MapReduce Entwurfsmuster – Numerische Aggregation (Standardabweichung 1/2)

MedianStdDevMapper.cs

public class MedianStdDevMapper : MapperBase
{
  public override void Map(string inputLine, MapperContext context)
  {
    var parsed = XmlUtils.ParseXml(inputLine);

    if (parsed == null 
        || !parsed.ContainsKey("CreationDate") 
        || !parsed.ContainsKey("Text"))
    {
      context.CoreContext.IncrementCounter(
        "Median / Std. Dev. Mapper", "Invalid Rows", 1);
      return;
    }

    DateTime creationDate;
    if (!DateTime.TryParse(parsed["CreationDate"], out creationDate))
    {
      context.CoreContext.IncrementCounter(
        "Median / Std. Dev. Mapper", "Invalid Creation Dates", 1);
      return;
    }

    var text = parsed["Text"];

    context.EmitKeyValue(
      creationDate.Hour.ToString(CultureInfo.InvariantCulture),
      text.Length.ToString(CultureInfo.InvariantCulture));
  }
}

MedianStdDevReducer.cs

public class MedianStdDevReducer : ReducerCombinerBase
{
  public override void Reduce(string key, 
    IEnumerable values, 
    ReducerCombinerContext context)
  {
    float sum = 0;
    int count = 0;

    var commentLengths = new List();

    foreach (var value in values.Select(float.Parse))
    {
      commentLengths.Add(value);
      sum += value;
      count++;
    }

    commentLengths.Sort((x, y) => x.CompareTo(y));

    // calculate median
    double median;
    if (count % 2 == 0)
    {
      // if commentLengths is an even value, average middle two elements
      median = (commentLengths[Convert.ToInt32(count / 2 - 1)]
        + commentLengths[Convert.ToInt32(count / 2)]) / 2.0f;
    } else {
      // else, set median to middle value
      median = commentLengths[Convert.ToInt32(count / 2)];
    }

    // calculate standard deviation
    var avg = sum / count;
    var sumOfSquares = commentLengths
      .Sum(commentLength => Math.Pow(commentLength - avg, 2));
    var stdDev = Math.Sqrt(sumOfSquares / (count - 1));

    context.EmitKeyValue(
      key,
      String.Format("{0}t{1}", median, stdDev));
  }
}