Defined in tensorflow/dragnn/core/ops/gen_dragnn_bulk_ops.py
.
Given a handle to a ComputeSession, advances until all states are final. Note
that, unlike AdvanceFromOracle, this op does mutate the master state, by advancing all of its states until they are final.
handle
: ATensor
of typestring
. handle to a ComputeSession.component
: An optionalstring
. Defaults to""
.name
: A name for the operation (optional).
A tuple of Tensor
objects (output_handle, gold_labels). *
output_handle
: A Tensor
of type string
. handle to updated
ComputeSession. * gold_labels
: A Tensor
of type int32
. [batch_size
- beam_size * max_num_steps] vector of oracle actions, where max_num_steps is the maximum number of steps in the oracle action sequences for every state in the batch of beams. Each sub-segment of length max_num_steps provides the oracle action sequence for the corresponding state in the batch of beams, padded with trailing -1s.