pyxDamerauLevenshtein
LICENSE
This software is licensed under the BSD 3Clause License. Please refer to the separate LICENSE file for the exact text of the license. You are obligated to give attribution if you use this code.
ABOUT
pyxDamerauLevenshtein implements the DamerauLevenshtein (DL) edit distance algorithm for Python in Cython for high performance. Courtesy Wikipedia:
In information theory and computer science, the DamerauLevenshtein distance (named after Frederick J. Damerau and Vladimir I. Levenshtein) is a "distance" (string metric) between two strings, i.e., finite sequence of symbols, given by counting the minimum number of operations needed to transform one string into the other, where an operation is defined as an insertion, deletion, or substitution of a single character, or a transposition of two adjacent characters.
This implementation is based on Michael Homer's pure Python implementation, which implements the optimal string alignment distance algorithm. It runs in O(N*M)
time using O(M)
space. It supports unicode characters.
REQUIREMENTS
This code requires Python 3.7+ and a C compiler such as GCC. Although the code was written in Cython, Cython is not required for installation.
INSTALL
pyxDamerauLevenshtein is available on PyPI at https://pypi.org/project/pyxDamerauLevenshtein/.
Install using pip:
pip install pyxDamerauLevenshtein
Install from source:
python setup.py install
or
pip install .
USING THIS CODE
The following methods are available:

Edit distance (
damerau_levenshtein_distance
) Compute the raw distance between two strings (i.e., the minumum number of operations necessary to transform one string into the other).
 Additionally, the distance between lists and tuples can also be computed.

Normalized edit distance (
normalized_damerau_levenshtein_distance
) Compute the ratio of the edit distance to the length of
max(string1, string2)
. 0.0 means that the sequences are identical, while 1.0 means that they have nothing in common. Note that this definition is the exact opposite ofdifflib.SequenceMatcher.ratio()
.
 Compute the ratio of the edit distance to the length of

Edit distance against a sequence of sequences (
damerau_levenshtein_distance_seqs
) Compute the raw distances between a sequence and each sequence within another sequence (e.g.,
list
,tuple
).
 Compute the raw distances between a sequence and each sequence within another sequence (e.g.,

Normalized edit distance against a sequence of sequences (
normalized_damerau_levenshtein_distance_seqs
) Compute the normalized distances between a sequence and each sequence within another sequence (e.g.,
list
,tuple
).
 Compute the normalized distances between a sequence and each sequence within another sequence (e.g.,
Basic use:
from pyxdameraulevenshtein import damerau_levenshtein_distance, normalized_damerau_levenshtein_distance
damerau_levenshtein_distance('smtih', 'smith') # expected result: 1
normalized_damerau_levenshtein_distance('smtih', 'smith') # expected result: 0.2
damerau_levenshtein_distance([1, 2, 3, 4, 5, 6], [7, 8, 9, 7, 10, 11, 4]) # expected result: 7
from pyxdameraulevenshtein import damerau_levenshtein_distance_seqs, normalized_damerau_levenshtein_distance_seqs
array = ['test1', 'test12', 'test123']
damerau_levenshtein_distance_seqs('test', array) # expected result: [1, 2, 3]
normalized_damerau_levenshtein_distance_seqs('test', array) # expected result: [0.2, 0.33333334, 0.42857143]
DIFFERENCES
Other Python DL implementations:
pyxDamerauLevenshtein differs from other Python implementations in that it is both fast via Cython and supports unicode. Michael Homer's implementation is fast for Python, but it is two orders of magnitude slower than this Cython implementation. jellyfish provides C implementations for a variety of string comparison metrics and is sometimes faster than pyxDamerauLevenshtein.
Python's builtin difflib.SequenceMatcher.ratio()
performs about an order of magnitude faster than Michael Homer's implementation but is still one order of magnitude slower than this DL implementation. difflib, however, uses a different algorithm (difflib uses the Ratcliff/Obershelp algorithm).
Performance differences (on Intel i72600 running at 3.4Ghz):
>>> import timeit
>>> #this implementation:
... timeit.timeit("damerau_levenshtein_distance('e0zdvfb840174ut74j2v7gabx1 5bs', 'qpk5vei 4tzo0bglx8rl7e 2h4uei7')", 'from pyxdameraulevenshtein import damerau_levenshtein_distance', number=500000)
7.417556047439575
>>> #Michael Homer's native Python implementation:
... timeit.timeit("dameraulevenshtein('e0zdvfb840174ut74j2v7gabx1 5bs', 'qpk5vei 4tzo0bglx8rl7e 2h4uei7')", 'from dameraulevenshtein import dameraulevenshtein', number=500000)
667.0276439189911
>>> #difflib
... timeit.timeit("difflib.SequenceMatcher(None, 'e0zdvfb840174ut74j2v7gabx1 5bs', 'qpk5vei 4tzo0bglx8rl7e 2h4uei7').ratio()", 'import difflib', number=500000)
135.41051697731018