from preprocess import Preprocess
from word2vec import Word2Vec
from news_embedding import NewsEmbeding
from han import HAN
if __name__ == '__main__':
PREPROCESSED = True
EMBEDDING_TRAINED = False
WORD_EMBEDDING_READY = False
preprocess = Preprocess()
if not PREPROCESSED:
preprocess.preprocess()
preprocess.load_data()
word2vec = Word2Vec(preprocess.data_dict)
if not EMBEDDING_TRAINED:
print('training word2vec...')
word2vec.train_model()
word2vec.load_model()
news_emb = NewsEmbeding(word2vec.model, preprocess.data_dict)
if not WORD_EMBEDDING_READY:
news_emb.embed()
news_emb.load_embeddings()
news_emb.get_max_corpus_date_count()
news_emb.pad_embeddings()
#print(news_emb.embedict_padded['AAPL'])
HAN = HAN(news_emb.emb_node_num, news_emb.date_num)
if FLAGS.run_type == "train":
print("Training...\n")
# create new graph set as default
with tf.Graph().as_default():
session_conf = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
session_conf.gpu_options.allocator_type = "BFC"
# create new session set it as default
sess = tf.Session(config=session_conf)
with sess.as_default():
# create new cnn model
han = HAN(max_seq_len=imdb.max_seq_len,
max_sent_len=imdb.max_sent_len,
num_classes=len(y_test[0]),
vocab_size=imdb.vocab_size,
embedding_size=FLAGS.embedding_dim,
max_grad_norm=FLAGS.max_grad_norm,
dropout_keep_proba=FLAGS.dropout_keep_proba,
learning_rate=FLAGS.learning_rate)
global_step = tf.Variable(0, name="global_step", trainable=False)
tvars = tf.trainable_variables()
grads, global_norm = tf.clip_by_global_norm(
tf.gradients(han.loss, tvars), han.max_grad_norm)
optimizer = tf.train.AdamOptimizer(han.learning_rate)
train_op = optimizer.apply_gradients(zip(grads, tvars),
name="train_op",
global_step=global_step)
# checkpoint model
saver = tf.train.Saver(tf.global_variables(),
EPOCH = 40
if __name__ == "__main__":
bf = BatchFeeder('train', 64, 15, 50)
tf.reset_default_graph()
config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=True)
with tf.device('/device:GPU:0'):
model = HAN(vocab_size=len(bf.vocab_encode_dict.keys()) + 1,
embedding_size=200,
classes=10,
word_cell=tf.nn.rnn_cell.GRUCell(50, name='word-gru'),
sentence_cell=tf.nn.rnn_cell.GRUCell(50,
name='sentence-gru'),
word_context_size=100,
sentence_context_size=100)
saver = tf.train.Saver()
with tf.Session(config=config) as sess:
writer = tf.summary.FileWriter('./han_graph',
graph=tf.get_default_graph())
sess.run(tf.global_variables_initializer())
for i in range(EPOCH):
for encoded_data, document_length_mask, sentence_length_mask, labels in bf:
feed_dict = {
model.inputs: encoded_data,
model.word_length: sentence_length_mask,
epochs = 10
print('Loading data...')
(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=max_features)
print(len(x_train), 'train sequences')
print(len(x_test), 'test sequences')
print('Pad sequences (samples x #sentence x #word)...')
x_train = sequence.pad_sequences(x_train, maxlen=maxlen_sentence * maxlen_word)
x_test = sequence.pad_sequences(x_test, maxlen=maxlen_sentence * maxlen_word)
x_train = x_train.reshape((len(x_train), maxlen_sentence, maxlen_word))
x_test = x_test.reshape((len(x_test), maxlen_sentence, maxlen_word))
print('x_train shape:', x_train.shape)
print('x_test shape:', x_test.shape)
print('Build model...')
model = HAN(maxlen_sentence, maxlen_word, max_features, embedding_dims)
model.compile('adam', 'binary_crossentropy', metrics=['accuracy'])
print('Train...')
early_stopping = EarlyStopping(monitor='val_accuracy', patience=3, mode='max')
model.fit(x_train,
y_train,
batch_size=batch_size,
epochs=epochs,
callbacks=[early_stopping],
validation_data=(x_test, y_test))
print('Test...')
result = model.predict(x_test)
(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words=max_features)
except:
(x_train, y_train), (x_test, y_test) = load_data(num_words=max_features)
logger.info('padding...')
x_train = sequence.pad_sequences(x_train, maxlen=max_seqs*max_words)
x_test = sequence.pad_sequences(x_test, maxlen=max_seqs*max_words)
x_train = x_train.reshape((len(x_train), max_seqs, max_words))
x_test = x_test.reshape((len(x_test), max_seqs, max_words))
logger.info('train data shape is: {}'.format(x_train.shape))
logger.info('test data shape is: {}'.format(x_test.shape))
logger.info('build model...')
model = HAN(max_features=max_features, max_words=max_words, max_seqs=max_seqs, emb_dim=emb_dim).build_model()
model.compile('adam', 'binary_crossentropy', ['acc'])
logger.info('training...')
earlystop = EarlyStopping(patience=3, monitor='val_acc', mode='max')
model.fit(x_train, y_train,
callbacks=[earlystop],
batch_size=batch_size,
epochs=epochs,
validation_data=[x_test, y_test])
logger.info('test...')
pred = model.predict(x_test)
logger.info(pred[:10])
logger.info(y_test[:10])