tf.edit_distance

TensorFlow 1 version

View source on GitHub

Computes the Levenshtein distance between sequences.

tf.edit_distance(
    hypothesis, truth, normalize=True, name='edit_distance'
)

This operation takes variable-length sequences (hypothesis and truth), each provided as a SparseTensor, and computes the Levenshtein distance. You can normalize the edit distance by length of truth by setting normalize to true.

For example:

Given the following input,

• hypothesis is a tf.SparseTensor of shape [2, 1, 1]
• truth is a tf.SparseTensor of shape [2, 2, 2]

hypothesis = tf.SparseTensor(
  [[0, 0, 0],
   [1, 0, 0]],
  ["a", "b"],
  (2, 1, 1))
truth = tf.SparseTensor(
  [[0, 1, 0],
   [1, 0, 0],
   [1, 0, 1],
   [1, 1, 0]],
   ["a", "b", "c", "a"],
   (2, 2, 2))
tf.edit_distance(hypothesis, truth, normalize=True)
<tf.Tensor: shape=(2, 2), dtype=float32, numpy=
array([[inf, 1. ],
       [0.5, 1. ]], dtype=float32)>

The operation returns a dense Tensor of shape [2, 2] with edit distances normalized by truth lengths.

For the following inputs,

# 'hypothesis' is a tensor of shape `[2, 1]` with variable-length values:
#   (0,0) = ["a"]
#   (1,0) = ["b"]
hypothesis = tf.sparse.SparseTensor(
    [[0, 0, 0],
     [1, 0, 0]],
    ["a", "b"],
    (2, 1, 1))

# 'truth' is a tensor of shape `[2, 2]` with variable-length values:
#   (0,0) = []
#   (0,1) = ["a"]
#   (1,0) = ["b", "c"]
#   (1,1) = ["a"]
truth = tf.sparse.SparseTensor(
    [[0, 1, 0],
     [1, 0, 0],
     [1, 0, 1],
     [1, 1, 0]],
    ["a", "b", "c", "a"],
    (2, 2, 2))

normalize = True

# The output would be a dense Tensor of shape `(2,)`, with edit distances
normalized by 'truth' lengths.
# output => array([0., 0.5], dtype=float32)