tf.contrib.layers.scattered_embedding_column

View source on GitHub

Creates an embedding column of a sparse feature using parameter hashing.

This is a useful shorthand when you have a sparse feature you want to use an embedding for, but also want to hash the embedding's values in each dimension to a variable based on a different hash.

Specifically, the i-th embedding component of a value v is found by retrieving an embedding weight whose index is a fingerprint of the pair (v,i).

An embedding column with sparse_column_with_hash_bucket such as

embedding_column(
  sparse_column_with_hash_bucket(column_name, bucket_size),
  dimension)

could be replaced by

scattered_embedding_column(
  column_name,
  size=bucket_size * dimension,
  dimension=dimension,
  hash_key=tf.contrib.layers.SPARSE_FEATURE_CROSS_DEFAULT_HASH_KEY)

for the same number of embedding parameters. This should hopefully reduce the impact of collisions, but adds the cost of slowing down training.

column_name A string defining sparse column name.
size An integer specifying the number of parameters in the embedding layer.
dimension An integer specifying dimension of the embedding.
hash_key Specify the hash_key that will be used by the FingerprintCat64 function to combine the crosses fingerprints on SparseFeatureCrossOp.
combiner A string specifying how to reduce if there are multiple entries in a single row. Currently "mean", "sqrtn" and "sum" are supported, with "mean" the default. "sqrtn" often achieves good accuracy, in particular with bag-of-words columns. Each of this can be thought as example level normalizations on the column:

  • "sum": do not normalize features in the column
  • "mean": do l1 normalization on features in the column
  • "sqrtn": do l2 normalization on features in the column For more information: tf.embedding_lookup_sparse.
initializer A variable initializer function to be used in embedding variable initialization. If not specified, defaults to tf.compat.v1.truncated_normal_initializer with mean 0 and standard deviation 0.1.

A _ScatteredEmbeddingColumn.

ValueError if dimension or size is not a positive integer; or if combiner is not supported.