https://github.com/nacnudus/changepoints

Explore changepoint detection on growing timeseries
https://github.com/nacnudus/changepoints

Last synced: 2 months ago
JSON representation

Explore changepoint detection on growing timeseries

• Host: GitHub
• URL: https://github.com/nacnudus/changepoints
• Owner: nacnudus
• Created: 2017-11-27T17:41:54.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2018-01-29T19:14:57.000Z (over 7 years ago)
• Last Synced: 2025-01-05T18:28:13.545Z (4 months ago)
• Language: R
• Size: 20.7 MB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Changepoints

Is automatic detection of changepoints in timeseries good enough to focus human

attention when reviewing large government reports that are released frequently?

![changepoints on a growing timeseries](https://github.com/nacnudus/changepoints/blob/master/slides/animation.gif)

This uses London Underground [performance data

almanac](https://tfl.gov.uk/corporate/publications-and-reports/underground-services-performance),

which contains dozens of performance metrics with monthly resolution over more

than ten years.  It's available in tidy form in the R package

[underground](https://github.com/nacnudus/underground).

Background changepoints blog

[post](https://www.r-bloggers.com/a-simple-intro-to-bayesian-change-point-analysis/)

with links to other materials

Method:

1. Apply various R changepoint detection packages

   ([changepoint](https://github.com/rkillick/changepoint/blob/master/NEWS),

   [ecp](https://cran.r-project.org/package=ecp),

   [bcp](https://cran.r-project.org/package=bcp) (Bayesian), Twitter's

   [BreakoutDetection](https://github.com/twitter/BreakoutDetection)) just to

   see what happens.

2. Evaluate the changepoints on shorter subsets of the series to see how stable

   they are from month to month.

3. Build a simple Twitter bot to tweet newly-detected changepoints as new data

   becomes available (proof of concept only, don't over-engineer this).

# `changepoint` package

A related Arxiv [paper](https://arxiv.org/pdf/1101.1438.pdf).

> As increasingly longer data sets are being collected, more and more

> applications require the detection of changes in the distributional properties

> of such data.

> ...

> In particular we focus on applications where we expect the number of

> changepoints to increase as we collect more data.

> ...

> At the time of writing Binary Segmentation proposed by Scott and Knott (1974)

> is arguably the most widely used changepoint search method. It is approximate

> in nature with an O(n log n) computational cost, where n is the number of data

> points.  While exact search algorithms exist for the most common forms of

> changepoint models, these have a much greater computational cost.

> ...

> We present a new approach to search for changepoints, which is exact and under

> mild conditions has a computational cost that is linear in the number of data

> points: the Pruned Exact Linear Time (PELT) method.

A [demo](https://rpubs.com/richkt/269908) on Rpubs.