Python tensorflow.enable_resource_variables() Examples

The following are 2 code examples of tensorflow.enable_resource_variables(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module tensorflow , or try the search function .
Example #1
Source File: model_runner.py    From multilabel-image-classification-tensorflow with MIT License 6 votes vote down vote up 
def __init__(self, ncf_dataset, params, num_train_steps, num_eval_steps,
               use_while_loop):
    self._num_train_steps = num_train_steps
    self._num_eval_steps = num_eval_steps
    self._use_while_loop = use_while_loop
    with tf.Graph().as_default() as self._graph:
      if params["use_xla_for_gpu"]:
        # The XLA functions we use require resource variables.
        tf.enable_resource_variables()
      self._ncf_dataset = ncf_dataset
      self._global_step = tf.train.create_global_step()
      self._train_model_properties = self._build_model(params, num_train_steps,
                                                       is_training=True)
      self._eval_model_properties = self._build_model(params, num_eval_steps,
                                                      is_training=False)

      initializer = tf.global_variables_initializer()
    self._graph.finalize()
    self._session = tf.Session(graph=self._graph)
    self._session.run(initializer)
Example #2
Source File: classifier_mnist_tpu_estimator.py    From kfac with Apache License 2.0 5 votes vote down vote up 
def main(argv):
  del argv  # Unused.

  # If using update_damping_immediately resource variables must be enabled.
  # (Although they probably will be by default on TPUs.)
  if FLAGS.update_damping_immediately:
    tf.enable_resource_variables()

  tf.set_random_seed(FLAGS.seed)
  # Invert using cholesky decomposition + triangular solve.  This is the only
  # code path for matrix inversion supported on TPU right now.
  kfac.utils.set_global_constants(posdef_inv_method='cholesky')
  kfac.fisher_factors.set_global_constants(
      eigenvalue_decomposition_threshold=10000)

  if not FLAGS.use_sua_approx:
    if FLAGS.use_custom_patches_op:
      kfac.fisher_factors.set_global_constants(
          use_patches_second_moment_op=True
          )
    else:
      # Temporary measure to save memory with giant batches:
      kfac.fisher_factors.set_global_constants(
          sub_sample_inputs=True,
          inputs_to_extract_patches_factor=0.1)

  config = make_tpu_run_config(
      FLAGS.master, FLAGS.seed, FLAGS.model_dir, FLAGS.iterations_per_loop,
      FLAGS.save_checkpoints_steps)

  estimator = contrib_tpu.TPUEstimator(
      use_tpu=True,
      model_fn=_model_fn,
      config=config,
      train_batch_size=FLAGS.batch_size,
      eval_batch_size=1024)

  estimator.train(
      input_fn=mnist_input_fn,
      max_steps=FLAGS.train_steps,
      hooks=[])