def train():
model_config = ModelConfig()
training_config = TrainConfig()
model_config.batch_size = 32
# Get model
model_fn = get_model_creation_fn(FLAGS.model_type)
# Create training directory.
train_dir = FLAGS.train_dir % (FLAGS.model_type, model_config.feat_type)
if not tf.gfile.IsDirectory(train_dir):
tf.logging.info("Creating training directory: %s", train_dir)
tf.gfile.MakeDirs(train_dir)
g = tf.Graph()
with g.as_default():
# Build the model.
model = model_fn(model_config, 'train')
model.build()
# Set up the learning rate.u
learning_rate = tf.constant(training_config.initial_learning_rate)
def _learning_rate_decay_fn(learn_rate, global_step):
return tf.train.exponential_decay(
learn_rate,
global_step,
decay_steps=training_config.decay_step,
decay_rate=training_config.decay_factor)
learning_rate_decay_fn = _learning_rate_decay_fn
train_op = tf.contrib.layers.optimize_loss(
loss=model.loss,
global_step=model.global_step,
learning_rate=learning_rate,
optimizer=training_config.optimizer,
clip_gradients=training_config.clip_gradients,
learning_rate_decay_fn=learning_rate_decay_fn)
# Set up the Saver for saving and restoring model checkpoints.
saver = tf.train.Saver(
max_to_keep=training_config.max_checkpoints_to_keep)
# create reader
reader = AttentionDataReader(batch_size=model_config.batch_size,
subset='trainval',
feat_type=model_config.feat_type)
# Run training.
training_util.train(train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver,
reader=reader,
feed_fn=model.fill_feed_dict)
def train():
training_config = TrainConfig()
# Create training directory.
train_dir = FLAGS.train_dir % ('v1', 'Fusion')
if not tf.gfile.IsDirectory(train_dir):
tf.logging.info("Creating training directory: %s", train_dir)
tf.gfile.MakeDirs(train_dir)
g = tf.Graph()
with g.as_default():
# Build the model.
model = RerankModel('train', version='v1', num_cands=5)
model.build()
# Set up the learning rate.u
learning_rate = tf.constant(training_config.initial_learning_rate)
def _learning_rate_decay_fn(learn_rate, global_step):
return tf.train.exponential_decay(
learn_rate,
global_step,
decay_steps=training_config.decay_step,
decay_rate=training_config.decay_factor,
staircase=False)
learning_rate_decay_fn = _learning_rate_decay_fn
train_op = tf.contrib.layers.optimize_loss(
loss=model.loss,
global_step=model.global_step,
learning_rate=learning_rate,
optimizer=training_config.optimizer,
clip_gradients=training_config.clip_gradients,
learning_rate_decay_fn=learning_rate_decay_fn)
# Set up the Saver for saving and restoring model checkpoints.
var_list = [
var for var in tf.global_variables() if 'Adam' not in var.name
]
saver = tf.train.Saver(
var_list, max_to_keep=training_config.max_checkpoints_to_keep)
# create reader
# reader = Reader(batch_size=32, subset='trainval', version='v1')
reader = Reader(batch_size=128 * 4, subset='trainval', version='v1')
# Run training.
training_util.train(train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver,
reader=reader,
feed_fn=model.fill_feed_dict,
debug_op=None)
def train():
model_config = ModelConfig()
training_config = TrainConfig()
# Get model
model_fn = get_model_creation_fn(FLAGS.model_type)
# Create training directory.
train_dir = FLAGS.train_dir % (FLAGS.model_trainset, FLAGS.model_type)
do_counter_sampling = FLAGS.version == 'v2'
if not tf.gfile.IsDirectory(train_dir):
tf.logging.info("Creating training directory: %s", train_dir)
tf.gfile.MakeDirs(train_dir)
g = tf.Graph()
with g.as_default():
# Build the model.
model = model_fn(model_config, phase='train')
model.build()
# Set up the learning rate
learning_rate = tf.constant(training_config.initial_learning_rate)
def _learning_rate_decay_fn(learn_rate, global_step):
return tf.train.exponential_decay(
learn_rate,
global_step,
decay_steps=training_config.decay_step,
decay_rate=training_config.decay_factor,
staircase=False)
learning_rate_decay_fn = _learning_rate_decay_fn
train_op = tf.contrib.layers.optimize_loss(
loss=model.loss,
global_step=model.global_step,
learning_rate=learning_rate,
optimizer=training_config.optimizer,
clip_gradients=training_config.clip_gradients,
learning_rate_decay_fn=learning_rate_decay_fn)
# Set up the Saver for saving and restoring model checkpoints.
saver = tf.train.Saver(
max_to_keep=training_config.max_checkpoints_to_keep)
# setup summaries
summary_op = tf.summary.merge_all()
# create reader
# model_name = os.path.split(train_dir)[1]
reader = Reader(
batch_size=32,
subset=FLAGS.model_trainset,
cst_file='vqa_replay_buffer/vqa_replay_low_rescore_prior_05_04.json')
# reader = Reader(batch_size=64,
# known_set='kprestval',
# unknown_set='kptrain', # 'kptrain'
# un_ratio=1,
# hide_label=False)
# Run training.
training_util.train(train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver,
reader=reader,
feed_fn=model.fill_feed_dict)
def train():
model_config = ModelConfig()
training_config = TrainConfig()
# Get model
# model_fn = get_model_creation_fn(FLAGS.model_type)
# Create training directory.
train_dir = FLAGS.train_dir % (FLAGS.version, FLAGS.model_type)
do_counter_sampling = FLAGS.version == 'v2'
if not tf.gfile.IsDirectory(train_dir):
tf.logging.info("Creating training directory: %s", train_dir)
tf.gfile.MakeDirs(train_dir)
g = tf.Graph()
with g.as_default():
# Build the model.
model = model_fn(model_config, phase='train')
model.build()
# Set up the learning rate
learning_rate = tf.constant(training_config.initial_learning_rate)
def _learning_rate_decay_fn(learn_rate, global_step):
return tf.train.exponential_decay(
learn_rate,
global_step,
decay_steps=training_config.decay_step,
decay_rate=training_config.decay_factor,
staircase=False)
learning_rate_decay_fn = _learning_rate_decay_fn
train_op = tf.contrib.layers.optimize_loss(
loss=model.loss,
global_step=model.global_step,
learning_rate=learning_rate,
optimizer=training_config.optimizer,
clip_gradients=training_config.clip_gradients,
learning_rate_decay_fn=learning_rate_decay_fn,
variables=model.qrd_vars)
# Set up the Saver for saving and restoring model checkpoints.
saver = tf.train.Saver(
max_to_keep=training_config.max_checkpoints_to_keep)
# setup summaries
summary_op = tf.summary.merge_all()
# create reader
model_name = os.path.split(train_dir)[1]
reader = Reader(batch_size=model_config.batch_size,
subset='trainval',
model_name=model_name,
epsilon=0.5,
feat_type='res5c',
version=FLAGS.version,
counter_sampling=do_counter_sampling)
# Run training.
training_util.train(train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver,
reader=reader,
feed_fn=model.fill_feed_dict,
summary_op=summary_op)
def train():
model_config = ModelConfig()
training_config = TrainConfig()
# training_config.initial_learning_rate = 0.01
training_config.decay_step = 100000
training_config.decay_factor = 0.1
# training_config.optimizer = lambda: tf.train.MomentumOptimizer(0.5, momentum=0.5)
# Get model
model_fn = get_model_creation_fn(FLAGS.model_type)
# Create training directory.
train_dir = FLAGS.train_dir % (FLAGS.version, FLAGS.model_type)
if FLAGS.sample_negative:
train_dir += '_sn'
if FLAGS.use_fb_data:
train_dir += '_fb'
if FLAGS.use_fb_bn:
train_dir += '_bn'
do_counter_sampling = FLAGS.version == 'v2'
if not tf.gfile.IsDirectory(train_dir):
tf.logging.info("Creating training directory: %s", train_dir)
tf.gfile.MakeDirs(train_dir)
g = tf.Graph()
with g.as_default():
# Build the model.
model_config.sample_negative = FLAGS.sample_negative
model_config.use_fb_bn = FLAGS.use_fb_bn
model = model_fn(model_config, phase='train')
model.build()
# Set up the learning rate
learning_rate = tf.constant(training_config.initial_learning_rate)
def _learning_rate_decay_fn(learn_rate, global_step):
return tf.train.exponential_decay(
learn_rate,
global_step,
decay_steps=training_config.decay_step,
decay_rate=training_config.decay_factor,
staircase=True)
# staircase=False)
learning_rate_decay_fn = _learning_rate_decay_fn
train_op = tf.contrib.layers.optimize_loss(
loss=model.loss,
global_step=model.global_step,
learning_rate=learning_rate,
# optimizer=tf.train.MomentumOptimizer(learning_rate, 0.9),
optimizer=training_config.optimizer,
clip_gradients=None,
learning_rate_decay_fn=learning_rate_decay_fn)
# Set up the Saver for saving and restoring model checkpoints.
saver = tf.train.Saver(
max_to_keep=training_config.max_checkpoints_to_keep)
# setup summaries
summary_op = tf.summary.merge_all()
# create reader
batch_size = 256 if FLAGS.sample_negative else 64
reader = Reader(batch_size=batch_size,
subset='kptrain',
sample_negative=FLAGS.sample_negative,
use_fb_data=FLAGS.use_fb_data)
# Run training.
training_util.train(train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver,
reader=reader,
feed_fn=model.fill_feed_dict,
summary_op=summary_op)
