ModelAverageCustomGetter
Defined in tensorflow/contrib/opt/python/training/model_average_optimizer.py
.
Custom_getter class is used to do.
For example, ma_custom_getter = ModelAverageCustomGetter(worker_device) with tf.device( tf.train.replica_device_setter( worker_device=worker_device, ps_device="/job:ps/cpu:0", cluster=cluster)), tf.variable_scope('',custom_getter=ma_custom_getter): hid_w = tf.get_variable( initializer=tf.truncated_normal( [IMAGE_PIXELS * IMAGE_PIXELS, FLAGS.hidden_units], stddev=1.0 / IMAGE_PIXELS), name="hid_w") hid_b = tf.get_variable(initializer=tf.zeros([FLAGS.hidden_units]), name="hid_b")
__init__
__init__(worker_device)
Create a new ElasticAverageCustomGetter
.
worker_device
: String. Name of the worker
job.__call__
__call__( getter, name, trainable, collections, *args, **kwargs )
Call self as a function.
© 2018 The TensorFlow Authors. All rights reserved.
Licensed under the Creative Commons Attribution License 3.0.
Code samples licensed under the Apache 2.0 License.
https://www.tensorflow.org/api_docs/python/tf/contrib/opt/ModelAverageCustomGetter