print("")
# Data Preparation
# ==================================================
# Load data
print("Loading data...")
x_text, y, category = data_helpers.load_data_and_labels(FLAGS.data_file)
print(category)
y = np.array(data_helpers.transform_labels(y, category))
# Build vocabulary
max_document_length = max([len(x) for x in x_text])
# vocab_processor = learn.preprocessing.VocabularyProcessor(max_document_length)
x = np.array(data_helpers.fit_transform(x_text, FLAGS.max_data_length))
# Randomly shuffle data
np.random.seed(10)
shuffle_indices = np.random.permutation(np.arange(len(y)))
x_shuffled = x[shuffle_indices, :]
y_shuffled = y[shuffle_indices]
# Split train/test set
# TODO: This is very crude, should use cross-validation
dev_sample_index = -1 * int(FLAGS.dev_sample_percentage * float(len(y)))
x_train, x_dev = x_shuffled[:dev_sample_index], x_shuffled[dev_sample_index:]
y_train, y_dev = y_shuffled[:dev_sample_index], y_shuffled[dev_sample_index:]
del x, y, x_shuffled, y_shuffled
# Data Preparation
# ==================================================
# Load data
print("Loading data...")
x_text, x_char, y_text, handcraft, bag_of_entity, vocab, w2v, data_pos, data_entity, pos_vocab, entity_vocab, train_test_dev, class_label = data_helpers.load_data_and_labels(FLAGS.Training_Data, FLAGS.Test_data_entity)
# Build vocabulary
# max_document_length = max([len(x.split(" ")) for x in x_text])
max_document_length = 39
char_length = 50
vocab_processor = np.zeros([len(x_text), max_document_length+1])
x = data_helpers.fit_transform(x_text, vocab_processor, vocab)
# Build vocabulary
vocab_processor_pos = np.zeros([len(data_pos), max_document_length+1])
x_pos = data_helpers.fit_transform_pos(data_pos, vocab_processor_pos)
# Build vocabulary
vocab_processor_entity = np.zeros([len(data_entity), max_document_length+1])
x_entity = data_helpers.fit_transform_pos(data_entity, vocab_processor_entity)
x_shuf, x_char_shuf, y_shuf, handcraft_shuf, bag_of_entity_shuf, x_pos_shuf, x_entity_shuf = x, x_char, y_text, handcraft, bag_of_entity, x_pos, x_entity
offset = int(x_shuf.shape[0] * 0)
x_shuffled, x_char_shuffled, y_shuffled, handcraft_shuffled, x_pos_shuffled, x_entity_shuffled, bag_of_entity_shuffled = x_shuf[offset:], x_char_shuf[offset:], y_shuf[offset:], handcraft_shuf[offset:], x_pos_shuf[offset:], x_entity_shuf[offset:], bag_of_entity_shuf[offset:]
# Parameters
# ==================================================
# Eval Parameters
tf.flags.DEFINE_string("checkpoint_dir", T.checkpoint_dir, "Checkpoint directory from training run")
FLAGS = tf.flags.FLAGS
x_raw, y_test = data_helpers.load_data_and_labels("./test/pos","./test/neg")
y_test = np.argmax(y_test, axis=1)
# Map data into vocabulary
vocab_processor = pickle.load( open(os.path.join(T.out_dir,"vocab"), "rb" ) )
x_list = [x.split(" ") for x in x_raw]
x_test = np.array(list(data_helpers.fit_transform(vocab_processor, x_list, max_document_length, n_gram)))
print("\nEvaluating...\n")
checkpoint_file = tf.train.latest_checkpoint(T.checkpoint_dir)
# Evaluation
# ==================================================
graph = tf.Graph()
with graph.as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
# Load the saved meta graph and restore variables
saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
FLAGS = tf.flags.FLAGS
# Data Preparation
# ==================================================
# Load data
print("Loading data...")
x_text, x_char, x_ib, x_pos, x_mtopic, x_features, x_spId, x_hub, y_text, handcraft, vocab, w2v, pos_vocab, train_dev_index, data_pred, class_label = data_helpers.load_data_and_labels(
FLAGS.Training_Data, FLAGS.Test_Data)
# Build vocabulary
max_document_length = max([len(x1.split(" ")) for x1 in x_text[0]])
print max_document_length
# max_document_length = 25
vocab_processor = np.zeros([len(x_text[0]), max_document_length])
x = data_helpers.fit_transform(x_text[0], vocab_processor, vocab)
vocab_processor = np.zeros([len(x_text[0]), max_document_length])
x_utt1 = data_helpers.fit_transform(x_text[1], vocab_processor, vocab)
vocab_processor = np.zeros([len(x_text[0]), max_document_length])
x_utt2 = data_helpers.fit_transform(x_text[2], vocab_processor, vocab)
vocab_processor = np.zeros([len(x_text[0]), max_document_length])
x_utt3 = data_helpers.fit_transform(x_text[3], vocab_processor, vocab)
x_shuf, x_utt1_shuf, x_utt2_shuf, x_utt3_shuf, x_char_shuf, x_char1_shuf, x_char2_shuf, x_ib1_shuf, x_ib2_shuf, x_pos0_shuf, x_pos1_shuf, x_pos2_shuf, x_pos3_shuf, x_spId0_shuf, x_spId1_shuf, x_spId2_shuf, x_spId3_shuf, x_hub0_shuf, x_hub1_shuf, x_hub2_shuf, x_mtp0_shuf, x_mtp1_shuf, x_mtp2_shuf, x_feat0_shuf, x_feat1_shuf, x_feat2_shuf, x_feat3_shuf, y_shuf, handcraft_shuf = \
x, x_utt1, x_utt2, x_utt3, x_char[0], x_char[1], x_char[2], x_ib[0], x_ib[1], x_pos[0], x_pos[1], x_pos[2], x_pos[3], x_spId[0], x_spId[1], x_spId[2], x_spId[3], x_hub[0], x_hub[1], x_hub[2], x_mtopic[0], x_mtopic[1], x_mtopic[2], x_features[0], x_features[1], x_features[2], x_features[3], y_text, handcraft
vocab_processor = np.zeros([len(x_text[0]), max_document_length])
x_pos0_shuf = data_helpers.fit_transform_pos(x_pos0_shuf, vocab_processor)
vocab_processor = np.zeros([len(x_text[0]), max_document_length])
x_pos1_shuf = data_helpers.fit_transform_pos(x_pos1_shuf, vocab_processor)
def __init__(self):
x_text, y = data_helpers.load_data_and_labels("./train/pos","./train/neg")
# Build vocabulary
x_list = [x.split(" ") for x in x_text]
vocab_processor = data_helpers.n_grams(x_list, max_word_cnt, n_gram)
print 'feed finished'
x = np.array(data_helpers.fit_transform(vocab_processor, x_list, max_document_length, n_gram))
# print x[0]
print 'fit transform finished'
# Randomly shuffle data
np.random.seed(10)
shuffle_indices = np.random.permutation(np.arange(len(y)))
x_shuffled = x[shuffle_indices]
y_shuffled = y[shuffle_indices]
# Split train/test set
# TODO: This is very crude, should use cross-validation
x_train, x_dev = x_shuffled[:-1000], x_shuffled[-1000:]
y_train, y_dev = y_shuffled[:-1000], y_shuffled[-1000:]
print("Vocabulary Size: {:d}".format(len(vocab_processor)))
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
# Training
# ==================================================
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
cnn = MLP(
sequence_length=x_train.shape[1],
num_classes=2,
vocab_size=len(vocab_processor),
embedding_size=FLAGS.embedding_dim,
filter_sizes=list(map(int, FLAGS.filter_sizes.split(","))),
num_filters = FLAGS.num_filters,
l2_reg_lambda=FLAGS.l2_reg_lambda)
# Define Training procedure
global_step = tf.Variable(0, name="global_step", trainable=False)
starter_learning_rate = 1e-3
learning_rate = tf.train.exponential_decay(starter_learning_rate, global_step,
3000, 0.96, staircase=True)
optimizer = tf.train.AdamOptimizer(starter_learning_rate)
grads_and_vars = optimizer.compute_gradients(cnn.loss)
train_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step)
# Keep track of gradient values and sparsity (optional)
grad_summaries = []
for g, v in grads_and_vars:
if g is not None:
grad_hist_summary = tf.histogram_summary("{}/grad/hist".format(v.name), g)
sparsity_summary = tf.scalar_summary("{}/grad/sparsity".format(v.name), tf.nn.zero_fraction(g))
grad_summaries.append(grad_hist_summary)
grad_summaries.append(sparsity_summary)
grad_summaries_merged = tf.merge_summary(grad_summaries)
# Output directory for models and summaries
timestamp = str(int(time.time()))
self.out_dir = os.path.abspath(os.path.join(os.path.curdir, "runs", timestamp))
print("Writing to {}\n".format(self.out_dir))
# Summaries for loss and accuracy
loss_summary = tf.scalar_summary("loss", cnn.loss)
acc_summary = tf.scalar_summary("accuracy", cnn.accuracy)
# Train Summaries
train_summary_op = tf.merge_summary([loss_summary, acc_summary, grad_summaries_merged])
train_summary_dir = os.path.join(self.out_dir, "summaries", "train")
train_summary_writer = tf.train.SummaryWriter(train_summary_dir, sess.graph)
# Dev summaries
dev_summary_op = tf.merge_summary([loss_summary, acc_summary])
dev_summary_dir = os.path.join(self.out_dir, "summaries", "dev")
dev_summary_writer = tf.train.SummaryWriter(dev_summary_dir, sess.graph)
# Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it
self.checkpoint_dir = os.path.abspath(os.path.join(self.out_dir, "checkpoints"))
checkpoint_prefix = os.path.join(self.checkpoint_dir, "model")
if not os.path.exists(self.checkpoint_dir):
os.makedirs(self.checkpoint_dir)
saver = tf.train.Saver(tf.all_variables())
# Write vocabulary
pickle.dump(vocab_processor, open(os.path.join(self.out_dir,"vocab"), "wb" ) )
# Initialize all variables
sess.run(tf.initialize_all_variables())
def train_step(x_batch, y_batch):
"""
A single training step
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
cnn.dropout_keep_prob: FLAGS.dropout_keep_prob
}
_, step, summaries, loss, accuracy = sess.run(
[train_op, global_step, train_summary_op, cnn.loss, cnn.accuracy],
feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
train_summary_writer.add_summary(summaries, step)
def dev_step(x_batch, y_batch, writer=None):
"""
Evaluates model on a dev set
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
cnn.dropout_keep_prob:1
}
step, summaries, loss, accuracy = sess.run(
[global_step, dev_summary_op, cnn.loss, cnn.accuracy],
feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
if writer:
writer.add_summary(summaries, step)
# Generate batches
t = list(zip(x_train, y_train))
batches = data_helpers.batch_iter(
t, FLAGS.batch_size, FLAGS.num_epochs)
# Training loop. For each batch...
for batch in batches:
x_batch, y_batch = zip(*batch)
train_step(x_batch, y_batch)
current_step = tf.train.global_step(sess, global_step)
if current_step % FLAGS.evaluate_every == 0:
print("\nEvaluation:")
dev_step(x_dev, y_dev, writer=dev_summary_writer)
print("")
if current_step % FLAGS.checkpoint_every == 0:
path = saver.save(sess, checkpoint_prefix, global_step=current_step)
print("Saved model checkpoint to {}\n".format(path))