Best way to randomize an array with .NET

Question

The following implementation uses the Fisher-Yates algorithm AKA the Knuth Shuffle. It runs in O(n) time and shuffles in place, so is better performing than the 'sort by random' technique, although it is more lines of code. See here for some comparative performance measurements. I have used System.Random, which is fine for non-cryptographic purposes.*

static class RandomExtensions
{
    public static void Shuffle<T> (this Random rng, T[] array)
    {
        int n = array.Length;
        while (n > 1) 
        {
            int k = rng.Next(n--);
            T temp = array[n];
            array[n] = array[k];
            array[k] = temp;
        }
    }
}

Usage:

var array = new int[] {1, 2, 3, 4};
var rng = new Random();
rng.Shuffle(array);
rng.Shuffle(array); // different order from first call to Shuffle

* For longer arrays, in order to make the (extremely large) number of permutations equally probable it would be necessary to run a pseudo-random number generator (PRNG) through many iterations for each swap to produce enough entropy. For a 500-element array only a very small fraction of the possible 500! permutations will be possible to obtain using a PRNG. Nevertheless, the Fisher-Yates algorithm is unbiased and therefore the shuffle will be as good as the RNG you use.

If you're on .NET 3.5, you can use the following IEnumerable coolness:

Random rnd=new Random();
string[] MyRandomArray = MyArray.OrderBy(x => rnd.Next()).ToArray();    

Edit: and here's the corresponding VB.NET code:

Dim rnd As New System.Random
Dim MyRandomArray = MyArray.OrderBy(Function() rnd.Next()).ToArray()

Second edit, in response to remarks that System.Random "isn't threadsafe" and "only suitable for toy apps" due to returning a time-based sequence: as used in my example, Random() is perfectly thread-safe, unless you're allowing the routine in which you randomize the array to be re-entered, in which case you'll need something like lock (MyRandomArray) anyway in order not to corrupt your data, which will protect rnd as well.

Also, it should be well-understood that System.Random as a source of entropy isn't very strong. As noted in the MSDN documentation, you should use something derived from System.Security.Cryptography.RandomNumberGenerator if you're doing anything security-related. For example:

using System.Security.Cryptography;

...

RNGCryptoServiceProvider rnd = new RNGCryptoServiceProvider();
string[] MyRandomArray = MyArray.OrderBy(x => GetNextInt32(rnd)).ToArray();

...

static int GetNextInt32(RNGCryptoServiceProvider rnd)
    {
        byte[] randomInt = new byte[4];
        rnd.GetBytes(randomInt);
        return Convert.ToInt32(randomInt[0]);
    }

You're looking for a shuffling algorithm, right?

Okay, there are two ways to do this: the clever-but-people-always-seem-to-misunderstand-it-and-get-it-wrong-so-maybe-its-not-that-clever-after-all way, and the dumb-as-rocks-but-who-cares-because-it-works way.

Dumb way

• Create a duplicate of your first array, but tag each string should with a random number.
• Sort the duplicate array with respect to the random number.

This algorithm works well, but make sure that your random number generator is unlikely to tag two strings with the same number. Because of the so-called Birthday Paradox, this happens more often than you might expect. Its time complexity is O(n log n).

Clever way

I'll describe this as a recursive algorithm:

To shuffle an array of size n (indices in the range [0..n-1]):

if n = 0

• do nothing

if n > 0

• (recursive step) shuffle the first n-1 elements of the array
• choose a random index, x, in the range [0..n-1]
• swap the element at index n-1 with the element at index x

The iterative equivalent is to walk an iterator through the array, swapping with random elements as you go along, but notice that you cannot swap with an element after the one that the iterator points to. This is a very common mistake, and leads to a biased shuffle.

Time complexity is O(n).

This algorithm is simple but not efficient, O(N²). All the "order by" algorithms are typically O(N log N). It probably doesn't make a difference below hundreds of thousands of elements but it would for large lists.

var stringlist = ... // add your values to stringlist

var r = new Random();

var res = new List<string>(stringlist.Count);

while (stringlist.Count >0)
{
   var i = r.Next(stringlist.Count);
   res.Add(stringlist[i]);
   stringlist.RemoveAt(i);
}

The reason why it's O(N²) is subtle: List.RemoveAt() is a O(N) operation unless you remove in order from the end.

You can also make an extention method out of Matt Howells. Example.

   namespace System
    {
        public static class MSSystemExtenstions
        {
            private static Random rng = new Random();
            public static void Shuffle<T>(this T[] array)
            {
                rng = new Random();
                int n = array.Length;
                while (n > 1)
                {
                    int k = rng.Next(n);
                    n--;
                    T temp = array[n];
                    array[n] = array[k];
                    array[k] = temp;
                }
            }
        }
    }

Then you can just use it like:

        string[] names = new string[] {
                "Aaron Moline1", 
                "Aaron Moline2", 
                "Aaron Moline3", 
                "Aaron Moline4", 
                "Aaron Moline5", 
                "Aaron Moline6", 
                "Aaron Moline7", 
                "Aaron Moline8", 
                "Aaron Moline9", 
            };
        names.Shuffle<string>();