tf.contrib.lookup.IdTableWithHashBuckets

Class IdTableWithHashBuckets

Inherits From: LookupInterface

Defined in tensorflow/python/ops/lookup_ops.py.

String to Id table wrapper that assigns out-of-vocabulary keys to buckets.

For example, if an instance of IdTableWithHashBuckets is initialized with a string-to-id table that maps:

  • emerson -> 0
  • lake -> 1
  • palmer -> 2

The IdTableWithHashBuckets object will performs the following mapping:

  • emerson -> 0
  • lake -> 1
  • palmer -> 2
  • <other term> -> bucket_id, where bucket_id will be between 3 and 3 + num_oov_buckets - 1, calculated by: hash(<term>) % num_oov_buckets + vocab_size

If input_tensor is ["emerson", "lake", "palmer", "king", "crimson"], the lookup result is [0, 1, 2, 4, 7].

If table is None, only out-of-vocabulary buckets are used.

Example usage:

num_oov_buckets = 3
input_tensor = tf.constant(["emerson", "lake", "palmer", "king", "crimnson"])
table = tf.IdTableWithHashBuckets(
    tf.HashTable(tf.TextFileIdTableInitializer(filename), default_value),
    num_oov_buckets)
out = table.lookup(input_tensor).
table.init.run()
print(out.eval())

The hash function used for generating out-of-vocabulary buckets ID is handled by hasher_spec.

__init__

__init__(
    table,
    num_oov_buckets,
    hasher_spec=tf.contrib.lookup.FastHashSpec,
    name=None,
    key_dtype=None
)

Construct a IdTableWithHashBuckets object.

Args:

  • table: Table that maps tf.string or tf.int64 keys to tf.int64 ids.
  • num_oov_buckets: Number of buckets to use for out-of-vocabulary keys.
  • hasher_spec: A HasherSpec to specify the hash function to use for assignation of out-of-vocabulary buckets (optional).
  • name: A name for the operation (optional).
  • key_dtype: Data type of keys passed to lookup. Defaults to table.key_dtype if table is specified, otherwise tf.string. Must be string or integer, and must be castable to table.key_dtype.

Raises:

  • ValueError: when table in None and num_oov_buckets is not positive.
  • TypeError: when hasher_spec is invalid.

Properties

init

The table initialization op.

key_dtype

The table key dtype.

name

The name of the table.

table_ref

Returns the table_ref of the underlying table, if one exists.

Only use the table_ref directly if you know what you are doing. The table_ref does not have the "hash bucket" functionality, as that is provided by this class.

One possible use of the table_ref is subtokenization, i.e. ops which dynamically decompose tokens into subtokens based on the contents of the table_ref.

Returns:

the underlying table_ref, or None if there is no underlying table

value_dtype

The table value dtype.

Methods

lookup

lookup(
    keys,
    name=None
)

Looks up keys in the table, outputs the corresponding values.

It assigns out-of-vocabulary keys to buckets based in their hashes.

Args:

  • keys: Keys to look up. May be either a SparseTensor or dense Tensor.
  • name: Optional name for the op.

Returns:

A SparseTensor if keys are sparse, otherwise a dense Tensor.

Raises:

  • TypeError: when keys doesn't match the table key data type.

size

size(name=None)

Compute the number of elements in this table.