created_at2024-02-12 14:30:45.518261
updated_at2024-02-23 16:25:30.554775
descriptionImplementations of string similarity metrics. Includes Hamming, Levenshtein, OSA, Damerau-Levenshtein, Jaro, Jaro-Winkler, and Sørensen-Dice.
Luiz Otavio Vilas Boas Oliveira (luizvbo)




Rust implementations of string similarity metrics:

The normalized versions return values between 0.0 and 1.0, where 1.0 means an exact match.

There are also generic versions of the functions for non-string inputs.

What is new?

This crate is heavily based on the strsim-rs crate, with some nice additions:

  • Gestalt pattern matching, the algorithm used by python difflib SequenceMatcher
  • Top-N matching, a method to retrieve the best N matches from a collection of choices.
  • Feature selection, allows you to select only the features (metrics) you want to use, reducing the memory footprint of your application.

Top-N Matching

The method get_top_n gets a list of the best matches from a collection of choices. This feature is inspired by the extractBests method from the Python fuzzywuzzy package (now thefuzz).

The get_top_n method takes a query string, an array of choice strings, a cutoff similarity score, an optional number of top matches to return, an optional string processor, and an optional similarity metric. It processes each choice and the query using the provided or default string processor, computes the similarity between the processed query and each processed choice using the provided or default similarity metric, and returns the top-N matches that have a similarity score greater than or equal to the cutoff.

Here's the signature of the get_top_n method:

extern crate fuzzt;
use fuzzt::{algorithms::NormalizedLevenshtein, get_top_n, processors::NullStringProcessor};

fn main() {
    let matches = get_top_n(
        &["apply", "apples", "ape", "applet", "applesauce"],
    assert_eq!(matches, ["apples", "applet", "apply"]);

Feature selection

fuzzt is designed with flexibility in mind, allowing you to select only the features you need for your specific use case. This can help to reduce the footprint of your application and optimize performance.

The crate includes the following features:

  • damerau_levenshtein
  • gestalt
  • hamming
  • jaro
  • levenshtein
  • optimal_string_alignment
  • sorensen_dice

By default, all features are included when you add fuzzt as a dependency. However, you can choose to include only specific features by listing them under the features key in your Cargo.toml file. For example:

fuzzt = { version = "*", default-features = false, features = ["levenshtein", "jaro"] }


Fuzzt is available on crates.io. Add it to your project:

cargo add fuzzt


Go to Docs.rs for the full documentation. You can also clone the repo, and run $ cargo doc --open.


extern crate fuzzt;

use fuzzt::{
    damerau_levenshtein, hamming, jaro, jaro_winkler, levenshtein, normalized_damerau_levenshtein,
    normalized_levenshtein, osa_distance, sequence_matcher, sorensen_dice,

fn main() {
    match hamming("hamming", "hammers") {
        Ok(distance) => assert_eq!(3, distance),
        Err(why) => panic!("{:?}", why),

    assert_eq!(levenshtein("kitten", "sitting"), 3);
    assert!((normalized_levenshtein("kitten", "sitting") - 0.571).abs() < 0.001);
    assert_eq!(osa_distance("ac", "cba"), 3);
    assert_eq!(damerau_levenshtein("ac", "cba"), 2);
    assert!((normalized_damerau_levenshtein("levenshtein", "löwenbräu") - 0.272).abs() < 0.001);
    assert_eq!(jaro("Friedrich Nietzsche", "Jean-Paul Sartre"), 0.3918859649122807);
        jaro_winkler("cheeseburger", "cheese fries"),
        sorensen_dice("web applications", "applications of the web"),
        sequence_matcher("this is a test", "this is a test!"),

Using the generic versions of the functions:

extern crate fuzzt;

use fuzzt::generic_levenshtein;

fn main() {
    assert_eq!(2, generic_levenshtein(&[1, 2, 3], &[0, 2, 5]));



The Hamming distance between two strings of equal length is the number of positions at which the corresponding symbols are different. It measures the minimum number of substitutions required to change one string into the other.


The Levenshtein distance is a string metric for measuring the difference between two sequences. It quantifies how many edits (insertions, deletions, or substitutions) you need to make to change one string into another. The normalized version of this metric gives you a proportion between 0 and 1, where 1 means the strings are identical.

Optimal String Alignment

The Optimal String Alignment (OSA), also known as the restricted Damerau-Levenshtein distance, computes the shortest distance considering only adjacent transpositions. This means it doesn't allow substrings to move as a block, unlike the Damerau-Levenshtein distance.


Damerau-Levenshtein distance is an extension of the Levenshtein distance, allowing for transpositions of two adjacent characters along with insertions, deletions, and substitutions. The normalized version gives a proportion between 0 and 1, where 1 means the strings are identical.

Jaro and Jaro-Winkler

The Jaro distance allows for transpositions and takes into account the number and order of common characters between two strings. The Jaro-Winkler distance is a modification of the Jaro distance that gives more favorable ratings to strings that match from the beginning.


This coefficient is a statistic used to gauge the similarity of two samples. It's calculated as twice the size of the intersection of the sets, divided by the sum of the sizes of the two sets.

Gestalt Pattern Matching

This is the algorithm used by Python's difflib.SequenceMatcher. It uses a heuristic called "Ratcliff/Obershelp" that computes the doubled number of matching characters divided by the total number of characters in the two strings. It's particularly good at detecting close matches and some types of typos.


If you don't want to install Rust itself, you can run $ ./dev for a development CLI if you have Docker installed.

Benchmarks require a Nightly toolchain. Run $ cargo +nightly bench.



Commit count: 0

cargo fmt