Module org.elasticsearch.tdigest
Package org.elasticsearch.tdigest

Class MergingDigest

java.lang.Object
org.elasticsearch.tdigest.TDigest
org.elasticsearch.tdigest.AbstractTDigest
org.elasticsearch.tdigest.MergingDigest
public class MergingDigest extends AbstractTDigest
Maintains a t-digest by collecting new points in a buffer that is then sorted occasionally and merged into a sorted array that contains previously computed centroids.

This can be very fast because the cost of sorting and merging is amortized over several insertion. If we keep N centroids total and have the input array is k long, then the amortized cost is something like

N/k + log k

These costs even out when N/k = log k. Balancing costs is often a good place to start in optimizing an algorithm. For different values of compression factor, the following table shows estimated asymptotic values of N and suggested values of k:

CompressionNk
507825
10015742
20031473
 Sizing considerations for t-digest

The virtues of this kind of t-digest implementation include:

  • No allocation is required after initialization
  • The data structure automatically compresses existing centroids when possible
  • No Java object overhead is incurred for centroids since data is kept in primitive arrays

The current implementation takes the liberty of using ping-pong buffers for implementing the merge resulting in a substantial memory penalty, but the complexity of an in place merge was not considered as worthwhile since even with the overhead, the memory cost is less than 40 bytes per centroid which is much less than half what the AVLTreeDigest uses and no dynamic allocation is required at all.

  • Field Details

    • useAlternatingSort

      public boolean useAlternatingSort

    • useTwoLevelCompression

      public boolean useTwoLevelCompression

    • useWeightLimit

      public static boolean useWeightLimit

  • Constructor Details

    • MergingDigest

      public MergingDigest(double compression)
      Allocates a buffer merging t-digest. This is the normally used constructor that allocates default sized internal arrays. Other versions are available, but should only be used for special cases.
      Parameters:
      compression - The compression factor

    • MergingDigest

      public MergingDigest(double compression, int bufferSize)
      If you know the size of the temporary buffer for incoming points, you can use this entry point.
      Parameters:
      compression - Compression factor for t-digest. Same as 1/\delta in the paper.
      bufferSize - How many samples to retain before merging.

    • MergingDigest

      public MergingDigest(double compression, int bufferSize, int size)
      Fully specified constructor. Normally only used for deserializing a buffer t-digest.
      Parameters:
      compression - Compression factor
      bufferSize - Number of temporary centroids
      size - Size of main buffer

  • Method Details

    • add

      public void add(double x, long w)
      Description copied from class: TDigest
      Adds a sample to a histogram.
      Specified by:
      add in class TDigest
      Parameters:
      x - The value to add.
      w - The weight of this point.

    • compress

      public void compress()
      Merges any pending inputs and compresses the data down to the public setting. Note that this typically loses a bit of precision and thus isn't a thing to be doing all the time. It is best done only when we want to show results to the outside world.
      Specified by:
      compress in class TDigest

    • size

      public long size()
      Description copied from class: TDigest
      Returns the number of points that have been added to this TDigest.
      Specified by:
      size in class TDigest
      Returns:
      The sum of the weights on all centroids.

    • cdf

      public double cdf(double x)
      Description copied from class: TDigest
      Returns the fraction of all points added which are ≤ x. Points that are exactly equal get half credit (i.e. we use the mid-point rule)
      Specified by:
      cdf in class TDigest
      Parameters:
      x - The cutoff for the cdf.
      Returns:
      The fraction of all data which is less or equal to x.

    • quantile

      public double quantile(double q)
      Description copied from class: TDigest
      Returns an estimate of a cutoff such that a specified fraction of the data added to this TDigest would be less than or equal to the cutoff.
      Specified by:
      quantile in class TDigest
      Parameters:
      q - The desired fraction
      Returns:
      The smallest value x such that cdf(x) ≥ q

    • centroidCount

      public int centroidCount()
      Specified by:
      centroidCount in class TDigest

    • centroids

      public Collection<Centroid> centroids()
      Description copied from class: TDigest
      A Collection that lets you go through the centroids in ascending order by mean. Centroids returned will not be re-used, but may or may not share storage with this TDigest.
      Specified by:
      centroids in class TDigest
      Returns:
      The centroids in the form of a Collection.

    • compression

      public double compression()
      Description copied from class: TDigest
      Returns the current compression factor.
      Specified by:
      compression in class TDigest
      Returns:
      The compression factor originally used to set up the TDigest.

    • getScaleFunction

      public ScaleFunction getScaleFunction()

    • setScaleFunction

      public void setScaleFunction(ScaleFunction scaleFunction)
      Overrides:
      setScaleFunction in class TDigest

    • byteSize

      public int byteSize()
      Description copied from class: TDigest
      Returns the number of bytes required to encode this TDigest using #asBytes().
      Specified by:
      byteSize in class TDigest
      Returns:
      The number of bytes required.

    • toString

      public String toString()
      Overrides:
      toString in class Object