Eddie

Fast and well-tested implementations of edit distance/string similarity metrics:

• Levenshtein,
• Damerau-Levenshtein,
• Hamming,
• Jaro,
• Jaro-Winkler.

Usage

Cargo.toml:

[dependencies]
eddie = "0.2"

Levenshtein:

use eddie::Levenshtein;
let lev = Levenshtein::new();
let dist = lev.distance("martha", "marhta");
assert_eq!(dist, 2);

Damerau-Levenshtein:

use eddie::DamerauLevenshtein;
let damlev = DamerauLevenshtein::new();
let dist = damlev.distance("martha", "marhta");
assert_eq!(dist, 1);

Hamming:

use eddie::Hamming;
let hamming = Hamming::new();
let dist = hamming.distance("martha", "marhta");
assert_eq!(dist, Some(2));

Jaro:

use eddie::Jaro;
let jaro = Jaro::new();
let sim = jaro.similarity("martha", "marhta");
assert!((sim - 0.94).abs() < 0.01);

Jaro-Winkler:

use eddie::JaroWinkler;
let jarwin = JaroWinkler::new();
let sim = jarwin.similarity("martha", "marhta");
assert!((sim - 0.96).abs() < 0.01);

Complementary metrics

The main metric methods are complemented with inverted and/or relative versions. The naming convention across the crate is following:

• distance — a number of edits required to transform one string to the other;
• rel_dist — a distance between two strings, relative to string length (inversion of similarity);
• similarity — similarity between two strings (inversion of relative distance).

Performance

Below are the tables with performance measurements for typical word lengths, including alternative implementations from crates.io that have Unicode support. Produced on Intel Core i5-4278U 2,6 GHz. All measurements are in microseconds.

Levenshtein

Damerau-Levenshtein

Hamming

Jaro

Jaro-Winkler