Running sum with Rx

Question

Hi,

There must be somebody out there who solved that already. Imagine I have a class that raises periodically an event about the change of a value (e.g. PropertyChanged) That value is nothing else than amount of money.

Now, I would like to make use of Rx so that I get the sum of the increasement of that last 10mins. e.g. BufferWithTime doesn't help, as I always need the last 10mins.

Any ideas how I can do this?

tia Martin

Answer 1

The solution below involves keeping the state of the relevant event data in the previous ten minutes in a list using Observable.Scan. State is maintained as a list of tuples with an int (money) and a DateTime as values.

var events = Observable.FromEvent<YourEventArgs>(
    h => SomeEvent += h, h => SomeEvent -= h);
var runningSums =
    events.Scan(new List<Tuple<int, DateTime>>(),
                (l, e) =>
                {
                    var now = DateTime.Now;
                    // Add last event data to list.
                    l.Add(Tuple.Create(e.EventArgs.Money, now));
                    // Return the correct part of the list (everything
                    // from the last ten minutes).
                    return l.Where(t => (now - t.Item2) <
                                   TimeSpan.FromMinutes(10)).ToList();
                 })
          .Select(l => l.Sum(t => t.Item1));
runningSums.Subscribe(sum => Console.WriteLine(sum));

EDIT: Example that doesn't return a new list for every event:

var events = Observable.FromEvent<YourEventArgs>(
    h => SomeEvent += h, h => SomeEvent -= h);
var runningSums =
    events.Scan(Tuple.Create(new List<Tuple<int, DateTime>>(),
                             DateTime.Now - TimeSpan.FromMinutes(10)),
                (l, e) =>
                {
                    var now = DateTime.Now;
                    l.Item1.Add(Tuple.Create(e.EventArgs.Nr, now));
                    // if (trimming-condition) then trim front of list...
                    return Tuple.Create(l.Item1, now - TimeSpan.FromMinutes(10));
                })
          .Select(l => l.Item1.Where(t => t.Item2 > l.Item2).Sum(t => t.Item1));
runningSums.Subscribe(sum => Console.WriteLine(sum));

Answer 2

Well, check out the following solution. It builds on previously presented solution here, but drops the pure functional style for the sake of efficiency (and readability, i think). It reuses aswell the built in type Timestamped to track the timing...

cheers

    public static class RxEntentsions
        {
            class TimeLimitedList<T>
            {
                public List<Timestamped<T>> Values = new List<Timestamped<T>>();
                TimeSpan span;
                public TimeLimitedList(TimeSpan sp) { span = sp; }
                public void Add(Timestamped<T> v)
                {
                    Values.Add(v);
                    Values.RemoveAll(a => a.Timestamp < (DateTime.Now - span));
                }
            }

            public static IObservable<List<Timestamped<TSource>>> SlidingWindow<TSource>(this IObservable<Timestamped<TSource>> source, TimeSpan slidingWindow)
            {
                return source.Scan0(new TimeLimitedList<TSource>(slidingWindow), (acc, v) => { acc.Add(v); return acc; }).Select(a => a.Values);
            }
        }


    static void Main(string[] args)
    {
        var gen = Observable.Interval(TimeSpan.FromSeconds(0.25d)).Timestamp();
        gen.SlidingWindow(TimeSpan.FromSeconds(1)).Subscribe(slw => {slw.ForEach(e=> Console.WriteLine(e)); Console.WriteLine("--------");});
        Console.ReadLine();
    }