Function _TensorArrayConcatGrad

tensorflow/python/ops/tensor_array_grad.py:198–224

Gradient for TensorArrayConcat. Args: op: Forward TensorArrayConcat op. grad: Gradient `Tensor` to TensorArrayConcat. Returns: A flow `Tensor`, which can be used in control dependencies to force the write of `grad` to the gradient `TensorArray`.

(op, grad, unused_lengths_grad)

@ops.RegisterGradient("TensorArrayConcatV2")
@ops.RegisterGradient("TensorArrayConcatV3")
def _TensorArrayConcatGrad(op, grad, unused_lengths_grad):
  """Gradient for TensorArrayConcat.

  Args:
    op: Forward TensorArrayConcat op.
    grad: Gradient `Tensor` to TensorArrayConcat.

  Returns:
    A flow `Tensor`, which can be used in control dependencies to
    force the write of `grad` to the gradient `TensorArray`.
  """
  # Note: the forward flow dependency in the call to grad() is necessary for
  # the case of dynamic sized TensorArrays.  When creating the gradient
  # TensorArray, the final size of the forward array must be known.
  # For this we need to wait until it has been created by depending on
  # the input flow of the original op.
  handle = op.inputs[0]
  flow = op.inputs[1]
  lengths = op.outputs[1]
  dtype = op.get_attr("dtype")
  grad_source = _GetGradSource(grad)
  g = (tensor_array_ops.TensorArray(dtype=dtype, handle=handle, flow=flow,
                                    colocate_with_first_write_call=False)
       .grad(source=grad_source, flow=flow))
  u_g = g.split(grad, lengths=lengths)
  # handle, flow_in
  return [None, u_g.flow]


@ops.RegisterGradient("TensorArraySplit")