allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement,
gpu_options=tf.GPUOptions(per_process_gpu_memory_fraction=0.8))
sess = tf.Session(config=session_conf)
with sess.as_default():
cnn = TextCNN(
sequence_length=FLAGS.
max_sequence_length, ##cnt the word num of the sentence
num_classes=y_train_shuffled.shape[1],
vocab_size=len(vocabulary),
embedding_size=embedding_dim,
filter_sizes=list(map(int, FLAGS.filter_sizes.split(","))),
num_filters=FLAGS.num_filters,
l2_reg_lambda=FLAGS.l2_reg_lambda)
if FLAGS.use_pretrain:
cnn.assign_embedding(sess, pretrained_embedding)
# Define Training procedure
global_step = tf.Variable(0, name="global_step",
trainable=False) #record the global step
#learning_rate = tf.train.exponential_decay(FLAGS.learning_rate,global_step=global_step,decay_steps=FLAGS.num_epochs,decay_rate=0.03)
optimizer = tf.train.AdamOptimizer(1e-3)
grads_and_vars = optimizer.compute_gradients(
cnn.loss) ##calculate gradient
train_op = optimizer.apply_gradients(grads_and_vars,
global_step=global_step,
name="train_op")
# Output directory for models and summaries
timestamp = str(int(time.time()))
sess = tf.Session(config=session_conf)
print(np.array(y_train))
with sess.as_default():
cnn = TextCNN(
sequence_length_arg1=x_train_arg1_word.shape[1],
sequence_length_arg2=x_train_arg2_word.shape[1],
num_classes=np.array(y_train).shape[1],
pos_vocab_size=len(pos_vocab),
word_vocab_size=len(word_vocab),
pos_embedding_size=pos_embedding_dim ,
word_embedding_size=word_embedding_dim,
filter_sizes=list(map(int, FLAGS.filter_sizes.split(","))),
num_filters=FLAGS.num_filters,
l2_reg_lambda=FLAGS.l2_reg_lambda)
if FLAGS.use_pretrain:
cnn.assign_embedding(sess, pretrained_pos_embedding,pretrained_word_embedding)
#cnn.assign_word_embedding(sess,pretrained_word_embedding)
global_step = tf.Variable(0, name="global_step", trainable=False)
optimizer = tf.train.AdamOptimizer(learning_rate=1e-3)
grads_and_vars = optimizer.compute_gradients(cnn.loss)
train_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step, name='train_op')
# var_list1 = [var for var in tf.trainable_variables() if
# 'word_embedding' in var.op.name]
# var_list2 = [var for var in tf.trainable_variables() if
# 'word_embedding' not in var.op.name]
# opt1 = tf.train.AdamOptimizer(0.001)
# opt2 = tf.train.AdamOptimizer(0.004)
#