Skip to content

Latest commit

 

History

History
27 lines (19 loc) · 1.21 KB

extract_fixed_features.md

File metadata and controls

27 lines (19 loc) · 1.21 KB

dragnn_ops.extract_fixed_features(handle, channel_id, component=None, name=None)

dragnn_ops.extract_fixed_features(handle, channel_id, component=None, name=None)

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

Given a handle to a ComputeSession and a channel index, outputs fixed features.

Fixed features returned as 3 vectors, 'indices', 'ids', and 'weights' of equal length. 'ids' specifies which rows should be looked up in the embedding matrix. 'weights' specifies a scale for each embedding vector. 'indices' is a sorted vector that assigns the same index to embedding vectors that should be summed together.

Args:

  • handle: A Tensor of type string. handle to a ComputeSession.
  • channel_id: An int. feature channel to extract features for.
  • component: An optional string. Defaults to "".
  • name: A name for the operation (optional).

Returns:

A tuple of Tensor objects (indices, ids, weights). * indices: A Tensor of type int32. row to add embeddings to. * ids: A Tensor of type int64. lookup indices into embedding matrices. * weights: A Tensor of type float32. weight for each looked up embedding.