tf.compat.v1.train.MonitoredSession

Session-like object that handles initialization, recovery and hooks.

tf.compat.v1.train.MonitoredSession(
    session_creator=None, hooks=None, stop_grace_period_secs=120
)

Example usage:

saver_hook = CheckpointSaverHook(...)
summary_hook = SummarySaverHook(...)
with MonitoredSession(session_creator=ChiefSessionCreator(...),
                      hooks=[saver_hook, summary_hook]) as sess:
  while not sess.should_stop():
    sess.run(train_op)

Initialization: At creation time the monitored session does following things in given order:

• calls hook.begin() for each given hook
• finalizes the graph via scaffold.finalize()
• create session
• initializes the model via initialization ops provided by Scaffold
• restores variables if a checkpoint exists
• launches queue runners
• calls hook.after_create_session()

Run: When run() is called, the monitored session does following things:

• calls hook.before_run()
• calls TensorFlow session.run() with merged fetches and feed_dict
• calls hook.after_run()
• returns result of session.run() asked by user
• if AbortedError or UnavailableError occurs, it recovers or reinitializes the session before executing the run() call again

Exit: At the close(), the monitored session does following things in order:

• calls hook.end()
• closes the queue runners and the session
• suppresses OutOfRange error which indicates that all inputs have been processed if the monitored_session is used as a context

How to set tf.compat.v1.Session arguments:

• In most cases you can set session arguments as follows:

MonitoredSession(
  session_creator=ChiefSessionCreator(master=..., config=...))

• In distributed setting for a non-chief worker, you can use following:

MonitoredSession(
  session_creator=WorkerSessionCreator(master=..., config=...))

See MonitoredTrainingSession for an example usage based on chief or worker.

Note: This is not a tf.compat.v1.Session. For example, it cannot do following:

• it cannot be set as default session.
• it cannot be sent to saver.save.
• it cannot be sent to tf.train.start_queue_runners.

Child Classes

class StepContext

Methods

close

View source

close()

run

View source

run(
    fetches, feed_dict=None, options=None, run_metadata=None
)

Run ops in the monitored session.

This method is completely compatible with the tf.Session.run() method.

run_step_fn

View source

run_step_fn(
    step_fn
)

Run ops using a step function.

with tf.Graph().as_default():
c = tf.compat.v1.placeholder(dtypes.float32)
v = tf.add(c, 4.0)
w = tf.add(c, 0.5)
def step_fn(step_context):
a = step_context.session.run(fetches=v, feed_dict={c: 0.5})
if a <= 4.5:
step_context.request_stop()
return step_context.run_with_hooks(fetches=w,
feed_dict={c: 0.1})

with tf.MonitoredSession() as session:
while not session.should_stop():
a = session.run_step_fn(step_fn)

should_stop

View source

should_stop()

__enter__

View source

__enter__()

__exit__

View source

__exit__(
    exception_type, exception_value, traceback
)