Package dev.frostlake.functions.aggregate

Class SquaredSumMoments

java.lang.Object
dev.frostlake.functions.aggregate.SquaredSumMoments
public final class SquaredSumMoments extends Object
The count, the sum and the sum of squares of a FLOAT series — the moments the account's variance family reads over FLOATs. VARIANCE is (n·Σx² − (Σx)²) / (n·(n − 1)) and VAR_POP the same numerator over n²; the STDDEV pair takes their roots. Each sum is added the way SUM adds a FLOAT column (see CompensatedSum) and read back with its last compensation applied.

Live-verified to the thirty-seventh decimal on every series measured — GENERATOR rows, UNION ALL branches, VALUES lists, a GROUP BY group and a window alike: 

   ten copies of 0.1        VARIANCE 0.0000000000000000024671622769447924062  (a running mean gives 0)
   1e8, 1e8 + 1, 1e8 + 2    VARIANCE 0                                        (the squares cancel)
   0.1, 0.2                 VARIANCE 0.0049999999999999975019981945933977840
   0.3, 0.7                 VARIANCE 0.0799999999999998490096686509787105024
   0.7, 0.3                 VARIANCE 0.0799999999999999600319711134943645447  (the order is read)

The sum of squares is the account's own SUM(x * x): over 0.3 then 0.7 that reads 0.5799999999999998490096686509787105024, and the variance above is exactly it less 0.5. What this does not model is the partitioning of a stored table: ten copies of 0.1 read back from one give twice the GENERATOR's variance, so a table's rows are not always met in the order they arrive here.

  • Constructor Details

    • SquaredSumMoments

      public SquaredSumMoments()

  • Method Details

    • add

      public void add(double value)
      Fold one value in.
      Parameters:
      value - the next value, in input order

    • merge

      public void merge(SquaredSumMoments other)
      Fold another series' moments in, its two sums added as terms.
      Parameters:
      other - the moments to absorb

    • count

      public long count()
      How many values were folded in.
      Returns:
      the count

    • sampleVariance

      public double sampleVariance()
      The sample variance (denominator n − 1); the caller checks there are two rows at least.
      Returns:
      the variance, never negative

    • populationVariance

      public double populationVariance()
      The population variance (denominator n); the caller checks there is a row at least.
      Returns:
      the variance, never negative

    • reset

      public void reset()
      Forget every value.