Numeric.StreamingHistogram

A very inefficient implementation of lossy online histograms based on the following paper:

• A Streaming Parallel Decision Tree Algorithm by Ben-Haim and Tom-Tov: http://www.jmlr.org/papers/volume11/ben-haim10a/ben-haim10a.pdf

NB This library is very experimental and inefficient. Numerical stability has not been tested (yet).

Documentation

empty

Arguments

:: Int

Maximum number of values to keep

-> Hist

The empty histogram

Creates an empty histogram to be updated with update

update

Arguments

:: Double

Value to add

-> Hist

Lossy histogram to update

-> Hist

Histogram with the value incorporated

Adds a single value to the histogram.

data Hist where

Constructors

Hist :: Int -> Map Double Int -> Hist

Fields

nBins :: Int

histData :: Map Double Int

Instances

Eq Hist

Show Hist

merge

Arguments

:: Hist

Histogram a

-> Hist

Histogram b

-> Maybe Hist

Returns Nothing if a and b disagree on how many values to keep.

Merges two histograms. Returns Nothing if they don't have the same number of values. TODO: Use more efficient algorithm.

sum

Arguments

:: Double

The cutoff value b

-> Hist

A histogram

-> Double

Estimated number of points

Estimates how many points lie in the interval [-infinity, b].