tf.nn.embedding_lookup

embedding_lookup(params, ids, partition_strategy='mod', name=None, validate_indices=True, max_norm=None)

    Looks up `ids` in a list of embedding tensors.

    This function is used to perform parallel lookups on the list of

    tensors in `params`.  It is a generalization of

    @{tf.gather}, where `params` is

    interpreted as a partitioning of a large embedding tensor.  `params` may be

    a `PartitionedVariable` as returned by using `tf.get_variable()` with a

    partitioner.

    If `len(params) > 1`, each element `id` of `ids` is partitioned between

    the elements of `params` according to the `partition_strategy`.

    In all strategies, if the id space does not evenly divide the number of

    partitions, each of the first `(max_id + 1) % len(params)` partitions will

    be assigned one more id.

    If `partition_strategy` is `"mod"`, we assign each id to partition

    `p = id % len(params)`. For instance,

    13 ids are split across 5 partitions as:

    `[[0, 5, 10], [1, 6, 11], [2, 7, 12], [3, 8], [4, 9]]`

    If `partition_strategy` is `"div"`, we assign ids to partitions in a

    contiguous manner. In this case, 13 ids are split across 5 partitions as:

    `[[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10], [11, 12]]`

    The results of the lookup are concatenated into a dense

    tensor. The returned tensor has shape `shape(ids) + shape(params)[1:]`.

    Args:

      params: A single tensor representing the complete embedding tensor,

        or a list of P tensors all of same shape except for the first dimension,

        representing sharded embedding tensors.  Alternatively, a

        `PartitionedVariable`, created by partitioning along dimension 0. Each

        element must be appropriately sized for the given `partition_strategy`.

      ids: A `Tensor` with type `int32` or `int64` containing the ids to be looked

        up in `params`.

      partition_strategy: A string specifying the partitioning strategy, relevant

        if `len(params) > 1`. Currently `"div"` and `"mod"` are supported. Default

        is `"mod"`.

      name: A name for the operation (optional).

      validate_indices: DEPRECATED. If this operation is assigned to CPU, values

        in `indices` are always validated to be within range.  If assigned to GPU,

        out-of-bound indices result in safe but unspecified behavior, which may

        include raising an error.

      max_norm: If provided, embedding values are l2-normalized to the value of

        max_norm.

    Returns:

      A `Tensor` with the same type as the tensors in `params`.

    Raises:

      ValueError: If `params` is empty.

embedding_lookup(params, ids)其实就是按照ids顺序返回params中的第ids行。

比如说，ids=[1,3,2],就是返回params中第1,3,2行。返回结果为由params的1,3,2行组成的tensor.

最近在看，一起学习。

输出：