def run(hparams, run_dir):
"""Run train/eval/test."""
train_dir = os.path.join(run_dir, 'train')
if FLAGS.mode == 'eval':
eval_dir = os.path.join(run_dir, 'eval')
if FLAGS.eval_dir:
eval_dir = os.path.join(eval_dir, FLAGS.eval_dir)
train_util.evaluate(train_dir=train_dir,
eval_dir=eval_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams)
elif FLAGS.mode == 'test':
checkpoint_path = (os.path.expanduser(FLAGS.checkpoint_path)
if FLAGS.checkpoint_path else
tf.train.latest_checkpoint(train_dir))
tf.logging.info('Testing with checkpoint: %s', checkpoint_path)
test_dir = os.path.join(run_dir, 'test')
train_util.test(checkpoint_path=checkpoint_path,
test_dir=test_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams)
elif FLAGS.mode == 'train':
train_util.train(train_dir=train_dir,
examples_path=FLAGS.examples_path,
hparams=hparams,
checkpoints_to_keep=FLAGS.checkpoints_to_keep,
num_steps=FLAGS.num_steps)
else:
raise ValueError('Invalid mode: {}'.format(FLAGS.mode))
def run(hparams, run_dir):
"""Run train/eval/test."""
train_dir = os.path.join(run_dir, 'train')
if FLAGS.mode == 'eval':
eval_dir = os.path.join(run_dir, 'eval')
if FLAGS.eval_dir:
eval_dir = os.path.join(eval_dir, FLAGS.eval_dir)
train_util.evaluate(
train_dir=train_dir,
eval_dir=eval_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams)
elif FLAGS.mode == 'test':
checkpoint_path = (os.path.expanduser(FLAGS.checkpoint_path)
if FLAGS.checkpoint_path else
tf.train.latest_checkpoint(train_dir))
tf.logging.info('Testing with checkpoint: %s', checkpoint_path)
test_dir = os.path.join(run_dir, 'test')
train_util.test(
checkpoint_path=checkpoint_path,
test_dir=test_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams)
elif FLAGS.mode == 'train':
train_util.train(
train_dir=train_dir,
examples_path=FLAGS.examples_path,
hparams=hparams,
checkpoints_to_keep=FLAGS.checkpoints_to_keep,
num_steps=FLAGS.num_steps)
else:
raise ValueError('Invalid mode: {}'.format(FLAGS.mode))
def run(config_map, data_fn, additional_trial_info):
"""Run training or evaluation."""
tf.logging.set_verbosity(FLAGS.log)
config = config_map[FLAGS.config]
model_dir = os.path.expanduser(FLAGS.model_dir)
hparams = config.hparams
# Command line flags override any of the preceding hyperparameter values.
hparams.parse(FLAGS.hparams)
if FLAGS.mode == 'train':
train_util.train(model_fn=config.model_fn,
data_fn=data_fn,
additional_trial_info=additional_trial_info,
master=FLAGS.master,
model_dir=model_dir,
use_tpu=FLAGS.use_tpu,
preprocess_examples=FLAGS.preprocess_examples,
hparams=hparams,
keep_checkpoint_max=FLAGS.keep_checkpoint_max,
num_steps=FLAGS.num_steps)
elif FLAGS.mode == 'eval':
train_util.evaluate(model_fn=config.model_fn,
data_fn=data_fn,
additional_trial_info=additional_trial_info,
master=FLAGS.master,
model_dir=model_dir,
name=FLAGS.eval_name,
preprocess_examples=FLAGS.preprocess_examples,
hparams=hparams,
num_steps=FLAGS.eval_num_steps)
else:
raise ValueError('Unknown/unsupported mode: %s' % FLAGS.mode)
def run(config_map, semisupervised_examples_map=None):
"""Run training or evaluation."""
tf.logging.set_verbosity(FLAGS.log)
# Validate data path flags.
if not FLAGS.examples_path and not FLAGS.semisupervised_examples_config:
raise ValueError('You must set flags for either `examples_path` or '
'`semisupervised_examples_config`.')
if FLAGS.examples_path and FLAGS.semisupervised_examples_config:
raise ValueError('You must only set one of either `examples_path` or '
'`semisupervised_examples_config`.')
if not FLAGS.examples_path and FLAGS.mode == 'eval':
raise ValueError('You must set flags for `examples_path` if in eval mode.')
semisupervised_configs = (
semisupervised_examples_map[FLAGS.semisupervised_examples_config]
if FLAGS.semisupervised_examples_config and semisupervised_examples_map
else None)
config = config_map[FLAGS.config]
model_dir = os.path.expanduser(FLAGS.model_dir)
hparams = config.hparams
# Command line flags override any of the preceding hyperparameter values.
hparams.parse(FLAGS.hparams)
if FLAGS.mode == 'train':
train_util.train(
model_fn=config.model_fn,
master=FLAGS.master,
model_dir=model_dir,
use_tpu=FLAGS.use_tpu,
examples_path=FLAGS.examples_path,
preprocess_examples=FLAGS.preprocess_examples,
hparams=hparams,
keep_checkpoint_max=FLAGS.keep_checkpoint_max,
num_steps=FLAGS.num_steps,
semisupervised_configs=semisupervised_configs)
elif FLAGS.mode == 'eval':
train_util.evaluate(
model_fn=config.model_fn,
master=FLAGS.master,
model_dir=model_dir,
name=FLAGS.eval_name,
examples_path=FLAGS.examples_path,
preprocess_examples=FLAGS.preprocess_examples,
hparams=hparams,
num_steps=FLAGS.eval_num_steps)
else:
raise ValueError('Unknown/unsupported mode: %s' % FLAGS.mode)
def run(hparams, run_dir):
"""Run train/eval/test."""
train_dir = os.path.join(run_dir, 'train')
if FLAGS.mode == 'eval':
eval_dir = os.path.join(run_dir, 'eval')
if FLAGS.eval_dir:
eval_dir = os.path.join(eval_dir, FLAGS.eval_dir)
train_util.evaluate(
train_dir=train_dir,
eval_dir=eval_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams,
master=FLAGS.master)
elif FLAGS.mode == 'test':
checkpoint_path = tf.train.latest_checkpoint(train_dir)
if FLAGS.checkpoint_path:
checkpoint_path = os.path.expanduser(FLAGS.checkpoint_path)
tf.logging.info('Testing with checkpoint: %s', checkpoint_path)
test_dir = os.path.join(run_dir, 'test')
train_util.test(
checkpoint_path=checkpoint_path,
test_dir=test_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams,
master=FLAGS.master)
elif FLAGS.mode == 'train':
train_util.train(
train_dir=train_dir,
examples_path=FLAGS.examples_path,
hparams=hparams,
checkpoints_to_keep=FLAGS.checkpoints_to_keep,
num_steps=FLAGS.num_steps,
master=FLAGS.master,
task=FLAGS.ps_task,
num_ps_tasks=FLAGS.num_ps_tasks)
def run(hparams, run_dir):
"""Run train/eval/test."""
train_dir = os.path.join(run_dir, 'train')
if FLAGS.mode == 'eval':
eval_dir = os.path.join(run_dir, 'eval')
if FLAGS.eval_dir:
eval_dir = os.path.join(eval_dir, FLAGS.eval_dir)
train_util.evaluate(train_dir=train_dir,
eval_dir=eval_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams,
master=FLAGS.master)
elif FLAGS.mode == 'test':
checkpoint_path = tf.train.latest_checkpoint(train_dir)
if FLAGS.checkpoint_path:
checkpoint_path = os.path.expanduser(FLAGS.checkpoint_path)
tf.logging.info('Testing with checkpoint: %s', checkpoint_path)
test_dir = os.path.join(run_dir, 'test')
train_util.test(checkpoint_path=checkpoint_path,
test_dir=test_dir,
examples_path=FLAGS.examples_path,
num_batches=FLAGS.eval_num_batches,
hparams=hparams,
master=FLAGS.master)
elif FLAGS.mode == 'train':
train_util.train(train_dir=train_dir,
examples_path=FLAGS.examples_path,
hparams=hparams,
checkpoints_to_keep=FLAGS.checkpoints_to_keep,
num_steps=FLAGS.num_steps,
master=FLAGS.master,
task=FLAGS.ps_task,
num_ps_tasks=FLAGS.num_ps_tasks)