Python tensorflow.python.ops.control_flow_ops._SwitchRefOrTensor() Examples

The following are 6 code examples of tensorflow.python.ops.control_flow_ops._SwitchRefOrTensor(). 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.python.ops.control_flow_ops , or try the search function .
Example #1
Source File: control_flow_ops_py_test.py    From deep_image_model with Apache License 2.0 5 votes vote down vote up 
def testRefSwitch(self):
    with self.test_session():
      v = tf.Variable(7)

      p = tf.constant(True)
      v1 = control_flow_ops._SwitchRefOrTensor(v.ref(), p)
      v2 = tf.assign(v1[1], 9)
      tf.global_variables_initializer().run()
      self.assertEqual(9, v2.eval())
Example #2
Source File: control_flow_grad.py    From lambda-packs with MIT License 4 votes vote down vote up 
def _MergeGrad(op, grad, _):
  """Gradients for a Merge op are calculated using a Switch op."""
  input_op = op.inputs[0].op
  graph = ops.get_default_graph()
  # pylint: disable=protected-access
  op_ctxt = control_flow_ops._GetOutputContext(input_op)
  grad_ctxt = graph._get_control_flow_context()
  # pylint: enable=protected-access
  if isinstance(op_ctxt, WhileContext):
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, grad_ctxt.pivot)
    # pylint: enable=protected-access
  elif isinstance(op_ctxt, CondContext):
    pred = op_ctxt.pred
    if grad_ctxt and grad_ctxt.grad_state:
      # This Merge node is part of a cond within a loop.
      # The backprop needs to have the value of this predicate for every
      # iteration. So we must have its values accumulated in the forward, and
      # use the accumulated values as the predicate for this backprop switch.
      grad_state = grad_ctxt.grad_state
      real_pred = grad_state.history_map.get(pred.name)
      if real_pred is None:
        # Remember the value of pred for every iteration.
        grad_ctxt = grad_state.grad_context
        grad_ctxt.Exit()
        history_pred = grad_state.AddForwardAccumulator(pred)
        grad_ctxt.Enter()

        # Add the stack pop op. If pred.op is in a (outer) CondContext,
        # the stack pop will be guarded with a switch.
        real_pred = grad_state.AddBackPropAccumulatedValue(history_pred, pred)
        grad_state.history_map[pred.name] = real_pred
      pred = real_pred
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, pred, name="cond_grad")
    # pylint: enable=protected-access
  else:
    num_inputs = len(op.inputs)
    cond = [math_ops.equal(op.outputs[1], i) for i in xrange(num_inputs)]
    # pylint: disable=protected-access
    return [control_flow_ops._SwitchRefOrTensor(grad, cond[i])[1]
            for i in xrange(num_inputs)]
    # pylint: enable=protected-access
Example #3
Source File: control_flow_grad.py    From auto-alt-text-lambda-api with MIT License 4 votes vote down vote up 
def _MergeGrad(op, grad, _):
  """Gradients for a Merge op are calculated using a Switch op."""
  input_op = op.inputs[0].op
  graph = ops.get_default_graph()
  # pylint: disable=protected-access
  op_ctxt = control_flow_ops._GetOutputContext(input_op)
  grad_ctxt = graph._get_control_flow_context()
  # pylint: enable=protected-access
  if isinstance(op_ctxt, WhileContext):
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, grad_ctxt.pivot)
    # pylint: enable=protected-access
  elif isinstance(op_ctxt, CondContext):
    pred = op_ctxt.pred
    if grad_ctxt and grad_ctxt.grad_state:
      # This Merge node is part of a cond within a loop.
      # The backprop needs to have the value of this predicate for every
      # iteration. So we must have its values accumulated in the forward, and
      # use the accumulated values as the predicate for this backprop switch.
      grad_state = grad_ctxt.grad_state
      real_pred = grad_state.history_map.get(pred.name)
      if real_pred is None:
        # Remember the value of pred for every iteration.
        grad_ctxt = grad_state.grad_context
        grad_ctxt.Exit()
        history_pred = grad_state.AddForwardAccumulator(pred)
        grad_ctxt.Enter()

        # Add the stack pop op. If pred.op is in a (outer) CondContext,
        # the stack pop will be guarded with a switch.
        real_pred = grad_state.AddBackPropAccumulatedValue(history_pred, pred)
        grad_state.history_map[pred.name] = real_pred
      pred = real_pred
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, pred, name="cond_grad")
    # pylint: enable=protected-access
  else:
    num_inputs = len(op.inputs)
    cond = [math_ops.equal(op.outputs[1], i) for i in xrange(num_inputs)]
    # pylint: disable=protected-access
    return [control_flow_ops._SwitchRefOrTensor(grad, cond[i])[1]
            for i in xrange(num_inputs)]
    # pylint: enable=protected-access
Example #4
Source File: control_flow_grad.py    From deep_image_model with Apache License 2.0 4 votes vote down vote up 
def _MergeGrad(op, grad, _):
  """Gradients for a Merge op are calculated using a Switch op."""
  input_op = op.inputs[0].op
  graph = ops.get_default_graph()
  # pylint: disable=protected-access
  op_ctxt = input_op._get_control_flow_context()
  grad_ctxt = graph._get_control_flow_context()
  # pylint: enable=protected-access
  if isinstance(op_ctxt, WhileContext):
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, grad_ctxt.pivot)
    # pylint: enable=protected-access
  elif isinstance(op_ctxt, CondContext):
    pred = op_ctxt.pred
    if grad_ctxt and grad_ctxt.grad_state:
      # This Merge node is part of a cond within a loop.
      # The backprop needs to have the value of this predicate for every
      # iteration. So we must have its values accumulated in the forward, and
      # use the accumulated values as the predicate for this backprop switch.
      grad_state = grad_ctxt.grad_state
      real_pred = grad_state.history_map.get(pred.name)
      if real_pred is None:
        # Remember the value of pred for every iteration.
        grad_ctxt = grad_state.grad_context
        grad_ctxt.Exit()
        history_pred = grad_state.AddForwardAccumulator(pred)
        grad_ctxt.Enter()

        # Add the stack pop op. If pred.op is in a (outer) CondContext,
        # the stack pop will be guarded with a switch.
        real_pred = grad_state.AddBackPropAccumulatedValue(history_pred, pred)
        grad_state.history_map[pred.name] = real_pred
      pred = real_pred
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, pred, name="cond_grad")
    # pylint: enable=protected-access
  else:
    num_inputs = len(op.inputs)
    cond = [math_ops.equal(op.outputs[1], i) for i in xrange(num_inputs)]
    # pylint: disable=protected-access
    return [control_flow_ops._SwitchRefOrTensor(grad, cond[i])[1]
            for i in xrange(num_inputs)]
    # pylint: enable=protected-access
Example #5
Source File: control_flow_grad.py    From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License 4 votes vote down vote up 
def _MergeGrad(op, grad, _):
  """Gradients for a Merge op are calculated using a Switch op."""
  input_op = op.inputs[0].op
  graph = ops.get_default_graph()
  # pylint: disable=protected-access
  op_ctxt = control_flow_ops._GetOutputContext(input_op)
  grad_ctxt = graph._get_control_flow_context()
  # pylint: enable=protected-access
  if isinstance(op_ctxt, WhileContext):
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, grad_ctxt.pivot)
    # pylint: enable=protected-access
  elif isinstance(op_ctxt, CondContext):
    pred = op_ctxt.pred
    if grad_ctxt and grad_ctxt.grad_state:
      # This Merge node is part of a cond within a loop.
      # The backprop needs to have the value of this predicate for every
      # iteration. So we must have its values accumulated in the forward, and
      # use the accumulated values as the predicate for this backprop switch.
      grad_state = grad_ctxt.grad_state
      real_pred = grad_state.history_map.get(pred.name)
      if real_pred is None:
        # Remember the value of pred for every iteration.
        grad_ctxt = grad_state.grad_context
        grad_ctxt.Exit()
        history_pred = grad_state.AddForwardAccumulator(pred)
        grad_ctxt.Enter()

        # Add the stack pop op. If pred.op is in a (outer) CondContext,
        # the stack pop will be guarded with a switch.
        real_pred = grad_state.AddBackpropAccumulatedValue(history_pred, pred)
        grad_state.history_map[pred.name] = real_pred
      pred = real_pred
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, pred, name="cond_grad")
    # pylint: enable=protected-access
  else:
    num_inputs = len(op.inputs)
    cond = [math_ops.equal(op.outputs[1], i) for i in xrange(num_inputs)]
    # pylint: disable=protected-access
    return [control_flow_ops._SwitchRefOrTensor(grad, cond[i])[1]
            for i in xrange(num_inputs)]
    # pylint: enable=protected-access
Example #6
Source File: control_flow_grad.py    From keras-lambda with MIT License 4 votes vote down vote up 
def _MergeGrad(op, grad, _):
  """Gradients for a Merge op are calculated using a Switch op."""
  input_op = op.inputs[0].op
  graph = ops.get_default_graph()
  # pylint: disable=protected-access
  op_ctxt = control_flow_ops._GetOutputContext(input_op)
  grad_ctxt = graph._get_control_flow_context()
  # pylint: enable=protected-access
  if isinstance(op_ctxt, WhileContext):
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, grad_ctxt.pivot)
    # pylint: enable=protected-access
  elif isinstance(op_ctxt, CondContext):
    pred = op_ctxt.pred
    if grad_ctxt and grad_ctxt.grad_state:
      # This Merge node is part of a cond within a loop.
      # The backprop needs to have the value of this predicate for every
      # iteration. So we must have its values accumulated in the forward, and
      # use the accumulated values as the predicate for this backprop switch.
      grad_state = grad_ctxt.grad_state
      real_pred = grad_state.history_map.get(pred.name)
      if real_pred is None:
        # Remember the value of pred for every iteration.
        grad_ctxt = grad_state.grad_context
        grad_ctxt.Exit()
        history_pred = grad_state.AddForwardAccumulator(pred)
        grad_ctxt.Enter()

        # Add the stack pop op. If pred.op is in a (outer) CondContext,
        # the stack pop will be guarded with a switch.
        real_pred = grad_state.AddBackPropAccumulatedValue(history_pred, pred)
        grad_state.history_map[pred.name] = real_pred
      pred = real_pred
    # pylint: disable=protected-access
    return control_flow_ops._SwitchRefOrTensor(grad, pred, name="cond_grad")
    # pylint: enable=protected-access
  else:
    num_inputs = len(op.inputs)
    cond = [math_ops.equal(op.outputs[1], i) for i in xrange(num_inputs)]
    # pylint: disable=protected-access
    return [control_flow_ops._SwitchRefOrTensor(grad, cond[i])[1]
            for i in xrange(num_inputs)]
    # pylint: enable=protected-access