Python embedding_ops.embedding_lookup_sparse函数代码示例

  def testIncompatibleShapes(self):
    with self.test_session():
      x, _, _ = _EmbeddingParams(1, 10, dtype=dtypes.float32)
      sp_ids = sparse_tensor.SparseTensor(
          constant_op.constant([[0, 0], [0, 1], [1, 0]], dtypes.int64),
          constant_op.constant([0, 1, 2], dtypes.int32),
          constant_op.constant([2, 2], dtypes.int64))
      sp_weights = sparse_tensor.SparseTensor(
          constant_op.constant([[0, 0], [0, 1]], dtypes.int64),
          constant_op.constant([12.0, 5.0], dtypes.float32),
          constant_op.constant([1, 2], dtypes.int64))

      with self.assertRaises(ValueError):
        embedding_ops.embedding_lookup_sparse(
            x, sp_ids, sp_weights, combiner="mean")

  def testGradientsEmbeddingLookupSparse(self):
    vocab_size = 12
    batch_size = 4
    param_shape = [2, 3]
    sp_ids, sp_weights, _, _, _ = (self._RandomIdsAndWeights(
        batch_size, vocab_size))

    for num_shards, combiner, dtype, ignore_weights in itertools.product(
        [1, 3], ["sum", "mean", "sqrtn"], [dtypes.float32,
                                           dtypes.float64], [True, False]):
      with self.test_session():
        x, params, _ = _EmbeddingParams(
            num_shards, vocab_size, shape=param_shape, dtype=dtype)

        y = embedding_ops.embedding_lookup_sparse(
            x,
            sp_ids,
            None if ignore_weights else sp_weights,
            combiner=combiner)
        x_name = [_PName(i) for i in range(num_shards)]
        x_init_value = [params[x_n + ":0"] for x_n in x_name]
        x_shape = [i.shape for i in x_init_value]
        y_shape = [batch_size] + list(params[_PName(0) + ":0"].shape[1:])
        err = gradient_checker.compute_gradient_error(
            x, x_shape, y, y_shape, x_init_value=x_init_value)
      self.assertLess(err, 1e-5 if dtype == dtypes.float64 else 2e-3)

  def testEmbeddingLookupSparse(self):
    vocab_size = 13
    batch_size = 10
    param_shape = [2, 5]
    expected_lookup_result_shape = [None] + param_shape

    sp_ids, sp_weights, ids, weights, vals_per_batch_entry = (
        self._RandomIdsAndWeights(batch_size, vocab_size))

    grouped_ids = self._GroupByBatchEntry(ids, vals_per_batch_entry)
    grouped_weights = self._GroupByBatchEntry(weights, vals_per_batch_entry)
    grouped_ignored_weights = self._GroupByBatchEntry(
        np.ones(np.sum(vals_per_batch_entry)), vals_per_batch_entry)

    for num_shards, combiner, dtype, ignore_weights in itertools.product(
        [1, 5], ["sum", "mean", "sqrtn"],
        [dtypes.float16, dtypes.bfloat16, dtypes.float32, dtypes.float64],
        [True, False]):

      with self.cached_session():
        p, params, feed_dict = _EmbeddingParams(
            num_shards, vocab_size, shape=param_shape, dtype=dtype)
        embedding_sum = embedding_ops.embedding_lookup_sparse(
            p,
            sp_ids,
            None if ignore_weights else sp_weights,
            combiner=combiner)

        self.assertEqual(embedding_sum.get_shape().as_list(),
                         expected_lookup_result_shape)
        if dtype in (dtypes.float16, dtypes.bfloat16):
          self.assertEqual(embedding_sum.dtype, dtypes.float32)
        else:
          self.assertEqual(embedding_sum.dtype, dtype)

        tf_embedding_sum = embedding_sum.eval(feed_dict=feed_dict)

        np_embedding_sum, np_weight_sum, np_weight_sq_sum = _EmbeddingResult(
            params,
            grouped_ids,
            num_shards,
            vocab_size,
            weight_vals=grouped_ignored_weights
            if ignore_weights else grouped_weights)
        if combiner == "mean":
          np_embedding_sum /= np.reshape(np_weight_sum, (batch_size, 1, 1))
        if combiner == "sqrtn":
          np_embedding_sum /= np.reshape(
              np.sqrt(np_weight_sq_sum), (batch_size, 1, 1))

        rtol = 1e-6
        if dtype == dtypes.bfloat16:
          rtol = 1e-2
        elif dtype == dtypes.float16:
          rtol = 1e-3
        atol = rtol
        self.assertAllClose(np_embedding_sum, tf_embedding_sum, rtol, atol)

def safe_embedding_lookup_sparse(embedding_weights,
                                 sparse_ids,
                                 sparse_weights=None,
                                 combiner=None,
                                 default_id=None,
                                 name=None,
                                 partition_strategy="div",
                                 max_norm=None):
  """Lookup embedding results, accounting for invalid IDs and empty features.

  The partitioned embedding in `embedding_weights` must all be the same shape
  except for the first dimension. The first dimension is allowed to vary as the
  vocabulary size is not necessarily a multiple of `P`.  `embedding_weights`
  may be a `PartitionedVariable` as returned by using `tf.get_variable()` with a
  partitioner.

  Invalid IDs (< 0) are pruned from input IDs and weights, as well as any IDs
  with non-positive weight. For an entry with no features, the embedding vector
  for `default_id` is returned, or the 0-vector if `default_id` is not supplied.

  The ids and weights may be multi-dimensional. Embeddings are always aggregated
  along the last dimension.

  Args:
    embedding_weights:  A list of `P` float tensors or values representing
        partitioned embedding tensors.  Alternatively, a `PartitionedVariable`,
        created by partitioning along dimension 0.  The total unpartitioned
        shape should be `[e_0, e_1, ..., e_m]`, where `e_0` represents the
        vocab size and `e_1, ..., e_m` are the embedding dimensions.
    sparse_ids: `SparseTensor` of shape `[d_0, d_1, ..., d_n]` containing the
        ids. `d_0` is typically batch size.
    sparse_weights: `SparseTensor` of same shape as `sparse_ids`, containing
        float weights corresponding to `sparse_ids`, or `None` if all weights
        are be assumed to be 1.0.
    combiner: A string specifying how to combine embedding results for each
        entry. Currently "mean", "sqrtn" and "sum" are supported, with "mean"
        the default.
    default_id: The id to use for an entry with no features.
    name: A name for this operation (optional).
    partition_strategy: A string specifying the partitioning strategy.
        Currently `"div"` and `"mod"` are supported. Default is `"div"`.
    max_norm: If not None, all embeddings are l2-normalized to max_norm before
        combining.


  Returns:
    Dense tensor of shape `[d_0, d_1, ..., d_{n-1}, e_1, ..., e_m]`.

  Raises:
    ValueError: if `embedding_weights` is empty.
  """
  if combiner is None:
    logging.warn("The default value of combiner will change from \"mean\" "
                 "to \"sqrtn\" after 2016/11/01.")
    combiner = "mean"
  if embedding_weights is None:
    raise ValueError("Missing embedding_weights %s." % embedding_weights)
  if isinstance(embedding_weights, variables.PartitionedVariable):
    embedding_weights = list(embedding_weights)  # get underlying Variables.
  if not isinstance(embedding_weights, list):
    embedding_weights = [embedding_weights]
  if len(embedding_weights) < 1:
    raise ValueError("Missing embedding_weights %s." % embedding_weights)

  dtype = sparse_weights.dtype if sparse_weights is not None else None
  if isinstance(embedding_weights, variables.PartitionedVariable):
    embedding_weights = list(embedding_weights)
  embedding_weights = [
      ops.convert_to_tensor(w, dtype=dtype) for w in embedding_weights
  ]

  contrib_tensor_util.assert_same_float_dtype(embedding_weights +
                                              [sparse_weights])

  with ops.name_scope(name, "embedding_lookup",
                      embedding_weights + [sparse_ids,
                                           sparse_weights]) as scope:
    # Reshape higher-rank sparse ids and weights to linear segment ids.
    original_shape = sparse_ids.dense_shape
    original_rank_dim = sparse_ids.dense_shape.get_shape()[0]
    original_rank = (
        array_ops.size(original_shape)
        if original_rank_dim.value is None
        else original_rank_dim.value)
    sparse_ids = sparse_ops.sparse_reshape(sparse_ids, [
        math_ops.reduce_prod(
            array_ops.slice(original_shape, [0], [original_rank - 1])),
        array_ops.gather(original_shape, original_rank - 1)])
    if sparse_weights is not None:
      sparse_weights = sparse_tensor.SparseTensor(
          sparse_ids.indices,
          sparse_weights.values, sparse_ids.dense_shape)

    # Prune invalid ids and weights.
    sparse_ids, sparse_weights = _prune_invalid_ids(sparse_ids, sparse_weights)

    # Fill in dummy values for empty features, if necessary.
    sparse_ids, is_row_empty = sparse_ops.sparse_fill_empty_rows(sparse_ids,
                                                                 default_id or
                                                                 0)
#.........这里部分代码省略.........

def safe_embedding_lookup_sparse(
    embedding_weights, sparse_ids, sparse_weights=None, combiner="mean",
    default_id=None, name=None, partition_strategy="div"):
  """Lookup embedding results, accounting for invalid IDs and empty features.

  The partitioned embedding in `embedding_weights` must all be the same shape
  except for the first dimension. The first dimension is allowed to vary as the
  vocabulary size is not necessarily a multiple of `P`.

  Invalid IDs (< 0) are pruned from input IDs and weights, as well as any IDs
  with non-positive weight. For an entry with no features, the embedding vector
  for `default_id` is returned, or the 0-vector if `default_id` is not supplied.

  Args:
    embedding_weights:  A list of `P` float tensors or values representing
        partitioned embedding tensors.
    sparse_ids: `SparseTensor` of shape `[batch_size, ?]` containing the ids.
    sparse_weights: `SparseTensor` of same shape as `sparse_ids`, containing
        float weights corresponding to `sparse_ids`, or `None` if all weights
        are be assumed to be 1.0.
    combiner: A string specifying how to combine embedding results for each
        entry. Currently "mean", "sqrtn" and "sum" are supported, with "mean"
        the default.
    default_id: The id to use for an entry with no features.
    name: A name for this operation (optional).
    partition_strategy: A string specifying the partitioning strategy.
        Currently `"div"` and `"mod"` are supported. Default is `"div"`.


  Returns:
    Dense tensor of shape `[batch_size, embed_dim]`.

  Raises:
    ValueError: if `embedding_weights` is empty.
  """
  if embedding_weights is None or len(embedding_weights) < 1:
    raise ValueError("Missing embedding_weights %s." % embedding_weights)

  dtype = sparse_weights.dtype if sparse_weights else None
  embedding_weights = [
      ops.convert_to_tensor(w, dtype=dtype) for w in embedding_weights]

  contrib_tensor_util.assert_same_float_dtype(
      embedding_weights + [sparse_weights])

  with ops.op_scope(
      embedding_weights + [sparse_ids, sparse_weights], name,
      "embedding_lookup") as scope:
    # Prune invalid ids and weights.
    sparse_ids, sparse_weights = _prune_invalid_ids(sparse_ids, sparse_weights)

    # Fill in dummy values for empty features, if necessary.
    sparse_ids, is_row_empty = sparse_ops.sparse_fill_empty_rows(
        sparse_ids, default_id or 0)
    if sparse_weights:
      sparse_weights, _ = sparse_ops.sparse_fill_empty_rows(
          sparse_weights, 1.0)

    result = tf_embedding_ops.embedding_lookup_sparse(
        embedding_weights, sparse_ids, sparse_weights, combiner=combiner,
        partition_strategy=partition_strategy,
        name=None if default_id is None else scope)

    if default_id is None:
      # Broadcast is_row_empty to the same shape as embedding_lookup_result,
      # for use in Select.
      is_row_empty = array_ops.tile(
          array_ops.reshape(is_row_empty, [-1, 1]),
          array_ops.pack([1, array_ops.shape(result)[1]]))

      result = math_ops.select(
          is_row_empty, array_ops.zeros_like(result), result, name=scope)

    return result