Python tensorflow.python.training.training.get_global_step() Examples

def _create_global_step(graph):
  graph = graph or ops.get_default_graph()
  if training.get_global_step(graph) is not None:
    raise ValueError('"global_step" already exists.')
  # Create in proper graph and base name_scope.
  with graph.as_default() as g, g.name_scope(None):
    return variable_scope.get_variable(
        ops.GraphKeys.GLOBAL_STEP,
        shape=[],
        dtype=dtypes.int64,
        initializer=init_ops.zeros_initializer(),
        trainable=False,
        use_resource=True,
        collections=[ops.GraphKeys.GLOBAL_VARIABLES, ops.GraphKeys.GLOBAL_STEP]) 

def begin(self):
    self._global_step_tensor = training_util.get_global_step()
    if self._global_step_tensor is None:
      raise RuntimeError('Global step should be created.')

    self._iterations_per_loop_var = _create_or_get_iterations_per_loop() 

def _create_global_step(graph):
  graph = graph or ops.get_default_graph()
  if training.get_global_step(graph) is not None:
    raise ValueError('"global_step" already exists.')
  # Create in proper graph and base name_scope.
  with graph.as_default() as g, g.name_scope(None):
    return variable_scope.get_variable(
        ops.GraphKeys.GLOBAL_STEP,
        shape=[],
        dtype=dtypes.int64,
        initializer=init_ops.zeros_initializer(),
        trainable=False,
        use_resource=True,
        collections=[ops.GraphKeys.GLOBAL_VARIABLES, ops.GraphKeys.GLOBAL_STEP]) 

def begin(self):
    self._global_step_tensor = training_util.get_global_step()
    if self._global_step_tensor is None:
      raise RuntimeError('Global step should be created.')

    self._iterations_per_loop_var = _create_or_get_iterations_per_loop() 

def _create_global_step(graph):
  graph = graph or ops.get_default_graph()
  if training.get_global_step(graph) is not None:
    raise ValueError('"global_step" already exists.')
  # Create in proper graph and base name_scope.
  with graph.as_default() as g, g.name_scope(None):
    return variable_scope.get_variable(
        ops.GraphKeys.GLOBAL_STEP,
        shape=[],
        dtype=dtypes.int64,
        initializer=init_ops.zeros_initializer(),
        trainable=False,
        use_resource=True,
        collections=[ops.GraphKeys.GLOBAL_VARIABLES, ops.GraphKeys.GLOBAL_STEP]) 

def begin(self):
    self._global_step_tensor = training_util.get_global_step()
    if self._global_step_tensor is None:
      raise RuntimeError('Global step should be created.')

    self._iterations_per_loop_var = _create_or_get_iterations_per_loop() 

def _create_and_assert_global_step(self, graph):
    """Creates and asserts properties of the global step.
    Args:
      graph: The graph in which to create the global step tensor.
    Returns:
      The global step `Tensor`.
    """
    step = self._create_global_step(graph)
    assert step == training.get_global_step()
    assert step.dtype.is_integer
    return step 

def _create_global_step(graph):
  graph = graph or ops.get_default_graph()
  if training.get_global_step(graph) is not None:
    raise ValueError('"global_step" already exists.')
  # Create in proper graph and base name_scope.
  with graph.as_default() as g, g.name_scope(None):
    return variable_scope.get_variable(
        ops.GraphKeys.GLOBAL_STEP,
        shape=[],
        dtype=dtypes.int64,
        initializer=init_ops.zeros_initializer(),
        trainable=False,
        use_resource=True,
        collections=[ops.GraphKeys.GLOBAL_VARIABLES, ops.GraphKeys.GLOBAL_STEP]) 

def begin(self):
    self._global_step_tensor = training_util.get_global_step()
    if self._global_step_tensor is None:
      raise RuntimeError('Global step should be created.')

    self._iterations_per_loop_var = _create_or_get_iterations_per_loop() 

def _create_and_assert_global_step(self, graph):
    """Creates and asserts properties of the global step.

    Args:
      graph: The graph in which to create the global step tensor.

    Returns:
      The global step `Tensor`.
    """
    step = self._create_global_step(graph)
    assert step == training.get_global_step()
    assert step.dtype.is_integer
    return step 

def _create_or_get_iterations_per_loop():
  """Creates or gets the iterations_per_loop variable.

  In TPUEstimator, the user provided computation, the model_fn, is wrapped
  inside a tf.while_loop for peak performance. The iterations of the loop are
  specified by this variable, which adjusts its value on the CPU after each TPU
  program execution and before the next TPU execution.

  The purpose of using a variable, rather then a constant, is to allow
  TPUEstimator adapt the TPU training iterations according to the final steps
  specified by users. For example, if the user sets the iterations_per_loop as 4
  in TPUConfig and steps as 10 in TPUEstimator.train(), the iterations_per_loop
  variable will have the following value before each TPU training.

      - 1-th TPU execution: iterations_per_loop = 4
      - 2-th TPU execution: iterations_per_loop = 4
      - 3-th TPU execution: iterations_per_loop = 2

  As model_fn increases the global step once per train_op invocation, the global
  step is 10 after all TPU executions, matching the steps=10 inputs passed in by
  users.

  Returns:
    A TF non-trainable resource variable.

  Raises:
    RuntimeError: If multi iterations_per_loop variables were found.
  """
  graph = ops.get_default_graph()
  collection_name = '{}_{}'.format(_TPU_ESTIMATOR, _ITERATIONS_PER_LOOP_VAR)
  iter_vars = graph.get_collection(collection_name)
  if len(iter_vars) == 1:
    return iter_vars[0]
  elif len(iter_vars) > 1:
    raise RuntimeError('Multiple iterations_per_loop_var in collection.')

  with ops.colocate_with(training_util.get_global_step()):
    with variable_scope.variable_scope(
        _TPU_ESTIMATOR, reuse=variable_scope.AUTO_REUSE):
      return variable_scope.get_variable(
          _ITERATIONS_PER_LOOP_VAR,
          initializer=init_ops.zeros_initializer(),
          shape=[],
          dtype=dtypes.int32,
          trainable=False,
          collections=[collection_name, ops.GraphKeys.LOCAL_VARIABLES],
          use_resource=True) 

def _create_or_get_iterations_per_loop():
  """Creates or gets the iterations_per_loop variable.

  In TPUEstimator, the user provided computation, the model_fn, is wrapped
  inside a tf.while_loop for peak performance. The iterations of the loop are
  specified by this variable, which adjusts its value on the CPU after each TPU
  program execution and before the next TPU execution.

  The purpose of using a variable, rather then a constant, is to allow
  TPUEstimator adapt the TPU training iterations according to the final steps
  specified by users. For example, if the user sets the iterations_per_loop as 4
  in TPUConfig and steps as 10 in TPUEstimator.train(), the iterations_per_loop
  variable will have the following value before each TPU training.

      - 1-th TPU execution: iterations_per_loop = 4
      - 2-th TPU execution: iterations_per_loop = 4
      - 3-th TPU execution: iterations_per_loop = 2

  As model_fn increases the global step once per train_op invocation, the global
  step is 10 after all TPU executions, matching the steps=10 inputs passed in by
  users.

  Returns:
    A TF non-trainable resource variable.

  Raises:
    RuntimeError: If multi iterations_per_loop variables were found.
  """
  graph = ops.get_default_graph()
  collection_name = '{}_{}'.format(_TPU_ESTIMATOR, _ITERATIONS_PER_LOOP_VAR)
  iter_vars = graph.get_collection(collection_name)
  if len(iter_vars) == 1:
    return iter_vars[0]
  elif len(iter_vars) > 1:
    raise RuntimeError('Multiple iterations_per_loop_var in collection.')

  with ops.colocate_with(training_util.get_global_step()):
    with variable_scope.variable_scope(
        _TPU_ESTIMATOR, reuse=variable_scope.AUTO_REUSE):
      return variable_scope.get_variable(
          _ITERATIONS_PER_LOOP_VAR,
          initializer=init_ops.zeros_initializer(),
          shape=[],
          dtype=dtypes.int32,
          trainable=False,
          collections=[collection_name, ops.GraphKeys.LOCAL_VARIABLES],
          use_resource=True) 

def _create_or_get_iterations_per_loop():
  """Creates or gets the iterations_per_loop variable.

  In TPUEstimator, the user provided computation, the model_fn, is wrapped
  inside a tf.while_loop for peak performance. The iterations of the loop are
  specified by this variable, which adjusts its value on the CPU after each TPU
  program execution and before the next TPU execution.

  The purpose of using a variable, rather then a constant, is to allow
  TPUEstimator adapt the TPU training iterations according to the final steps
  specified by users. For example, if the user sets the iterations_per_loop as 4
  in TPUConfig and steps as 10 in TPUEstimator.train(), the iterations_per_loop
  variable will have the following value before each TPU training.

      - 1-th TPU execution: iterations_per_loop = 4
      - 2-th TPU execution: iterations_per_loop = 4
      - 3-th TPU execution: iterations_per_loop = 2

  As model_fn increases the global step once per train_op invocation, the global
  step is 10 after all TPU executions, matching the steps=10 inputs passed in by
  users.

  Returns:
    A TF non-trainable resource variable.

  Raises:
    RuntimeError: If multi iterations_per_loop variables were found.
  """
  graph = ops.get_default_graph()
  collection_name = '{}_{}'.format(_TPU_ESTIMATOR, _ITERATIONS_PER_LOOP_VAR)
  iter_vars = graph.get_collection(collection_name)
  if len(iter_vars) == 1:
    return iter_vars[0]
  elif len(iter_vars) > 1:
    raise RuntimeError('Multiple iterations_per_loop_var in collection.')

  with ops.colocate_with(training_util.get_global_step()):
    with variable_scope.variable_scope(
        _TPU_ESTIMATOR, reuse=variable_scope.AUTO_REUSE):
      return variable_scope.get_variable(
          _ITERATIONS_PER_LOOP_VAR,
          initializer=init_ops.zeros_initializer(),
          shape=[],
          dtype=dtypes.int32,
          trainable=False,
          collections=[collection_name, ops.GraphKeys.LOCAL_VARIABLES],
          use_resource=True) 

def _create_or_get_iterations_per_loop():
  """Creates or gets the iterations_per_loop variable.

  In TPUEstimator, the user provided computation, the model_fn, is wrapped
  inside a tf.while_loop for peak performance. The iterations of the loop are
  specified by this variable, which adjusts its value on the CPU after each TPU
  program execution and before the next TPU execution.

  The purpose of using a variable, rather then a constant, is to allow
  TPUEstimator adapt the TPU training iterations according to the final steps
  specified by users. For example, if the user sets the iterations_per_loop as 4
  in TPUConfig and steps as 10 in TPUEstimator.train(), the iterations_per_loop
  variable will have the following value before each TPU training.

      - 1-th TPU execution: iterations_per_loop = 4
      - 2-th TPU execution: iterations_per_loop = 4
      - 3-th TPU execution: iterations_per_loop = 2

  As model_fn increases the global step once per train_op invocation, the global
  step is 10 after all TPU executions, matching the steps=10 inputs passed in by
  users.

  Returns:
    A TF non-trainable resource variable.

  Raises:
    RuntimeError: If multi iterations_per_loop variables were found.
  """
  graph = ops.get_default_graph()
  collection_name = '{}_{}'.format(_TPU_ESTIMATOR, _ITERATIONS_PER_LOOP_VAR)
  iter_vars = graph.get_collection(collection_name)
  if len(iter_vars) == 1:
    return iter_vars[0]
  elif len(iter_vars) > 1:
    raise RuntimeError('Multiple iterations_per_loop_var in collection.')

  with ops.colocate_with(training_util.get_global_step()):
    with variable_scope.variable_scope(
        _TPU_ESTIMATOR, reuse=variable_scope.AUTO_REUSE):
      return variable_scope.get_variable(
          _ITERATIONS_PER_LOOP_VAR,
          initializer=init_ops.zeros_initializer(),
          shape=[],
          dtype=dtypes.int32,
          trainable=False,
          collections=[collection_name, ops.GraphKeys.LOCAL_VARIABLES],
          use_resource=True)