forked from nadre/clickbait_detection
-
Notifications
You must be signed in to change notification settings - Fork 0
/
clickbait_text_cnn.py
407 lines (323 loc) · 14.5 KB
/
clickbait_text_cnn.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
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
import tensorflow as tf
from tensorflow.python.client import device_lib
import numpy as np
import pandas as pd
import functools
import datetime
import name_gen as ng
import gensim
import os
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "6,7"
np.random.seed(1338)
DTYPE = 'float32'
RUN_NAME = ng.get_name()
LOG_DIR = 'logs/{}/'.format(RUN_NAME)
CHECKPOINT_DIR = 'checkpoints/{}/'.format(RUN_NAME)
DATA_DIR = 'data/'
def lazy_property(function):
"""
http://danijar.com/structuring-your-tensorflow-models/
http://www.wildml.com/2015/12/implementing-a-cnn-for-text-classification-in-tensorflow/
paper: http://arxiv.org/abs/1408.5882
:param function:
:return:
"""
attribute = '_cache_' + function.__name__
@property
@functools.wraps(function)
def decorator(self):
if not hasattr(self, attribute):
# print(function.__name__)
setattr(self, attribute, function(self))
return getattr(self, attribute)
return decorator
class Model:
def __init__(self, name, sequence_length, output_size, vocab_size=int(3e6), train_batch_size=100, test_batch_size=100,
embedding_size=300, num_filters=64, max_filter_length=15, beta=0.005, dropout_keep_prob=0.75,
embedding_name='unknown', learning_rate=0.05, info=''):
self.name = name
self.date = datetime.datetime.now().strftime('%Y-%m-%d %H:%M')
self.embedding_name = embedding_name
self.learning_rate = learning_rate
self.train_batch_size = train_batch_size
self.test_batch_size = test_batch_size
self.output_size = output_size
self.vocab_size = vocab_size
self.embedding_size = embedding_size
self.filter_sizes = [fs for fs in range(1, max_filter_length)]
self.num_filters = num_filters
self.pooling_layer_output_size = self.num_filters * len(self.filter_sizes)
self.sequence_length = sequence_length
self.dropout_keep_prob = dropout_keep_prob
self.beta = beta
self.info = info
self.lowest_log_loss = 9e10
self.best_step = 0
self._sequence_placeholder = tf.placeholder(tf.int32, shape=(None, self.sequence_length),
name='sequence_placeholder')
self._target_placeholder = tf.placeholder(tf.float32, shape=(None, self.output_size),
name='target_placeholder')
self._dropout_keep_prob_placeholder = tf.placeholder(tf.float32, name="dropout_keep_prob")
self.embedding_placeholder
self.embedding_lookup
self.convolution_and_max_pooling
self.prediction
self.optimize
self.merged_summaries
self.output_weights_and_bias
self.global_step
self.l2_loss
def get_info(self):
info = ''
for attr, value in self.__dict__.items():
if not attr.startswith('_') and not callable(value):
info += '{}: {}\n'.format(attr, value)
return info
def save_info(self, info_dir, fname):
os.makedirs(info_dir, exist_ok=True)
with open(info_dir + fname, 'w') as text_file:
print(self.get_info(), file=text_file)
@lazy_property
def prediction(self):
pooling = self.convolution_and_max_pooling
with tf.name_scope('prediction'):
weights, bias = self.output_weights_and_bias
activation = tf.matmul(pooling, weights) + bias
softmax_out = tf.nn.softmax(activation)
summarize_variable('activation', activation)
summarize_variable('softmax_out', softmax_out)
summarize_variable('bias', bias)
summarize_variable('weights', weights)
summarize_variable('l2_loss', self.l2_loss)
return softmax_out
@lazy_property
def predict_one(self, placeholder):
pooling = self.convolution_and_max_pooling(placeholder)
weights, bias = self.output_weights_and_bias
activation = tf.matmul(pooling, weights) + bias
softmax_out = tf.nn.softmax(activation)
return softmax_out, tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
@lazy_property
def convolution_and_max_pooling(self, placeholder=None):
if placeholder is None:
embedding_lookup = self.embedding_lookup
else:
embedding_lookup = self.embedding_lookup(placeholder)
pooled_outputs = []
for i, filter_size in enumerate(self.filter_sizes):
with tf.name_scope('convolution-maxpool-%s' % filter_size):
# Convolution Layer
filter_shape = [filter_size, self.embedding_size, 1, self.num_filters]
W = tf.Variable(tf.truncated_normal(filter_shape, stddev=0.1), name='W')
b = tf.Variable(tf.constant(0.1, shape=[self.num_filters]), name='b')
conv = tf.nn.conv2d(
embedding_lookup,
W,
strides=[1, 1, 1, 1],
padding='VALID',
name='convolution')
# Apply nonlinearity
h = tf.nn.relu(tf.nn.bias_add(conv, b), name='relu')
# Max-pooling over the outputs
pooled = tf.nn.max_pool(
h,
ksize=[1, self.sequence_length - filter_size + 1, 1, 1],
strides=[1, 1, 1, 1],
padding='VALID',
name='pooling')
pooled_outputs.append(pooled)
# Combine all the pooled features
with tf.name_scope('combine_and_reshape'):
h_pool = tf.concat(pooled_outputs, 3)
h_pool_flat = tf.reshape(h_pool, [-1, self.pooling_layer_output_size])
with tf.name_scope('dropout'):
h_pool_flat = tf.nn.dropout(h_pool_flat, self._dropout_keep_prob_placeholder)
return h_pool_flat
@lazy_property
def output_weights_and_bias(self):
weights = tf.get_variable(
"output_weights",
shape=[self.pooling_layer_output_size, self.output_size],
initializer=tf.contrib.layers.xavier_initializer())
bias = tf.Variable(tf.constant(0.1, shape=[self.output_size]), name="output_bias")
return weights, bias
@lazy_property
def embedding_lookup(self, placeholder=None):
if placeholder is None:
sequence_placeholder = self._sequence_placeholder
else:
sequence_placeholder = placeholder
with tf.device('/:cpu0'):
embedding = tf.Variable(self.embedding_placeholder, trainable=False)
embedding_lookup = tf.nn.embedding_lookup(embedding, sequence_placeholder)
embedding_lookup_expanded = tf.expand_dims(embedding_lookup, -1)
return embedding_lookup_expanded
@lazy_property
def embedding_placeholder(self):
"""
https://stackoverflow.com/questions/35394103/initializing-tensorflow-variable-with-an-array-larger-than-2gb
:return:
"""
return tf.placeholder(tf.float32, shape=(self.vocab_size, self.embedding_size),
name='embedding_placeholder')
@lazy_property
def optimize(self):
with tf.name_scope('train'):
loss = self.log_loss + self.beta * self.l2_loss
optimizer = tf.train.AdagradOptimizer(self.learning_rate)
return optimizer.minimize(loss)
@lazy_property
def global_step(self):
return tf.Variable(0, name="global_step", trainable=False)
@lazy_property
def mse(self):
return tf.losses.mean_squared_error(self._target_placeholder, self.prediction)
@lazy_property
def mse_mean(self):
return tf.reduce_mean(self.mse)
@lazy_property
def log_loss(self):
return tf.losses.log_loss(labels=self._target_placeholder, predictions=self.prediction)
@lazy_property
def log_loss_mean(self):
return tf.reduce_mean(self.log_loss)
@lazy_property
def l2_loss(self):
weights, bias = self.output_weights_and_bias
return tf.nn.l2_loss(weights) + tf.nn.l2_loss(bias)
@lazy_property
def l2_loss_mean(self):
return tf.reduce_mean(self.l2_loss)
@lazy_property
def merged_summaries(self):
return tf.summary.merge_all()
@lazy_property
def checkpoint(self):
if not os.path.exists(CHECKPOINT_DIR):
os.makedirs(CHECKPOINT_DIR)
return tf.train.Saver(tf.global_variables())
def summarize_variable(name_scope, var):
"""Attach a lot of summaries to a Tensor (for TensorBoard visualization)."""
with tf.name_scope(name_scope + '_summaries'):
mean = tf.reduce_mean(var)
tf.summary.scalar('mean', mean)
with tf.name_scope('stddev'):
stddev = tf.sqrt(tf.reduce_mean(tf.square(var - mean)))
tf.summary.scalar('stddev', stddev)
tf.summary.scalar('max', tf.reduce_max(var))
tf.summary.scalar('min', tf.reduce_min(var))
tf.summary.histogram('histogram', var)
def get_random_batch(batch_size, data, labels):
assert batch_size < data.shape[0]
data_batch = data.sample(batch_size)
label_batch = labels.loc[data_batch.index]
return data_batch, label_batch
def get_batch(data, labels, batch_size, step):
num_samples = data.shape[0]
assert batch_size < num_samples
start = step * batch_size
if start > (num_samples - 1):
start %= num_samples
end = start + batch_size
if end > (num_samples - 1):
end = num_samples - 1
return data[start:end], labels[start:end]
def get_vocab_and_pretrained_embedding(path_to_model, binary=False):
model = gensim.models.KeyedVectors.load_word2vec_format(path_to_model, binary=binary)
embedding = model.syn0
vocab = model.vocab
return vocab, embedding
def load_data(embedding_name):
truth = pd.read_pickle(DATA_DIR+embedding_name+'_labels.pickle')
tokens = pd.read_pickle(DATA_DIR+embedding_name+'_indices.pickle')
return tokens, truth
def sample_test_set(data, labels, fraction):
"""
https://stackoverflow.com/questions/17260109/sample-two-pandas-dataframes-the-same-way
"""
assert data.shape[0] == labels.shape[0]
indices = np.random.binomial(1, fraction, size=data.shape[0]).astype('bool')
train_data = data[~indices]
test_data = data[indices]
train_labels = labels[~indices]
test_labels = labels[indices]
return train_data, test_data, train_labels, test_labels
def evaluate_test_set(model, sess, test_data, test_labels, train_step, summary_writer):
test_set_size = test_data.shape[0]
num_test_steps = int(test_set_size/model.test_batch_size) + 1
errors = {
'mse': [],
'log_loss': [],
'l2_loss': []
}
for test_step in range(num_test_steps):
test_data_batch, test_label_batch = get_batch(data=test_data, labels=test_labels,
batch_size=model.test_batch_size, step=test_step)
mse, log_loss, l2_loss, summary = sess.run([
model.mse_mean,
model.log_loss_mean,
model.l2_loss_mean,
model.merged_summaries
], feed_dict={model._sequence_placeholder: test_data_batch,
model._target_placeholder: test_label_batch,
model._dropout_keep_prob_placeholder: 1.0})
errors['mse'].append(mse)
errors['log_loss'].append(log_loss)
errors['l2_loss'].append(l2_loss)
print('\n\n'
'Train Step: {}\n'
'Test Step: {}\n'
'MSE {:6.10f}\n'
'Log Loss {:6.10f}\n'
'L2 Loss {:6.10f}\n'
.format(train_step, test_step, mse, log_loss, l2_loss))
summary_writer.add_summary(summary, train_step + test_step)
error_description_df = pd.DataFrame.from_dict(errors).describe()
summary = tf.Summary()
print(RUN_NAME+'#'*(80 - len(RUN_NAME)))
for key in error_description_df.keys():
for measurement in ['mean', 'std']:
print('{} {} : {:6.10f}'.format(key, measurement, error_description_df[key][measurement]))
tag = 'test_{}_{}'.format(key, measurement)
summary.value.add(tag=tag, simple_value=error_description_df[key][measurement])
print('#'*80)
summary_writer.add_summary(summary, train_step)
log_loss = error_description_df['log_loss']['mean']
if log_loss < model.lowest_log_loss:
model.lowest_log_loss = log_loss
model.best_step = train_step
model.save_info(LOG_DIR, RUN_NAME + '.txt')
model.checkpoint.save(sess=sess, save_path=CHECKPOINT_DIR+'step_{}.ckpt'.format(train_step))
def main(embedding_name):
print(device_lib.list_local_devices())
tokens, truth = load_data(embedding_name)
num_instances, sequence_length = tokens.shape
_, output_size = truth.shape
train_data, test_data, train_labels, test_labels = sample_test_set(tokens, truth, 0.1)
test_set_size = test_data.shape[0]
num_instances -= test_set_size
print('loading embedding...')
vocab, embedding = get_vocab_and_pretrained_embedding(DATA_DIR + embedding_name + '.bin', binary=True)
print('...done.')
vocab_size, embedding_size = embedding.shape
model = Model(RUN_NAME, sequence_length, output_size,
vocab_size=vocab_size, embedding_size=embedding_size, embedding_name=embedding_name)
model.info = 'log loss optimization'
print(model.get_info())
model.save_info(LOG_DIR, RUN_NAME + '.txt')
sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
summary_writer = tf.summary.FileWriter(LOG_DIR, sess.graph)
sess.run(tf.global_variables_initializer(), feed_dict={model.embedding_placeholder: embedding})
print('started running: ' + RUN_NAME)
for train_step in range(100000):
train_data_batch, train_label_batch = get_random_batch(model.train_batch_size,
data=train_data, labels=train_labels)
sess.run([model.optimize], feed_dict={model._sequence_placeholder: train_data_batch,
model._target_placeholder: train_label_batch,
model._dropout_keep_prob_placeholder: model.dropout_keep_prob})
# if train_step != 0 and train_step % 500 == 0:
if train_step % 500 == 0:
evaluate_test_set(model, sess, test_data, test_labels, train_step, summary_writer)
if __name__ == '__main__':
main(embedding_name='googlenews300')