tf.io.SparseFeature | TensorFlow v2.16.1 (original) (raw)

Configuration for parsing a sparse input feature from an Example.

View aliases

Compat aliases for migration

tf.compat.v1.SparseFeature, tf.compat.v1.io.SparseFeature

tf.io.SparseFeature(
    index_key, value_key, dtype, size, already_sorted=False
)

Note, preferably use VarLenFeature (possibly in combination with aSequenceExample) in order to parse out SparseTensors instead ofSparseFeature due to its simplicity.

Closely mimicking the SparseTensor that will be obtained by parsing anExample with a SparseFeature config, a SparseFeature contains a

value_key: The name of key for a Feature in the Example whose parsedTensor will be the resulting SparseTensor.values.
index_key: A list of names - one for each dimension in the resultingSparseTensor whose indices[i][dim] indicating the position of the i-th value in the dim dimension will be equal to the i-th value in the Feature with key named index_key[dim] in the Example.
size: A list of ints for the resulting SparseTensor.dense_shape.

For example, we can represent the following 2D SparseTensor

SparseTensor(indices=[[3, 1], [20, 0]],
             values=[0.5, -1.0]
             dense_shape=[100, 3])

with an Example input proto

features {
  feature { key: "val" value { float_list { value: [ 0.5, -1.0 ] } } }
  feature { key: "ix0" value { int64_list { value: [ 3, 20 ] } } }
  feature { key: "ix1" value { int64_list { value: [ 1, 0 ] } } }
}

and SparseFeature config with 2 index_keys

SparseFeature(index_key=["ix0", "ix1"],
              value_key="val",
              dtype=tf.float32,
              size=[100, 3])

Fields
index_key	A single string name or a list of string names of index features. For each key the underlying feature's type must be int64 and its length must always match that of the value_key feature. To represent SparseTensors with a dense_shape of rank higher than 1 a list of length rank should be used.
value_key	Name of value feature. The underlying feature's type must be dtype and its length must always match that of all the index_keys' features.
dtype	Data type of the value_key feature.
size	A Python int or list thereof specifying the dense shape. Should be a list if and only if index_key is a list. In that case the list must be equal to the length of index_key. Each for each entry i all values in the index_key[i] feature must be in [0, size[i]).
already_sorted	A Python boolean to specify whether the values invalue_key are already sorted by their index position. If so skip sorting. False by default (optional).

Attributes
index_key	A namedtuple alias for field number 0
value_key	A namedtuple alias for field number 1
dtype	A namedtuple alias for field number 2
size	A namedtuple alias for field number 3
already_sorted	A namedtuple alias for field number 4