How to calculate simple moving average faster in C#?

问题

What is the fastest library/algorithm for calculating simple moving average? I wrote my own, but it takes too long on 330 000 items decimal dataset.

period / time(ms)
20 / 300;
60 / 1500;
120 / 3500.

Here is the code of my method:

public decimal MA_Simple(int period, int ii) {
    if (period != 0 && ii > period) {
        //stp.Start();
        decimal summ = 0;
        for (int i = ii; i > ii - period; i--) {
            summ = summ + Data.Close[i];
        }
        summ = summ / period;
        //stp.Stop();
        //if (ii == 1500) System.Windows.Forms.MessageBox.Show((stp.ElapsedTicks * 1000.0) / Stopwatch.Frequency + " ms");
        return summ;
    } else return -1;
}

The Data.Close[] is a fixed size(1 000 000) decimal array.

回答1:

Your main problem is that you throw away too much information for each iteration. If you want to run this fast, you need to keep a buffer of the same size as the frame length.

This code will run moving averages for your whole dataset:

(Not real C# but you should get the idea)

decimal buffer[] = new decimal[period];
decimal output[] = new decimal[data.Length];
current_index = 0;
for (int i=0; i<data.Length; i++)
    {
        buffer[current_index] = data[i]/period;
        decimal ma = 0.0;
        for (int j=0;j<period;j++)
            {
                ma += buffer[j];
            }
        output[i] = ma;
        current_index = (current_index + 1) % period;
    }
return output;

Please note that it may be tempting to keep a running cumsum instead of keeping the whole buffer and calculating the value for each iteration, but this does not work for very long data lengths as your cumulative sum will grow so big that adding small additional values will result in rounding errors.

回答2:

    public class MovingAverage  
    {
        private Queue<Decimal> samples = new Queue<Decimal>();
        private int windowSize = 16;
        private Decimal sampleAccumulator;
        public Decimal Average { get; private set; }

        /// <summary>
        /// Computes a new windowed average each time a new sample arrives
        /// </summary>
        /// <param name="newSample"></param>
        public void ComputeAverage(Decimal newSample)
        {
            sampleAccumulator += newSample;
            samples.Enqueue(newSample);

            if (samples.Count > windowSize)
            {
                sampleAccumulator -= samples.Dequeue();
            }

            Average = sampleAccumulator / samples.Count;
        }
    }

回答3:

If the data is static, you can preprocess the array to make moving average queries very fast:

decimal[] GetCSum(decimal[] data) {
    decimal csum[] = new decimal[data.Length];
    decimal cursum = 0;
    for(int i=0; i<data.Length; i++) {
        cursum += data[i];
        csum[i] = cursum;
    }
    return csum;
}

Now the moving average calculation is easy and fast:

decimal CSumMovingAverage(decimal[] csum, int period, int ii) {
    if(period == 0 || ii <= period)
        return -1;
    return csum[ii] - csum[ii - period];
}

回答4:

These days, the Math DotNet library has a class called RunningStatistics that will do this for you. If you want to do it over the last "X" items only, use MovingStatistics instead.

Both will calculate running averages, variance, and standard deviation, on the fly with one-pass only and without storing extra copies of the data.

回答5:

The current (accepted) solution contains an inner loop. It would be more efficient to remove this as well. You can see how this is achieved here:

How to efficiently calculate a moving Standard Deviation

回答6:

// simple moving average
int moving_average(double *values, double *&averages, int size, int periods)
{
    double sum = 0;
    for (int i = 0; i < size; i ++)
        if (i < periods) {
            sum += values[i];
            averages[i] = (i == periods - 1) ? sum / (double)periods : 0;
        } else {
            sum = sum - values[i - periods] + values[i];
            averages[i] = sum / (double)periods;
        }
    return (size - periods + 1 > 0) ? size - periods + 1 : 0;
}

One C function, 13 lines of codes, simple moving average. Example of usage:

double *values = new double[10]; // the input
double *averages = new double[10]; // the output
values[0] = 55;
values[1] = 113;
values[2] = 92.6;
...
values[9] = 23;
moving_average(values, averages, 10, 5); // 5-day moving average

回答7:

This is MA I'm using in my app.

double[] MovingAverage(int period, double[] source)
{
    var ma = new double[source.Length];

    double sum = 0;
    for (int bar = 0; bar < period; bar++)
        sum += source[bar];

    ma[period - 1] = sum/period;

    for (int bar = period; bar < source.Length; bar++)
        ma[bar] = ma[bar - 1] + source[bar]/period
                              - source[bar - period]/period;

    return ma;
}

Once you have it calculated for the whole data series, you can grab a particular value instantly.

回答8:

Here's how I tried it. But warning I'm a complete amateur so this may be completely wrong.

List<decimal> MovingAverage(int period, decimal[] Data)
{
     decimal[] interval = new decimal[period];
     List<decimal> MAs = new List<decimal>();

     for (int i=0, i < Data.length, i++)
     {
          interval[i % period] = Data[i];
          if (i > period - 1)
          {
               MAs.Add(interval.Average());
          }
     }
     return MAs;
}

Should return a list of decimals containing the moving averages for your data.

回答9:

How aboutQueue ?

using System.Collections.Generic;
using System.Linq;

public class MovingAverage
{
    private readonly Queue<decimal> _queue;
    private readonly int _period;

    public MovingAverage(int period)
    {
        _period = period;
        _queue = new Queue<decimal>(period);
    }

    public double Compute(decimal x)
    {
        if (_queue.Count >= _period)
        {
            _queue.Dequeue();
        }

        _queue.Enqueue(x);

        return _queue.Average();
    }
}

Usage:

MovingAverage ma = new MovingAverage(3);

foreach(var val in new decimal[1,2,3,4,5,6,7,8,9])
{
   Console.WriteLine(ma.Compute(val));
}

回答10:

You don't need to keep a running queue. Just pick the latest new entry to the window and drop off the older entry. Notice that this only uses one loop and no extra storage other than a sum.

  // n is the window for your Simple Moving Average
  public List<double> GetMovingAverages(List<Price> prices, int n)
  {
    var movingAverages = new double[prices.Count];
    var runningTotal = 0.0d;       

    for (int i = 0; i < prices.Count; ++i)
    {
      runningTotal += prices[i].Value;
      if( i - n >= 0) {
        var lost = prices[i - n].Value;
        runningTotal -= lost;
        movingAverages[i] = runningTotal / n;
      }
    }
    return movingAverages.ToList();
  }

回答11:

I find the provide answers a bit to memory hungry, and slow, you asked for fast. Add 2 fields one to keep the running total and one for the times the value changed as average is the sum/count of a list of values. I added a Add method, however you can also just use variables in a method….

public class Sample
{
    private decimal sum = 0;
    private uint count = 0;

    public void Add(decimal value)
    {
        sum += value;
        count++;
    }

    public decimal AverageMove => count > 0 ? sum / count : 0;
}

to make it thread safe:

public class ThreadSafeSample
{
private decimal sum = 0;
private uint count = 0;

private static object locker = new object();
public void Add(decimal value)
{
    lock (locker)
    {
        sum += value;
        count++;
    }
}

public decimal AverageMove => count > 0 ? sum / count : 0;

}

回答12:

/// <summary>
/// Fast low CPU usage moving average based on floating point math
/// Note: This algorithm algorithm compensates for floating point error by re-summing the buffer for every 1000 values
/// </summary>
public class FastMovingAverageDouble
{
    /// <summary>
    /// Adjust this as you see fit to suit the scenario
    /// </summary>
    const int MaximumWindowSize = 100;

    /// <summary>
    /// Adjust this as you see fit
    /// </summary>
    const int RecalculateEveryXValues = 1000;

    /// <summary>
    /// Initializes moving average for specified window size
    /// </summary>
    /// <param name="_WindowSize">Size of moving average window between 2 and MaximumWindowSize 
    /// Note: this value should not be too large and also bear in mind the possibility of overflow and floating point error as this class internally keeps a sum of the values within the window</param>
    public FastMovingAverageDouble(int _WindowSize)
    {
        if (_WindowSize < 2)
        {
            _WindowSize = 2;
        }
        else if (_WindowSize > MaximumWindowSize)
        {
            _WindowSize = MaximumWindowSize;
        }
        m_WindowSize = _WindowSize;
    }
    private object SyncRoot = new object();
    private Queue<double> Buffer = new Queue<double>();
    private int m_WindowSize;
    private double m_MovingAverage = 0d;
    private double MovingSum = 0d;
    private bool BufferFull;
    private int Counter = 0;

    /// <summary>
    /// Calculated moving average
    /// </summary>
    public double MovingAverage
    {
        get
        {
            lock (SyncRoot)
            {
                return m_MovingAverage;
            }
        }
    }

    /// <summary>
    /// Size of moving average window set by constructor during intialization
    /// </summary>
    public int WindowSize
    {
        get
        {
            return m_WindowSize;
        }
    }

    /// <summary>
    /// Add new value to sequence and recalculate moving average seee <see cref="MovingAverage"/>
    /// </summary>
    /// <param name="NewValue">New value to be added</param>
    public void AddValue(int NewValue)
    {
        lock (SyncRoot)
        {
            Buffer.Enqueue(NewValue);
            MovingSum += NewValue;
            if (!BufferFull)
            {
                int BufferSize = Buffer.Count;
                BufferFull = BufferSize == WindowSize;
                m_MovingAverage = MovingSum / BufferSize;
            }
            else
            {
                Counter += 1;
                if (Counter > RecalculateEveryXValues)
                {
                    MovingSum = 0;
                    foreach (double BufferValue in Buffer)
                    {
                        MovingSum += BufferValue;
                    }
                    Counter = 0;
                }
                MovingSum -= Buffer.Dequeue();
                m_MovingAverage = MovingSum / WindowSize;
            }
        }
    }
}

回答13:

Tested with Dotnet Core 3 & Linq:

int period = 20;
for(int k=0;data.Count()-period;k++){
   decimal summe = data.Skip(k).Take(period).Sum();
   summe /= (decimal)period;
}

It does rely on Linq and its internal optimization, did not time it.
Uses Skip() and Take() as a "rangeBetween" solution for moving average and then divide the summe by the period quantity.
*The for loop is upper capped to avoid incomplete sum operations.
Reference (C# Microsoft): Skip(), Take(), Sum();

来源：https://stackoverflow.com/questions/12884600/how-to-calculate-simple-moving-average-faster-in-c

标签

algorithm

financial

moving-average