Trouble Implementing a Sliding Window in Rx

Question

I created a SlidingWindow operator for reactive extensions because I want to easily monitor things like rolling averages, etc. As a simple example, I want to s

Answer 1

Just source.Window(count, 1) - or source.Buffer(count, 1) It be a window/buffer of "count" items, sliding by one.

Answer 2

Using your original test, with an argument of 3 for count, this gives the desired results:

public static IObservable<IList<T>> SlidingWindow<T>(
    this IObservable<T> source, int count)
{
    return source.Buffer(count, 1)
                 .Where(list => list.Count == count);
}

Testing like this:

var source = Observable.Range(1, 5);
var query = source.SlidingWindow(3);
using (query.Subscribe(i => Console.WriteLine(string.Join(",", i))))
{

}

Output:

1,2,3
2,3,4
3,4,5

Answer 3

Try this instead - I'd have to sit and have a think about it's relative performance, but it's at least likely as good, and way easier to read:

public static IObservable<IList<T>> SlidingWindow<T>(
       this IObservable<T> src, 
       int windowSize)
{
    var feed = src.Publish().RefCount();    
    // (skip 0) + (skip 1) + (skip 2) + ... + (skip nth) => return as list  
    return Observable.Zip(
       Enumerable.Range(0, windowSize)
           .Select(skip => feed.Skip(skip))
           .ToArray());
}

Test rig:

var source = Observable.Range(0, 10);
var query = source.SlidingWindow(3);
using(query.Subscribe(Console.WriteLine))
{               
    Console.ReadLine();
}

Output:

ListOf(0,1,2)
ListOf(1,2,3)
ListOf(2,3,4)
ListOf(3,4,5)
ListOf(4,5,6)
...

EDIT: As an aside, I find myself compulsively .Publish().RefCount()ing ever since being burned once by not doing it...I don't think it's strictly required here, tho.

EDIT for yzorg:

If you augment the method like so, you'll see the runtime behavior more clearly:

public static IObservable<IList<T>> SlidingWindow<T>(
    this IObservable<T> src, 
    int windowSize)
{
    var feed = src.Publish().RefCount();    
    // (skip 0) + (skip 1) + (skip 2) + ... + (skip nth) => return as list  
    return Observable.Zip(
    Enumerable.Range(0, windowSize)
        .Select(skip => 
        {
            Console.WriteLine("Skipping {0} els", skip);
            return feed.Skip(skip);
        })
        .ToArray());
}

Answer 4

The sliding window implementations here was not adequate for my idea of a sliding window. The closest was using Buffer(N, 1) but is a problem since it waits for the first N items before it emits the first result then slides beyond the end of the sequence. I want up to N items emitted at a time.

I ended up with this implementation:

public static IObservable<IList<T>> SlidingWindow<T>(this IObservable<T> obs, int windowSize) =>
    Observable.Create<IList<T>>(observer =>
    {
        var buffer = new CircularBuffer<T>(windowSize);
        return obs.Subscribe(
            value =>
            {
                buffer.Add(value);
                observer.OnNext(buffer.ToList());
            },
            ex => observer.OnError(ex),
            () => observer.OnCompleted()
        );
    });

I was initially using a queue for the buffer but wanted to use something a bit more lightweight.

public class CircularBuffer<T> : IReadOnlyList<T>
{
    private readonly T[] buffer;
    private int offset;
    private int count;
    public CircularBuffer(int bufferSize) => this.buffer = new T[bufferSize];
    public int Capacity => buffer.Length;
    public int Count => count;
    public T this[int index] => index < 0 || index >= count
        ? throw new ArgumentOutOfRangeException(nameof(index))
        : buffer[(offset + index) % buffer.Length];
    public void Add(T value)
    {
        buffer[(offset + count) % buffer.Length] = value;
        if (count < buffer.Length) count++;
        else offset = (offset + 1) % buffer.Length;
    }
    public IEnumerator<T> GetEnumerator()
    {
        for (var i = 0; i < count; i++)
            yield return buffer[(offset + i) % buffer.Length];
    }
    IEnumerator IEnumerable.GetEnumerator() => GetEnumerator();
}

It would yield sequences for Observable.Range(0, 10).SlidingWindow(3):

 0,1,2,3,4,5,6,7,8,9
[0]
[0,1]
[0,1,2]
  [1,2,3]
    [2,3,4]
      [3,4,5]
        [4,5,6]
          [5,6,7]
            [6,7,8]
              [7,8,9]