Skip to content

Instantly share code, notes, and snippets.

@andrei-m
Last active January 12, 2024 23:00
Star You must be signed in to star a gist
Save andrei-m/982927 to your computer and use it in GitHub Desktop.
Levenshtein distance between two given strings implemented in JavaScript and usable as a Node.js module
/*
Copyright (c) 2011 Andrei Mackenzie
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
// Compute the edit distance between the two given strings
exports.getEditDistance = function(a, b){
if(a.length == 0) return b.length;
if(b.length == 0) return a.length;
var matrix = [];
// increment along the first column of each row
var i;
for(i = 0; i <= b.length; i++){
matrix[i] = [i];
}
// increment each column in the first row
var j;
for(j = 0; j <= a.length; j++){
matrix[0][j] = j;
}
// Fill in the rest of the matrix
for(i = 1; i <= b.length; i++){
for(j = 1; j <= a.length; j++){
if(b.charAt(i-1) == a.charAt(j-1)){
matrix[i][j] = matrix[i-1][j-1];
} else {
matrix[i][j] = Math.min(matrix[i-1][j-1] + 1, // substitution
Math.min(matrix[i][j-1] + 1, // insertion
matrix[i-1][j] + 1)); // deletion
}
}
}
return matrix[b.length][a.length];
};
@cowdude
Copy link

cowdude commented Mar 6, 2013

Thanks for the implementation. Does wonders, exactly what I needed for my NodeJS script.

@dotnetCarpenter
Copy link

Sorry if this is dumb question, but what exactly is the use case?

@andrei-m
Copy link
Author

@dotnetCarpenter I've used this as part of a simple suggestion system. It works by calculating edit distances between some input string and all known acceptable inputs. If the edit distance of any is within an acceptable range, the system can offer a suggestion - e.g. input 'JvaaScript' - "Did you mean 'JavaScript'?"

@its-your-favorite
Copy link

Also mhosen's implementation is incorrect. Provides distance between "house" and "house" as 5, should be 0.

@achatha
Copy link

achatha commented Aug 28, 2015

mhosen's implementation is actually computing hamming distance and NOT Levenshtein.

@pablobarria
Copy link

This is great. Thanks!

@HamletDRC
Copy link

If I use this in my own MIT licensed github project... then what do I need to do to attribute the code?

@bdelespierre
Copy link

bdelespierre commented Jun 8, 2016

the exact same algorithm, condensed for your courtesy:

String.prototype.levenstein = function(string) {
    var a = this, b = string + "", m = [], i, j, min = Math.min;

    if (!(a && b)) return (b || a).length;

    for (i = 0; i <= b.length; m[i] = [i++]);
    for (j = 0; j <= a.length; m[0][j] = j++);

    for (i = 1; i <= b.length; i++) {
        for (j = 1; j <= a.length; j++) {
            m[i][j] = b.charAt(i - 1) == a.charAt(j - 1)
                ? m[i - 1][j - 1]
                : m[i][j] = min(
                    m[i - 1][j - 1] + 1, 
                    min(m[i][j - 1] + 1, m[i - 1 ][j]))
        }
    }

    return m[b.length][a.length];
}

@scottgelin
Copy link

@bdelespierre - Thanks - that is a useful snippet, although there is a slight mistake when compared with the original -- missed the + 1 for the deletion case -- here is your snippet updated with that small fix

String.prototype.levenstein = function(string) {
    var a = this, b = string + "", m = [], i, j, min = Math.min;

    if (!(a && b)) return (b || a).length;

    for (i = 0; i <= b.length; m[i] = [i++]);
    for (j = 0; j <= a.length; m[0][j] = j++);

    for (i = 1; i <= b.length; i++) {
        for (j = 1; j <= a.length; j++) {
            m[i][j] = b.charAt(i - 1) == a.charAt(j - 1)
                ? m[i - 1][j - 1]
                : m[i][j] = min(
                    m[i - 1][j - 1] + 1, 
                    min(m[i][j - 1] + 1, m[i - 1 ][j] + 1))
        }
    }

    return m[b.length][a.length];
}

@nahidakbar
Copy link

nahidakbar commented Jun 16, 2016

Cache matrix for even faster performance.

@ermauliks
Copy link

Thanks for sharing :)

@rd4k1
Copy link

rd4k1 commented Aug 19, 2016

Having four loops was really bugging me.
Also, Math.min can take more than two numbers

var levenshtein = function(a, b){
    if(!a || !b) return (a || b).length;
    var m = [];
    for(var i = 0; i <= b.length; i++){
        m[i] = [i];
        if(i === 0) continue;
        for(var j = 0; j <= a.length; j++){
            m[0][j] = j;
            if(j === 0) continue;
            m[i][j] = b.charAt(i - 1) == a.charAt(j - 1) ? m[i - 1][j - 1] : Math.min(
                m[i-1][j-1] + 1,
                m[i][j-1] + 1,
                m[i-1][j] + 1
            );
        }
    }
    return m[b.length][a.length];
};

@milot-mirdita
Copy link

Here a version that only needs O(min(m,n)) memory, instead of the original's O(m*n):

var levenshtein = function(a, b) {
  if(a.length == 0) return b.length; 
  if(b.length == 0) return a.length;

  // swap to save some memory O(min(a,b)) instead of O(a)
  if(a.length > b.length) {
    var tmp = a;
    a = b;
    b = tmp;
  }

  var row = [];
  // init the row
  for(var i = 0; i <= a.length; i++){
    row[i] = i;
  }

  // fill in the rest
  for(var i = 1; i <= b.length; i++){
    var prev = i;
    for(var j = 1; j <= a.length; j++){
      var val;
      if(b.charAt(i-1) == a.charAt(j-1)){
        val = row[j-1]; // match
      } else {
        val = Math.min(row[j-1] + 1, // substitution
                       prev + 1,     // insertion
                       row[j] + 1);  // deletion
      }
      row[j - 1] = prev;
      prev = val;
    }
    row[a.length] = prev;
  }

  return row[a.length];
}

Runtime should be the same. Code is also licensed as MIT, same as OP.

@kigiri
Copy link

kigiri commented Nov 26, 2016

I did some test and it turns out caching Math.min or passing it more than 2 arguments was a huge preformance loss (60% slower on V8)
My guess is that v8 has some highly optimised Math.min that takes only 2 args and he isn't detecting it if we don't call min from Math.
Caching the .length of the strings was actualy slower, == vs === made no significative differences.

so after some iterations this was the fastests i was able to get :

const levenshtein = (a, b) => {
  if (a.length === 0) return b.length
  if (b.length === 0) return a.length
  let tmp, i, j, prev, val, row
  // swap to save some memory O(min(a,b)) instead of O(a)
  if (a.length > b.length) {
    tmp = a
    a = b
    b = tmp
  }

  row = Array(a.length + 1)
  // init the row
  for (i = 0; i <= a.length; i++) {
    row[i] = i
  }

  // fill in the rest
  for (i = 1; i <= b.length; i++) {
    prev = i
    for (j = 1; j <= a.length; j++) {
      if (b[i-1] === a[j-1]) {
        val = row[j-1] // match
      } else {
        val = Math.min(row[j-1] + 1, // substitution
              Math.min(prev + 1,     // insertion
                       row[j] + 1))  // deletion
      }
      row[j - 1] = prev
      prev = val
    }
    row[a.length] = prev
  }
  return row[a.length]
}

Which is mainly just milto-mirdita version with the 2 Math.min, hoisting the variables, and fixed length array Array(a.length + 1)
wierdly enough it does a significative difference.
still MIT

@rksm
Copy link

rksm commented Jan 14, 2017

@kigiri

row isn't declared.

I suggest let tmp, i, j, prev, val => let tmp, i, j, prev, val, row;

@jsdevtom
Copy link

jsdevtom commented Feb 9, 2017

Here is the jsperf test of @kigiri's 50% faster solution with @rksm's row variable declaration.

@DerekZiemba
Copy link

DerekZiemba commented Apr 13, 2017

Here's a JSBench of almost every implementation on this post. I managed to cut another 5% off kigiri's fastest in everything except firefox.
https://jsperf.com/levenshtein-distance-bench/1

Below is the fastest. It's based off of kirgiri's fastest but improves it by caching array lengths and reducing and reusing variables where possible.

edit: updated with fix from @pukahontas

function dziemba_levenshtein(a, b) {
  var tmp;
  if (a.length === 0) { return b.length; }
  if (b.length === 0) { return a.length; }
  if (a.length > b.length) { tmp = a; a = b; b = tmp; }

  var i, j, res, alen = a.length, blen = b.length, row = Array(alen);
  for (i = 0; i <= alen; i++) { row[i] = i; }

  for (i = 1; i <= blen; i++) {
    res = i;
    for (j = 1; j <= alen; j++) {
      tmp = row[j - 1];
      row[j - 1] = res;
      res = b[i - 1] === a[j - 1] ? tmp : Math.min(tmp + 1, Math.min(res + 1, row[j] + 1));
    }
    row[j - 1] = res;
  }
  return res;
}

@mattotoole0
Copy link

Thanks for all the hard work guys, very helpful!

@jonha892
Copy link

@DerekZiemba

I'm afraid there is a bug in your code. For my example ("badbadnotgood", "s") it returned 2 instead of 13.

@isaklafleur
Copy link

Have you seen this git?
JS-Levenshtein

$ npm run bench
         655,992 op/s » js-levenshtein
         542,796 op/s » leven
         497,966 op/s » talisman
         386,839 op/s » levenshtein-edit-distance
         254,941 op/s » fast-levenshtein
          69,857 op/s » levenshtein-component
          21,688 op/s » levdist
          24,631 op/s » ld
          21,834 op/s » natural
          13,984 op/s » levenshtein

@isaklafleur
Copy link

Anyone know how to implement this in Javascript?
String Matching and Clustering

I'm thinking of the clustering part. So calculate the score I'm looking at this package. a javascript port of the popular Python package fuzzywuzzy

@ansarisufiyan777
Copy link

ansarisufiyan777 commented Jul 12, 2017

function fuzzyMatched (comparer, comparitor, matchCount) {
var isMatched = false;
a = comparer.trim().toLowerCase();
b = comparitor.trim().toLowerCase();
if(a.length == 0) return b.length;
if(b.length == 0) return a.length;
var matrix = [];

// increment along the first column of each row
var i;
for(i = 0; i <= b.length; i++){
    matrix[i] = [i];
}

// increment each column in the first row
var j;
for(j = 0; j <= a.length; j++){
    matrix[0][j] = j;
}

// Fill in the rest of the matrix
for(i = 1; i <= b.length; i++){
    for(j = 1; j <= a.length; j++){
        if(b.charAt(i-1) == a.charAt(j-1)){
            matrix[i][j] = matrix[i-1][j-1];
        } else {
            matrix[i][j] = Math.min(matrix[i-1][j-1] + 1, // substitution
                                    Math.min(matrix[i][j-1] + 1, // insertion
                                             matrix[i-1][j] + 1)); // deletion
        }
    }
}
var fuzzyDistance = matrix[b.length][a.length];
var cLength = Math.max(a.length, b.length);
var score = 1.0 - (fuzzyDistance / cLength);
if (score > matchCount)
    isMatched = true;

return isMatched;

}

In the above method youll have to pass two parameters and the maximum fuzzy match count, for example if you are passing 0.9 then the it will check match count 90% and gives you the result true or false.

fuzzyMatched("abc","abc1",0.7)
true
fuzzyMatched("abc","abc1",0.8)
false

@slidenerd
Copy link



String.prototype.levenstein = function(string) {
    var a = this, b = string + "", m = [], i, j, min = Math.min;

    if (!(a && b)) return (b || a).length;

    for (i = 0; i <= b.length; m[i] = [i++]);
    for (j = 0; j <= a.length; m[0][j] = j++);

    for (i = 1; i <= b.length; i++) {
        for (j = 1; j <= a.length; j++) {
            m[i][j] = b.charAt(i - 1) == a.charAt(j - 1)
                ? m[i - 1][j - 1]
                : m[i][j] = min(
                    m[i - 1][j - 1] + 1, 
                    min(m[i][j - 1] + 1, m[i - 1 ][j]))
        }
    }

    return m[b.length][a.length];
}

function findNearestMatchingSymbol(symbol, coins){
  let combinations = []
  for(let i = 0; i < symbol.length; i++){
    for(let j = i + 1; j <= symbol.length; j++){
      combinations.push(symbol.substring(i,j))
    }
  }

  let scores = []
  coins.forEach(coin=>{
    for(let current of combinations){
      if(coin.includes(current)){
        scores.push({
          symbol: current,
          coin: coin,
          score: current.levenstein(coin)
        }) 
      }
    }
  })
  scores.sort((a,b)=>a.score - b.score)
  console.log(scores)
}

findNearestMatchingSymbol('BTR', coins.map(i=>i.symbol))

This gets all the close symbols but takes too many iterations

@teleranek
Copy link

teleranek commented Jan 20, 2018

Providing that string is shorter than 128 chars, this version based on kigiri's is ~25% faster (chrome 63):

const levenshtein = (a,b)=>{
  let alen = a.length;
  let blen = b.length;
  if (alen === 0) return blen;
  if (blen=== 0) return alen;
  let tmp, i, j, prev, val, row, ma, mb, mc, md, bprev;

  if (alen> blen) {
    tmp = a;
    a = b;
    b = tmp;
  }

  row = new Int8Array(alen+1);
  // init the row
  for (i = 0; i <= alen; i++) {
    row[i] = i;
  }

  // fill in the rest
  for (i = 1; i <= blen; i++) {
    prev = i;
    bprev = b[i - 1]
    for (j = 1; j <= alen; j++) {
      if (bprev === a[j - 1]) {
        val = row[j-1];
      } else {
          ma = prev+1;
          mb = row[j]+1;
          mc = ma - ((ma - mb) & ((mb - ma) >> 7));
          md = row[j-1]+1;
          val = mc - ((mc - md) & ((md - mc) >> 7));
      }
      row[j - 1] = prev;
      prev = val;
    }
    row[alen] = prev;
  }
  return row[alen];
}

https://jsperf.com/levenshteinteleranek/

@mirow
Copy link

mirow commented Feb 2, 2018

@teleranek, your solution is broken as well.

levenshtein('helli', 'elli') returns 3, while levenshtein('elli', 'helli') returns 1

@apappas1129
Copy link

👏 Awesome!🤟

@pukahontas
Copy link

@DerekZiemba
Found the error in your algorithm as pointed out by @jonha892
It was missing the last write of the result to row so that value wasn't available the next iteration.

function dziemba_levenshtein(a, b){
	var tmp;
	if (a.length === 0) { return b.length; }
	if (b.length === 0) { return a.length; }
	if (a.length > b.length) { tmp = a; a = b; b = tmp; }

	var i, j, res, alen = a.length, blen = b.length, row = Array(alen);
	for (i = 0; i <= alen; i++) { row[i] = i; }

	for (i = 1; i <= blen; i++) {
		res = i;
		for (j = 1; j <= alen; j++) {
			tmp = row[j - 1];
			row[j - 1] = res;
			res = Math.min(tmp + (b[i - 1] !== a[j - 1]), res + 1, row[j] + 1);
		}
		row[j - 1] = res; // This was the missing line
	}
	return res;
}

@henning410
Copy link

henning410 commented Nov 29, 2022

@milot-mirdita

Hey i wanna ask, if it is okay for you if I use your code in an own npm package, where i calculate the character-error-rate for ASR.
I can also give you credits on the code

@milot-mirdita
Copy link

@henning410 Sure! its MIT licensed, you don't have to ask for permission as long as you stick to the license terms.

@DMacredasilva
Copy link

boa noite consigo rodar bot mais nao consigo fazer nenhuma da funçoes rodar

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment