Fast JavaScript Max/Min | Hacker News Reader

Fast JavaScript Max/Min

ejohn.org

62 pointsDanielRibeiro17 comments

I used his technique for a while but then I had to find the max and min of an array of 1 million values which is more than the maximum number of arguments to a function in most javascript implementation so I rewrote it a such:

Array.prototype.min = function() {

  var increment = 50000;

  if(this.length > increment){

    var reduced_array = [];

    for(var i=0;i<this.length;i+=increment) {

      reduced_array.push(Math.min.apply(Math,

this.slice(i,i+increment-1))); }

  }else {

    return Math.min.apply(Math, this);


  }

  return reduced_array.min();

}; Array.prototype.max = function(array) {

  var increment = 50000;

  if(this.length > increment){

    var reduced_array = [];

    for(var i=0;i<this.length;i+=increment) {

      reduced_array.push(Math.max.apply(Math, this.slice(i,i+increment-1)));

    }

  }else {

    return Math.max.apply(Math, this);
  }

I'm sure someone can suggest a better way.

They way you've done it seems fine, although if you're trying to find minimums of such large arrays in javascript I'd assume you're doing something wrong.

Javascript really isn't a great choice of language for big processing tasks; stuff like that should be moved away where possible (granted it isn't always an option).

Yeah I would not be doing it unless it was necessary. When I wrote that I was parsing a large file using javascript (from the client filesystem) and the round trip to the server would have been longer than this little function takes.

It might be faster to write the code, but it's definitely not the fastest way to find the minimum/maximum value. Some simple test cases I wrote show a simple for loop iterator over an array to be 2x faster than Math.min.apply in Firefox 4, and 3x faster in Chrome 10.

http://jsperf.com/fastest-array-min-max

If you take out the (currently unused: it is updated but never referred to afterwards) the first-index-of-the-smallest-value-was-found-at variable the difference in FF is even larger.

These are the results on FF 3.6.16 on my desktop:

    ~33.5 ops/sec (million test, using Array.min.apply)
    ~57.0 o/s     (with index var)
    ~86.5 o/s     (without index var)

I'm a little surprised it makes that much difference (I assumed the two set operations, one of which I removed, would be less significant than that in the execution timings compared to the array object lookup in the comparison, given JS arrays aren't actually arrays strictly speaking). The difference may be less significant (zero, in fact) with a JS engine that does dead-code analysis that successfully works out that min_i is set in the loop but otherwise never used.

Ah yes, there was no need of using min_i in my benchmark.

That's a pretty big difference, and I wouldn't have expected that much of a difference for assigning one extra variable in the loop.

I started to feel pretty smug [0], at first. How could John Resig not know this? Then I saw that it was from 2007. Sheesh, I probably first learned about this trick from that post.

[0] http://stackoverflow.com/questions/5020954/adding-functions-...

Underscore.js is pretty good about delegating to built-in functions when available in its little utility functions. http://documentcloud.github.com/underscore/docs/underscore.h...

They surely missed this one though.

as Underscore.js do this:

     _.max = function(obj, iterator, context) {
        if (!iterator && _.isArray(obj)) return Math.max.apply(Math, obj);
        var result = {computed : -Infinity};
        each(obj, function(value, index, list) {
          var computed = iterator ? iterator.call(context, value, index, list) : value;
          computed >= result.computed && (result = {value : value, computed : computed});
        });
        return result.value;
     };


     _.min = function(obj, iterator, context) {
        if (!iterator && _.isArray(obj)) return Math.min.apply(Math, obj);
        var result = {computed : Infinity};
        each(obj, function(value, index, list) {
          var computed = iterator ? iterator.call(context, value, index, list) : value;
          computed < result.computed && (result = {value : value, computed : computed});
        });
        return result.value;
     };

While John do that:

    Array.max = function( array ){
      return Math.max.apply( Math, array );
    };
    Array.min = function( array ){
      return Math.min.apply( Math, array );
    };

The Underscore functions do exactly the same as Resig's in the simple case (when passed just an array of numbers). They're longer because they accept custom iterators for non-numerical arrays.

The real reason why Underscore doesn't use "Math.max.apply" and "Math.min.apply", is that they only work on small arrays of numbers. After you get beyond 40-50,000-ish (last time I checked), some browsers will throw an exception because you can't pass that many arguments to a function.

I remember running into this in real-world code, trying to calculate the brightest pixel in a <canvas> imageData array.

> Posted: January 13th, 2007

That's four years ago.

> 4 points by clvv 2 hours ago | link | parent | flag

That's 2 hours ago.

In the post, he asks this question: "What's the fastest way to find the largest, or smallest, number in an array?"

I guess he means fastest code to write, not the fastest code to run. If you have, say, a million numbers, I'd try doing this using concurrency. If there are two CPUs (or cores), find the maximum in the bottom half of the array and the maximum in the top half in parallel. Then return the maximum of the two maximums.

I don't know much JavaScript - does it have built-in support for concurrency?

I don't like it if it is not documented that Math.max can take more than two arguments. Can we count on all future implementations to accept more than two arguments?

It is documented everywhere I looked. Check out pg162 of http://www.ecma-international.org/publications/files/ECMA-ST..., for example.

OK - in the article (or the comments) it said that the documentation everywhere says two arguments only.

Reply on news.ycombinator.com