def main(_):
if FLAGS.model == 'dmn':
word2vec = load_glove(FLAGS.glove_size)
words = WordTable(word2vec, FLAGS.glove_size)
from models.old.dmn import DMN
elif FLAGS.model == 'dmn+':
words = WordTable()
from models.new.dmn_plus import DMN
elif FLAGS.model == 'mc_dmn+':
word2vec = load_glove(FLAGS.glove_size)
words = WordTable(word2vec, FLAGS.glove_size)
from models.new.mc_dmn_plus import DMN
elif FLAGS.model == 'dmn_embed':
words = WordTable()
from models.old.dmn_embedding import DMN
else:
print('Unknown model type: %s' % FLAGS.model)
return
# Read data
train = read_babi(FLAGS.data_dir, FLAGS.task, 'train', FLAGS.batch_size,
words)
test = read_babi(FLAGS.data_dir, FLAGS.task, 'test', FLAGS.batch_size,
words)
val = train.split_dataset(FLAGS.val_ratio)
FLAGS.max_sent_size, FLAGS.max_ques_size, FLAGS.max_fact_count = get_max_sizes(
train, test, val)
print('Word count: %d, Max sentence len : %d' %
(words.vocab_size, FLAGS.max_sent_size))
# Modify save dir
FLAGS.save_dir += '/task_%d/' % FLAGS.task
if not os.path.exists(FLAGS.save_dir):
os.makedirs(FLAGS.save_dir) #, exist_ok=True)
with tf.Session() as sess:
model = DMN(FLAGS, words)
sess.run(tf.initialize_all_variables())
if FLAGS.test:
model.load(sess)
model.eval(sess, test, name='Test')
else:
if FLAGS.load: model.load(sess)
model.train(sess, train, val)
def main(_):
train_ds, test_ds = read_data.read_babi(FLAGS.batch_size, FLAGS.data_dir,
FLAGS.task)
train_ds, val_ds = read_data.split_val(train_ds, FLAGS.val_ratio)
train_ds.name, val_ds.name, test_ds.name = 'train', 'val', 'test'
FLAGS.vocab_size = test_ds.vocab_size
FLAGS.max_sent_size, FLAGS.max_ques_size = read_data.get_max_sizes(
train_ds, val_ds, test_ds)
# FIXME : adhoc for now!
FLAGS.max_sent_size = max(FLAGS.max_sent_size, FLAGS.max_ques_size)
FLAGS.train_num_batches = train_ds.num_batches
FLAGS.val_num_batches = val_ds.num_batches
FLAGS.test_num_batches = test_ds.num_batches
if not os.path.exists(FLAGS.save_dir):
os.mkdir(FLAGS.save_dir)
if FLAGS.linear_start:
FLAGS.num_epochs = FLAGS.ls_num_epochs
FLAGS.init_lr = FLAGS.ls_init_lr
if FLAGS.draft:
FLAGS.num_layers = 1
FLAGS.num_epochs = 1
FLAGS.eval_period = 1
FLAGS.ls_duration = 1
FLAGS.train_num_batches = 1
FLAGS.test_num_batches = 1
FLAGS.save_period = 1
pprint(FLAGS.__flags)
graph = tf.Graph()
model = N2NModel(graph, FLAGS)
with tf.Session(graph=graph) as sess:
sess.run(tf.initialize_all_variables())
if FLAGS.train:
writer = tf.train.SummaryWriter(FLAGS.log_dir, sess.graph)
if FLAGS.load:
model.load(sess)
model.train(sess, writer, train_ds, val_ds)
else:
model.load(sess)
model.eval(sess, test_ds)
def main(_):
train_ds, test_ds = read_data.read_babi(FLAGS.batch_size, FLAGS.data_dir, FLAGS.task)
train_ds, val_ds = read_data.split_val(train_ds, FLAGS.val_ratio)
train_ds.name, val_ds.name, test_ds.name = 'train', 'val', 'test'
FLAGS.vocab_size = test_ds.vocab_size
FLAGS.max_sent_size, FLAGS.max_ques_size = read_data.get_max_sizes(train_ds, val_ds, test_ds)
# FIXME : adhoc for now!
FLAGS.max_sent_size = max(FLAGS.max_sent_size, FLAGS.max_ques_size)
FLAGS.train_num_batches = train_ds.num_batches
FLAGS.val_num_batches = val_ds.num_batches
FLAGS.test_num_batches = test_ds.num_batches
if not os.path.exists(FLAGS.save_dir):
os.mkdir(FLAGS.save_dir)
if FLAGS.linear_start:
FLAGS.num_epochs = FLAGS.ls_num_epochs
FLAGS.init_lr = FLAGS.ls_init_lr
if FLAGS.draft:
FLAGS.num_layers = 1
FLAGS.num_epochs = 1
FLAGS.eval_period = 1
FLAGS.ls_duration = 1
FLAGS.train_num_batches = 1
FLAGS.test_num_batches = 1
FLAGS.save_period = 1
pprint(FLAGS.__flags)
graph = tf.Graph()
model = N2NModel(graph, FLAGS)
with tf.Session(graph=graph) as sess:
sess.run(tf.initialize_all_variables())
if FLAGS.train:
writer = tf.train.SummaryWriter(FLAGS.log_dir, sess.graph)
if FLAGS.load:
model.load(sess)
model.train(sess, writer, train_ds, val_ds)
else:
model.load(sess)
model.eval(sess, test_ds)
def main(_):
if FLAGS.model == 'dmn':
word2vec = load_glove(FLAGS.glove_size)
words = WordTable(word2vec, FLAGS.glove_size)
from models.old.dmn import DMN
elif FLAGS.model == 'dmn+':
words = WordTable()
from models.new.dmn_plus import DMN
elif FLAGS.model == 'dmn_embed':
words = WordTable()
from models.old.dmn_embedding import DMN
else:
print('Unknown model type: %s' % FLAGS.model)
return
# Read data
train = read_babi(FLAGS.data_dir, FLAGS.task, 'train', FLAGS.batch_size, words)
test = read_babi(FLAGS.data_dir, FLAGS.task, 'test', FLAGS.batch_size, words)
val = train.split_dataset(FLAGS.val_ratio)
FLAGS.max_sent_size, FLAGS.max_ques_size, FLAGS.max_fact_count = get_max_sizes(train, test, val)
print('Word count: %d, Max sentence len : %d' % (words.vocab_size, FLAGS.max_sent_size))
# Modify save dir
FLAGS.save_dir += '/task_%d/' % FLAGS.task
if not os.path.exists(FLAGS.save_dir):
os.makedirs(FLAGS.save_dir, exist_ok=True)
with tf.Session() as sess:
model = DMN(FLAGS, words)
sess.run(tf.initialize_all_variables())
if FLAGS.test:
model.load(sess)
model.eval(sess, test, name='Test')
else:
if FLAGS.load: model.load(sess)
model.train(sess, train, val)