Streaming approximate histograms in Go

build status

This package provides Streaming Approximate Histograms for efficient quantile approximations.

The histograms in this package are based on the algorithms found in Ben-Haim & Yom-Tov’s A Streaming Parallel Decision Tree Algorithm (PDF). Histogram bins do not have a preset size. As values stream into the histogram, bins are dynamically added and merged.

Another implementation can be found in the Apache Hive project (see NumericHistogram).

An example:


The accurate method of calculating quantiles (like percentiles) requires data to be sorted. Streaming histograms make it possible to approximate quantiles without sorting (or even individually storing) values.

NumericHistogram is the more basic implementation of a streaming histogram. WeightedHistogram implements bin values as exponentially-weighted moving averages.

A maximum bin size is passed as an argument to the constructor methods. A larger bin size yields more accurate approximations at the cost of increased memory utilization and performance.

A picture of kittens:

stack of kittens

Getting started

Using in your own code

$ go get
import ""

Running tests and making modifications

Get the code into your workspace:

$ cd $GOPATH
$ git clone [email protected]:VividCortex/gohistogram.git ./src/

You can run the tests now:

$ cd src/
$ go test .

API Documentation

Full source documentation can be found here.


We only accept pull requests for minor fixes or improvements. This includes:

  • Small bug fixes
  • Typos
  • Documentation or comments

Please open issues to discuss new features. Pull requests for new features will be rejected, so we recommend forking the repository and making changes in your fork for your use case.


Copyright (c) 2013 VividCortex

Released under MIT License. Check LICENSE file for details.


Airflow job is started before scheduled time

Consider every possible pair, remove NaN entries and obtain the linear regression