Skip to content

Latest commit

 

History

History
33 lines (25 loc) · 1.67 KB

bulk_fixed_features.md

File metadata and controls

33 lines (25 loc) · 1.67 KB
Raw

dragnn_ops.bulk_fixed_features(handle, num_channels, component=None, name=None)

dragnn_ops.bulk_fixed_features(handle, num_channels, component=None, name=None)

Defined in tensorflow/dragnn/core/ops/gen_dragnn_bulk_ops.py.

Given a handle to a ComputeSession and a component name, outputs fixed features

for the entire oracle path of the component. Unlike ExtractFixedFeatures, this op mutates the master state, advancing all of its states until they are final. For every channel, indices[channel], ids[channel], and weights[channel] have the same length, ie. the number of predicates, ordered by batch, beam, step.

Args:

  • handle: A Tensor of type string. handle to a ComputeSession.
  • num_channels: An int that is >= 1.
  • component: An optional string. Defaults to "".
  • name: A name for the operation (optional).

Returns:

A tuple of Tensor objects (output_handle, indices, ids, weights, num_steps). * output_handle: A Tensor of type string. handle to the same ComputeSession after advancement. indices (num_channels vectors of int32): if indices[i] = j, then embedding_sum[j] += embedding_matrix[ids[i]] * weights[i]. ids (num_channels vectors of int64): ids to lookup in embedding matrices. weights (num_channels vectors of float): weight for each embedding. num_steps (int32 scalar): batch was unrolled for these many steps. * indices: A list of num_channels Tensor objects of type int32. * ids: A list of num_channels Tensor objects of type int64. * weights: A list of num_channels Tensor objects of type float32. * num_steps: A Tensor of type int32.