def monitored_estimator(FLAGS, worker_count, task_index, cluster, is_chief,
target, init_checkpoint, train_file, dev_file,
checkpoint_dir, **kargs):
estimator_fn(FLAGS, worker_count, task_index, is_chief, target,
init_checkpoint, train_file, dev_file, checkpoint_dir,
FLAGS.is_debug, **kargs)
def monitored_estimator(FLAGS, worker_count, task_index, cluster, is_chief,
target, init_checkpoint, train_file, dev_file,
checkpoint_dir, **kargs):
if kargs.get("running_type", "train") == "train":
print("==begin to train==")
estimator_fn(FLAGS, worker_count, task_index, is_chief, target,
init_checkpoint, train_file, dev_file, checkpoint_dir,
FLAGS.is_debug, **kargs)
def monitored_estimator(worker_count, task_index, cluster, is_chief, target,
init_checkpoint, train_file, dev_file, checkpoint_dir):
if worker_count >= 1 and FLAGS.opt_type == "ps":
available_worker_device = "/job:worker/task:%d" % (task_index)
with tf.device(
tf.train.replica_device_setter(
worker_device=available_worker_device, cluster=cluster)):
estimator_fn(FLAGS, worker_count, task_index, is_chief, target,
init_checkpoint, train_file, dev_file, checkpoint_dir,
FLAGS.is_debug)
else:
estimator_fn(FLAGS, worker_count, task_index, is_chief, target,
init_checkpoint, train_file, dev_file, checkpoint_dir,
FLAGS.is_debug)