Class Monitor
Defined in tensorflow/contrib/distribute/python/monitor.py.
Executes training steps, recovers and checkpoints.
Note that this class is particularly preliminary, experimental, and expected to change.
__init__
__init__(
step_callable,
session=None
)
Initialize the Monitor with components for executing training steps.
Args:
step_callable: a trainingStepthat's capable of signaling when done.session: aSessioninstance that's needed for graph mode.
Raises:
ValueError: ifsessionwas provided for eager mode or not provided for graph mode.
Methods
tf.contrib.distribute.Monitor.run_steps
run_steps(num_steps=None)