How to implement sublime text like fuzzy search?

Question

How can i implement a sublime-like fuzzy search on select2?

Example, typing \"sta jav sub\" would match \"Stackoverflow javascript sublime like\"

Answer 1

had difficulties with new select2, here what worked

 $("#foo").select2({
   matcher: matcher
 });

function matcher(params, data) {
  // return all opts if seachbox is empty
  if(!params.term) {
    return data;
  } else if(data) {
    var term = params.term.toUpperCase();
    var option = data.text.toUpperCase();
    var j = -1; // remembers position of last found character

    // consider each search character one at a time
    for (var i = 0; i < term.length; i++) {
      var l = term[i];
      if (l == ' ') continue;     // ignore spaces

      j = option.indexOf(l, j+1);     // search for character & update position
      if (j == -1) return false;  // if it's not found, exclude this item
    }
    return data; // return option
  }
}

Answer 2

Here's an alternate matching function. http://jsfiddle.net/trevordixon/pXzj3/4/

function match(search, text) {
    search = search.toUpperCase();
    text = text.toUpperCase();

    var j = -1; // remembers position of last found character

    // consider each search character one at a time
    for (var i = 0; i < search.length; i++) {
        var l = search[i];
        if (l == ' ') continue;     // ignore spaces

        j = text.indexOf(l, j+1);     // search for character & update position
        if (j == -1) return false;  // if it's not found, exclude this item
    }
    return true;
}

This one's faster (according to this test in Chrome), which may start to matter if you're filtering a lot of items.

Answer 3

I wrote some that works very much long Sublime Text's fuzzy match. Achieving this requires a few things.

First, match all characters from a pattern in sequence. Second, score matches such that certain matched characters are worth more points than other.

I came up with a few factors to check for. "CamelCase" letters or letters following a separator (space or underscore) are worth a lot of points. Consecutive matches are worth more. Results found near the start are worth more.

A critically important trick is to find the best matching character. Which is not necessarily the first. Consider fuzzy_match("tk", "The Black Knight"). There are two Ks which could be matched. The second is worth more points because it follows a space.

JavaScript code is below. There is some nuance which is described in more detail in a blog post. There's also an interactive demo. And full source (includes demo, plus C++ implementation) on GitHub.

• Blog Post
• Interactive Demo

• GitHub

  // Returns [bool, score, formattedStr]
  // bool: true if each character in pattern is found sequentially within str
  // score: integer; higher is better match. Value has no intrinsic meaning. Range varies with pattern. 
  //        Can only compare scores with same search pattern.
  // formattedStr: input str with matched characters marked in <b> tags. Delete if unwanted.
  
  function fuzzy_match(pattern, str) {
      // Score consts
      var adjacency_bonus = 5;                // bonus for adjacent matches
      var separator_bonus = 10;               // bonus if match occurs after a separator
      var camel_bonus = 10;                   // bonus if match is uppercase and prev is lower
      var leading_letter_penalty = -3;        // penalty applied for every letter in str before the first match
      var max_leading_letter_penalty = -9;    // maximum penalty for leading letters
      var unmatched_letter_penalty = -1;      // penalty for every letter that doesn't matter
  
      // Loop variables
      var score = 0;
      var patternIdx = 0;
      var patternLength = pattern.length;
      var strIdx = 0;
      var strLength = str.length;
      var prevMatched = false;
      var prevLower = false;
      var prevSeparator = true;       // true so if first letter match gets separator bonus
  
      // Use "best" matched letter if multiple string letters match the pattern
      var bestLetter = null;
      var bestLower = null;
      var bestLetterIdx = null;
      var bestLetterScore = 0;
  
      var matchedIndices = [];
  
      // Loop over strings
      while (strIdx != strLength) {
          var patternChar = patternIdx != patternLength ? pattern.charAt(patternIdx) : null;
          var strChar = str.charAt(strIdx);
  
          var patternLower = patternChar != null ? patternChar.toLowerCase() : null;
          var strLower = strChar.toLowerCase();
          var strUpper = strChar.toUpperCase();
  
          var nextMatch = patternChar && patternLower == strLower;
          var rematch = bestLetter && bestLower == strLower;
  
          var advanced = nextMatch && bestLetter;
          var patternRepeat = bestLetter && patternChar && bestLower == patternLower;
          if (advanced || patternRepeat) {
              score += bestLetterScore;
              matchedIndices.push(bestLetterIdx);
              bestLetter = null;
              bestLower = null;
              bestLetterIdx = null;
              bestLetterScore = 0;
          }
  
          if (nextMatch || rematch) {
              var newScore = 0;
  
              // Apply penalty for each letter before the first pattern match
              // Note: std::max because penalties are negative values. So max is smallest penalty.
              if (patternIdx == 0) {
                  var penalty = Math.max(strIdx * leading_letter_penalty, max_leading_letter_penalty);
                  score += penalty;
              }
  
              // Apply bonus for consecutive bonuses
              if (prevMatched)
                  newScore += adjacency_bonus;
  
              // Apply bonus for matches after a separator
              if (prevSeparator)
                  newScore += separator_bonus;
  
              // Apply bonus across camel case boundaries. Includes "clever" isLetter check.
              if (prevLower && strChar == strUpper && strLower != strUpper)
                  newScore += camel_bonus;
  
              // Update patter index IFF the next pattern letter was matched
              if (nextMatch)
                  ++patternIdx;
  
              // Update best letter in str which may be for a "next" letter or a "rematch"
              if (newScore >= bestLetterScore) {
  
                  // Apply penalty for now skipped letter
                  if (bestLetter != null)
                      score += unmatched_letter_penalty;
  
                  bestLetter = strChar;
                  bestLower = bestLetter.toLowerCase();
                  bestLetterIdx = strIdx;
                  bestLetterScore = newScore;
              }
  
              prevMatched = true;
          }
          else {
              // Append unmatch characters
              formattedStr += strChar;
  
              score += unmatched_letter_penalty;
              prevMatched = false;
          }
  
          // Includes "clever" isLetter check.
          prevLower = strChar == strLower && strLower != strUpper;
          prevSeparator = strChar == '_' || strChar == ' ';
  
          ++strIdx;
      }
  
      // Apply score for last match
      if (bestLetter) {
          score += bestLetterScore;
          matchedIndices.push(bestLetterIdx);
      }
  
      // Finish out formatted string after last pattern matched
      // Build formated string based on matched letters
      var formattedStr = "";
      var lastIdx = 0;
      for (var i = 0; i < matchedIndices.length; ++i) {
          var idx = matchedIndices[i];
          formattedStr += str.substr(lastIdx, idx - lastIdx) + "<b>" + str.charAt(idx) + "</b>";
          lastIdx = idx + 1;
      }
      formattedStr += str.substr(lastIdx, str.length - lastIdx);
  
      var matched = patternIdx == patternLength;
      return [matched, score, formattedStr];
  }
  

Answer 4

albertein's answer doesn't match Trevor's version because the original function performs matching on character basis and not on word basis. Here's a simpler one matching on character basis:

$("#element").select2({
  matcher: function(term, text, opts) {
    var pattern = term.replace(/\s+/g, '').split('').join('.*');
    text.match(new RegExp(pattern, 'i'))
  }
})

Answer 5

var fuzzysearch = function (querystrings, values) {
    return !querystrings.some(function (q) {
        return !values.some(function (v) {
            return v.toLocaleLowerCase().indexOf(q) !== -1;
        });
    });
}

Example searching for title and author in book collection http://jsfiddle.net/runjep/r887etnh/2/

For a 9kb alternative which ranks the search result: http://kiro.me/projects/fuse.html

You may need a polyfill for the 'some' function https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Array/some

var books = [{
    id: 1,
    title: 'The Great Gatsby',
    author: 'F. Scott Fitzgerald'
}, {
    id: 2,
    title: 'The DaVinci Code',
    author: 'Dan Brown'
}, {
    id: 3,
    title: 'Angels & Demons',
    author: 'Dan Brown'
}];
search = function () {
    var queryarray = document.getElementById('inp').value.trim().toLowerCase().split(' ');
    var res = books.filter(function (b) {
        return fs(queryarray, [b.title, b.author]);
    });
    document.getElementById('res').innerHTML = res.map(function (b) {
        return b.title + ' <i> ' + b.author + '</i>';
    }).join('<br/> ');
}
fs = function (qs, vals) {
    return !qs.some(function (q) {
        return !vals.some(function (v) {
            return v.toLocaleLowerCase().indexOf(q) !== -1;
        });
    });
}

<input id="inp" />
<button id="but" onclick="search()">Search</button>
<div id="res"></div>

Answer 6

function fuzzyMe(term, query) {
  var score = 0;
  var termLength = term.length;
  var queryLength = query.length;
  var highlighting = '';
  var ti = 0;
  // -1 would not work as this would break the calculations of bonus
  // points for subsequent character matches. Something like
  // Number.MIN_VALUE would be more appropriate, but unfortunately
  // Number.MIN_VALUE + 1 equals 1...
  var previousMatchingCharacter = -2;
  for (var qi = 0; qi < queryLength && ti < termLength; qi++) {
    var qc = query.charAt(qi);
    var lowerQc = qc.toLowerCase();

    for (; ti < termLength; ti++) {
      var tc = term.charAt(ti);

      if (lowerQc === tc.toLowerCase()) {
        score++;

        if ((previousMatchingCharacter + 1) === ti) {
          score += 2;
        }

        highlighting += "<em>" + tc + "</em>";
        previousMatchingCharacter = ti;
        ti++;
        break;
      } else {
        highlighting += tc;
      }
    }
  }

  highlighting += term.substring(ti, term.length);

  return {
    score: score,
    term: term,
    query: query,
    highlightedTerm: highlighting
  };
}

The above takes care of the fuzziness. Then you can just iterate over all your select 2 elements

$("#element").select2({
  matcher: function(term, text, opt) {
    return fuzzyMe(term, text).highlightedTerm;
  }
});

Credit for fuzzy code -: https://github.com/bripkens/fuzzy.js