def format20n_input(trained_classes, untrained_classes):
datasets1 = data_helpers.get_datasets_20newsgroup(
subset="test",
categories=trained_classes,
remove=('headers', 'footers', 'quotes'))
x_raw1, y_raw1 = data_helpers.load_data_labels(datasets1)
y_test1 = np.argmax(y_raw1, axis=1)
labels1 = datasets1['target_names']
dataset2 = data_helpers.get_datasets_20newsgroup(
subset="test",
categories=untrained_classes,
remove=('headers', 'footers', 'quotes'))
x_raw2, y_test2 = data_helpers.load_data_labels(dataset2)
y_test2 = np.add(np.argmax(y_test2, axis=1), len(trained_classes))
labels2 = dataset2['target_names']
x_net = x_raw1 + x_raw2
y_net = np.append(y_test1, y_test2)
labels = labels1 + labels2
return (x_net, y_net, labels)
def load_data_and_labels(self):
# Load data
print("Loading embedding and dataset...")
embedding_name = None
dataset_name = self.cfg["datasets"]["default"]
if FLAGS.enable_word_embeddings and self.cfg['word_embeddings'][
'default'] is not None:
embedding_name = self.cfg['word_embeddings']['default']
embedding_dimension = self.cfg['word_embeddings'][embedding_name][
'dimension']
else:
embedding_dimension = FLAGS.embedding_dim
datasets = None
if dataset_name == "mrpolarity":
datasets = data_helpers.get_datasets_mrpolarity(
self.cfg["datasets"][dataset_name]["positive_data_file"]
["path"], self.cfg["datasets"][dataset_name]
["negative_data_file"]["path"])
elif dataset_name == "20newsgroup":
datasets = data_helpers.get_datasets_20newsgroup(
subset="train",
categories=self.cfg["datasets"][dataset_name]["categories"],
shuffle=self.cfg["datasets"][dataset_name]["shuffle"],
random_state=self.cfg["datasets"][dataset_name]
["random_state"])
elif dataset_name == "localdata":
datasets = data_helpers.get_datasets_localdata(
container_path=self.cfg["datasets"][dataset_name]
["container_path"],
categories=self.cfg["datasets"][dataset_name]["categories"],
shuffle=self.cfg["datasets"][dataset_name]["shuffle"],
random_state=self.cfg["datasets"][dataset_name]
["random_state"])
print("Loaded dataset: {}. [Embedding: {}].\nLoading labels".format(
dataset_name, embedding_name))
x_text, y = data_helpers.load_data_labels(datasets)
self.x_text = x_text
self.y = y
self.embedding_name = embedding_name
self.embedding_dimension = embedding_dimension
self.dataset_name = dataset_name
print("Done load_data_and_labels()")
dataset = "amazon"
#trained on 20 classes and tested on 20 classes
training_classes = [
'Amplifier', 'Automotive', 'Battery', 'Beauty', 'Cable', 'Camera',
'CDPlayer', 'Clothing', 'Computer', 'Conditioner', 'Fan', 'Flashlight',
'Graphics Card', 'Headphone', 'Home Improvement', 'Jewelry', 'Kindle',
'Kitchen', 'Lamp', 'Luggage'
]
#dataset = "20newsgroup"
if dataset == "20newsgroup":
datasets = data_helpers.get_datasets_20newsgroup(
subset="test",
categories=training_classes,
remove=('headers', 'footers', 'quotes'))
x_raw, y_test = data_helpers.load_data_labels(datasets)
y_test = np.argmax(y_test, axis=1)
else:
datasets = data_helpers.get_datasets_localdata(
"./amazon/test",
categories=training_classes) # TODO: tweak parameters in the future
x_raw, y_test = data_helpers.load_data_labels(
datasets) # text is stored in x_test; # labels are stored in y
y_test = np.argmax(y_test, axis=1)
print("length of y_test", y_test[:20])
print("Total number of test examples: {}".format(len(y_test)))
print(datasets['target_names'])
print("{}={}".format(attr.upper(), value))
print("")
datasets = None
# CHANGE THIS: Load data. Load your own data here
dataset_name = cfg["datasets"]["default"]
if FLAGS.eval_train:
if dataset_name == "mrpolarity":
datasets = data_helpers.get_datasets_mrpolarity(
cfg["datasets"][dataset_name]["positive_data_file"]["path"],
cfg["datasets"][dataset_name]["negative_data_file"]["path"])
elif dataset_name == "20newsgroup":
datasets = data_helpers.get_datasets_20newsgroup(
subset="test",
categories=cfg["datasets"][dataset_name]["categories"],
shuffle=cfg["datasets"][dataset_name]["shuffle"],
random_state=cfg["datasets"][dataset_name]["random_state"])
elif dataset_name == "abstract":
datasets = data_helpers.get_datasets_abstract("data/")
elif dataset_name == 'intents':
datasets = data_helpers.get_datasets_intentst("data/")
x_text, y = data_helpers.load_data_labels(datasets)
x_raw, y_test = data_helpers.load_data_labels(datasets)
y_test = np.argmax(y_test, axis=1)
print("Total number of test examples: {}".format(len(y_test)))
else:
if dataset_name == "mrpolarity":
x_raw = [
"a masterpiece four years in the making", "everything is off."
]
'sci.crypt',
'sci.electronics',
'sci.med',
'sci.space',
'soc.religion.christian',
'talk.politics.guns',
'talk.politics.mideast',
'talk.politics.misc', 'talk.religion.misc']
training_classes = ['comp.graphics', 'alt.atheism', 'comp.sys.mac.hardware', 'misc.forsale', 'rec.autos']
# load data
print("Loading data...")
if dataset == "20newsgroup":
datasets = data_helpers.get_datasets_20newsgroup(subset='train', categories=training_classes, remove=()) # TODO: use the remove parameter
x_text, y_train = data_helpers.load_data_labels_remove_SW(datasets)
else:
dataset = data_helpers.get_datasets_localdata("./data/20newsgroup", categories=None) # TODO: tweak parameters in the future
x_text, y_train = data_helpers.load_data_labels(dataset) # text is stored in x_test; # labels are stored in y
# Build vocabulary
max_document_length = max([len(x.split(" ")) for x in x_text]) # TODO: should be hardcoded to save time
print("Max document length: {}".format(max_document_length))
vocab_processor = learn.preprocessing.VocabularyProcessor(max_document_length)
x_train = np.array(list(vocab_processor.fit_transform(x_text)))
# Randomly shuffle data
# np.random.seed(10)
# shuffle_indices = np.random.permutation(np.arange(len(y)))
def main():
import time
start_time = time.time()
FLAGS = flagClass()
with open("config.yml", 'r') as ymlfile:
cfg = yaml.load(ymlfile)
dataset_name = cfg["datasets"]["default"]
if FLAGS.enable_word_embeddings and cfg['word_embeddings'][
'default'] is not None:
embedding_name = cfg['word_embeddings']['default']
embedding_dimension = cfg['word_embeddings'][embedding_name][
'dimension']
else:
embedding_dimension = FLAGS.embedding_dim
# Data Preparation
# ==================================================
# Load data
print("Loading data...")
datasets = None
if dataset_name == "mrpolarity":
datasets = data_helpers.get_datasets_mrpolarity(
cfg["datasets"][dataset_name]["positive_data_file"]["path"],
cfg["datasets"][dataset_name]["negative_data_file"]["path"])
elif dataset_name == 'spamham':
datasets = data_helpers.get_datasets_mrpolarity(
cfg["datasets"][dataset_name]["spam_file"]["path"],
cfg["datasets"][dataset_name]["ham_file"]["path"])
elif dataset_name == "20newsgroup":
datasets = data_helpers.get_datasets_20newsgroup(
subset="train",
categories=cfg["datasets"][dataset_name]["categories"],
shuffle=cfg["datasets"][dataset_name]["shuffle"],
random_state=cfg["datasets"][dataset_name]["random_state"])
elif dataset_name == "dbpedia":
datasets = data_helpers.get_datasets_dbpedia(
cfg["datasets"][dataset_name]["train_file"]["path"],
cfg["datasets"][dataset_name]["train_file"]["limit"])
elif dataset_name == "email":
datasets = data_helpers.get_datasets_email(
container_path=cfg["datasets"][dataset_name]["container_path"],
categories=cfg["datasets"][dataset_name]["categories"],
shuffle=cfg["datasets"][dataset_name]["shuffle"],
random_state=cfg["datasets"][dataset_name]["random_state"])
elif dataset_name == "localdata":
datasets = data_helpers.get_datasets_localdata(
container_path=cfg["datasets"][dataset_name]["container_path"],
categories=cfg["datasets"][dataset_name]["categories"],
shuffle=cfg["datasets"][dataset_name]["shuffle"],
random_state=cfg["datasets"][dataset_name]["random_state"])
x_text, y = data_helpers.load_data_labels(datasets)
# Build vocabulary
# To limit memory usage, you can cut off input text to first 40 words
# Other research has shown that first 40 words in text (IMDB dataset?)
# were representative of the content of the sentence for classification
# purposes - Comment out one of the two lines below
# max_document_length = max([len(x.split(" ")) for x in x_text])
max_document_length = 40 # read up to 40 words from each sentence
vocab_processor = learn.preprocessing.VocabularyProcessor(
max_document_length)
x = np.array(list(vocab_processor.fit_transform(x_text)))
# 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:]
print("Vocabulary Size: {:d}".format(len(vocab_processor.vocabulary_)))
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
print('Sequence_length={}'.format(x_train.shape[1]))
# 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 = TextCNN(sequence_length=x_train.shape[1],
num_classes=y_train.shape[1],
vocab_size=len(vocab_processor.vocabulary_),
embedding_size=embedding_dimension,
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)
optimizer = tf.train.AdamOptimizer(cnn.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.summary.histogram(
"{}/grad/hist".format(v.name), g)
sparsity_summary = tf.summary.scalar(
"{}/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.summary.merge(grad_summaries)
# Output directory for models and summaries
timestamp = str(int(time.time()))
out_dir = os.path.abspath(
os.path.join(os.path.curdir, "runs", timestamp))
print("Writing to {}\n".format(out_dir))
# Summaries for loss and accuracy
loss_summary = tf.summary.scalar("loss", cnn.loss)
acc_summary = tf.summary.scalar("accuracy", cnn.accuracy)
# Train Summaries
train_summary_op = tf.summary.merge(
[loss_summary, acc_summary, grad_summaries_merged])
train_summary_dir = os.path.join(out_dir, "summaries", "train")
train_summary_writer = tf.summary.FileWriter(
train_summary_dir, sess.graph)
# Dev summaries
dev_summary_op = tf.summary.merge([loss_summary, acc_summary])
dev_summary_dir = os.path.join(out_dir, "summaries", "dev")
dev_summary_writer = tf.summary.FileWriter(dev_summary_dir,
sess.graph)
# Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it
checkpoint_dir = os.path.abspath(
os.path.join(out_dir, "checkpoints"))
checkpoint_prefix = os.path.join(checkpoint_dir, "model")
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
saver = tf.train.Saver(tf.global_variables(),
max_to_keep=FLAGS.num_checkpoints)
# Write vocabulary
vocab_processor.save(os.path.join(out_dir, "vocab"))
# Initialize all variables
sess.run(tf.global_variables_initializer())
if FLAGS.enable_word_embeddings and cfg['word_embeddings'][
'default'] is not None:
vocabulary = vocab_processor.vocabulary_
initW = None
if embedding_name == 'word2vec':
# load embedding vectors from the word2vec
print("Load word2vec file {}".format(
cfg['word_embeddings']['word2vec']['path']))
initW = data_helpers.load_embedding_vectors_word2vec(
vocabulary, cfg['word_embeddings']['word2vec']['path'],
cfg['word_embeddings']['word2vec']['binary'])
print("word2vec file has been loaded")
elif embedding_name == 'glove':
# load embedding vectors from the glove
print("Load glove file {}".format(
cfg['word_embeddings']['glove']['path']))
initW = data_helpers.load_embedding_vectors_glove(
vocabulary, cfg['word_embeddings']['glove']['path'],
embedding_dimension)
print("glove file has been loaded\n")
sess.run(cnn.W.assign(initW))
def train_step(x_batch, y_batch, learning_rate):
"""
A single training step
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
cnn.dropout_keep_prob: FLAGS.dropout_keep_prob,
cnn.learning_rate: learning_rate
}
_, 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}, learning_rate {:g}".
format(time_str, step, loss, accuracy, learning_rate))
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: 1.0
}
step, summaries, loss, accuracy, gr = sess.run([
global_step, dev_summary_op, cnn.loss, cnn.accuracy,
cnn.grad
], feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}, gr {}".format(
time_str, step, loss, accuracy, gr))
if writer:
writer.add_summary(summaries, step)
# Generate batches
batches = data_helpers.batch_iter(list(zip(x_train, y_train)),
FLAGS.batch_size,
FLAGS.num_epochs)
print("Number of epochs: {}".format(FLAGS.num_epochs))
num_batches_per_epoch = int(
(len(list(zip(x_train, y_train))) - 1) / FLAGS.batch_size) + 1
print("Batches per epoch: {}".format(num_batches_per_epoch))
print("Batch size: {}".format(FLAGS.batch_size))
# It uses dynamic learning rate with a high value at the beginning to speed up the training
max_learning_rate = 0.005
min_learning_rate = 0.0001
decay_speed = FLAGS.decay_coefficient * len(
y_train) / FLAGS.batch_size
# Training loop. For each batch...
counter = 0
for batch in batches:
learning_rate = min_learning_rate + (
max_learning_rate - min_learning_rate) * math.exp(
-counter / decay_speed)
counter += 1
x_batch, y_batch = zip(*batch)
train_step(x_batch, y_batch, learning_rate)
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))
print("runtime was " + str(time.time() - start_time))
FLAGS = tf.flags.FLAGS
FLAGS._parse_flags()
print("\nParameters:")
for attr, value in sorted(FLAGS.__flags.items()):
print("{}={}".format(attr.upper(), value))
print("")
# Data Preparation
# ==================================================
# Load data
print("Loading data...")
#import newsgroups data
datasets = data_helpers.get_datasets_20newsgroup(subset="train",
categories=['comp.graphics', 'comp.os.ms-windows.misc',
'comp.sys.ibm.pc.hardware', 'comp.sys.mac.hardware',
'comp.windows.x'],
shuffle=True)
x_text, y = data_helpers.load_data_and_labels(datasets)
# Build vocabulary
max_document_length = max([len(x.split(" ")) for x in x_text])
vocab_processor = learn.preprocessing.VocabularyProcessor(max_document_length)
x = np.array(list(vocab_processor.fit_transform(x_text)))
# 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]