x) = a * 2norm(x/a)
# Scale over the full sequence, dims (1, 2)
alpha = tf.reduce_max(tf.abs(x), (1, 2), keepdims=True) + 1e-12
l2_norm = alpha * tf.sqrt(
tf.reduce_sum(tf.pow(x / alpha, 2), (1, 2), keepdims=True) + 1e-6)
x_unit = x / l2_norm
return norm_length * x_unit
def normalize(emb, weights):
# weights = vocab_freqs / tf.reduce_sum(vocab_freqs) ?? 这个实现没问题吗
print("Weights: ", weights)
mean = tf.reduce_sum(weights * emb, 0, keep_dims=True)
var = tf.reduce_sum(weights * tf.pow(emb - mean, 2.), 0, keep_dims=True)
stddev = tf.sqrt(1e-6 + var)
return (emb - mean) / stddev
class AdversarialClassifier(object):
def __init__(self, config):
self.max_len = config["max_len"]
self.hidden_size = config["hidden_size"]
self.vocab_size = config["vocab_size"]
self.embedding_size = config["embedding_size"]
self.n_class = config["n_class"]
self.learning_rate = config["learning_rate"]
self.epsilon = config["epsilon"]
# placeholder
self.x = tf.placeholder(tf.int32, [None, self.max_len])
self.label = tf.placeholder(tf.int32, [None])
self.keep_prob = tf.placeholder(tf.float32)
def _add_perturbation(self, embedded, loss):
"""Adds gradient to embedding and recomputes classification loss."""
grad, = tf.gradients(
loss,
embedded,
aggregation_method=tf.AggregationMethod.EXPERIMENTAL_ACCUMULATE_N)
grad = tf.stop_gradient(grad)
perturb = scale_l2(grad, self.epsilon)
return embedded + perturb
def _get_freq(self, vocab_freq, word2idx):
"""get a frequency dict format as {word_idx: word_freq}"""
words = vocab_freq.keys()
freq = [0] * self.vocab_size
for word in words:
word_idx = word2idx.get(word)
word_freq = vocab_freq[word]
freq[word_idx] = word_freq
return freq
def build_graph(self, vocab_freq, word2idx):
vocab_freqs = tf.constant(self._get_freq(vocab_freq, word2idx),
dtype=tf.float32, shape=(self.vocab_size, 1))
weights = vocab_freqs / tf.reduce_sum(vocab_freqs)
embeddings_var = tf.Variable(tf.random_uniform([self.vocab_size, self.embedding_size], -1.0, 1.0),
trainable=True, name="embedding_var")
embedding_norm = normalize(embeddings_var, weights)
batch_embedded = tf.nn.embedding_lookup(embedding_norm, self.x)
W = tf.Variable(tf.random_normal([self.hidden_size], stddev=0.1))
W_fc = tf.Variable(tf.truncated_normal([self.hidden_size, self.n_class], stddev=0.1))
b_fc = tf.Variable(tf.constant(0., shape=[self.n_class]))
def cal_loss_logit(embedded, keep_prob, reuse=True, scope="loss"):
with tf.variable_scope(scope, reuse=reuse) as scope:
rnn_outputs, _ = bi_rnn(BasicLSTMCell(self.hidden_size),
BasicLSTMCell(self.hidden_size),
inputs=embedded, dtype=tf.float32)
# Attention
H = tf.add(rnn_outputs[0], rnn_outputs[1]) # fw + bw
M = tf.tanh(H) # M = tanh(H) (batch_size, seq_len, HIDDEN_SIZE)
# alpha (bs * sl, 1)
alpha = tf.nn.softmax(tf.matmul(tf.reshape(M, [-1, self.hidden_size]),
tf.reshape(W, [-1, 1])))
r = tf.matmul(tf.transpose(H, [0, 2, 1]), tf.reshape(alpha, [-1, self.max_len,
1])) # supposed to be (batch_size * HIDDEN_SIZE, 1)
r = tf.squeeze(r)
h_star = tf.tanh(r)
drop = tf.nn.dropout(h_star, keep_prob)
# Fully connected layer（dense layer)
y_hat = tf.nn.xw_plus_b(drop, W_fc, b_fc)
return y_hat, tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y_hat, labels=self.label))
logits, self.cls_loss = cal_loss_logit(batch_embedded, self.keep_prob, reuse=False)
embedding_perturbated = self._add_perturbation(batch_embedded, self.cls_loss)
adv_logits, self.adv_loss = cal_loss_logit(embedding_perturbated, self.keep_prob, reuse=True)
self.loss = self.cls_loss + self.adv_loss
# optimization
loss_to_minimize = self.loss
tvars = tf.trainable_variables()
gradients = tf.gradients(loss_to_minimize, tvars, aggregation_method=tf.AggregationMethod.EXPERIMENTAL_TREE)
grads, global_norm = tf.clip_by_global_norm(gradients, 1.0)
self.global_step = tf.Variable(0, name="global_step", trainable=False)
self.optimizer = tf.train.AdamOptimizer(learning_rate=self.learning_rate)
self.train_op = self.optimizer.apply_gradients(zip(grads, tvars), global_step=self.global_step,
name='train_step')
self.prediction = tf.argmax(tf.nn.softmax(logits), 1)
print("graph built successfully!")
if __name__ == '__main__':
# load data
x_train, y_train = load_data("../dbpedia_data/dbpedia_csv/train.csv", sample_ratio=1e-2, one_hot=False)
x_test, y_test = load_data("../dbpedia_data/dbpedia_csv/test.csv", one_hot=False)
# data preprocessing
x_train, x_test, vocab_freq, word2idx, vocab_size = \
data_preprocessing_with_dict(x_train, x_test, max_len=32)
print("train size: ", len(x_train))
print("vocab size: ", vocab_size)
# split dataset to test and dev
x_test, x_dev, y_test, y_dev, dev_size, test_size = \
split_dataset(x_test, y_test, 0.1)
print("Validation Size: ", dev_size)
config = {
"max_len": 32,
"hidden_size": 64,
"vocab_size": vocab_size,
"embedding_size": 128,
"n_class": 15,
"learning_rate": 1e-3,
"batch_size": 32,
"train_epoch": 10,
"epsilon": 5,
}
classifier = AdversarialClassifier(config)
classifier.build_graph(vocab_freq, word2idx)
# auto GPU growth, avoid occupy all GPU memory
tf_config = tf.ConfigProto()
tf_config.gpu_options.allow_growth = True
sess = tf.Session(config=tf_config)
sess.run(tf.global_variables_initializer())
dev_batch = (x_dev, y_dev)
start = time.time()
for e in range(config["train_epoch"]):
t0 = time.time()
print("Epoch %d start !" % (e + 1))
for x_batch, y_batch in fill_feed_dict(x_train, y_train, config["batch_size"]):
return_dict = run_train_step(classifier, sess, (x_batch, y_batch))
t1 = time.time()
print("Train Epoch time: %.3f s" % (t1 - t0))
dev_acc = run_eval_step(classifier, sess, dev_batch)
print("validation accuracy: %.3f " % dev_acc)
print("Training finished, time consumed : ", time.time() - start, " s")
print("Start evaluating: \n")
cnt = 0
test_acc = 0
for x_batch, y_batch in fill_feed_dict(x_test, y_test, config["batch_size"]):
acc = run_eval_step(classifier, sess, (x_batch, y_batch))
test_acc += acc
cnt += 1
print("Test accuracy : %f %%" % (test_acc / cnt * 100))