Segment Data With Maximum Likelihood

Given a likelihood provided by the user, this package applies it to a given matrix dataset in order to find change points in the data that maximize the sum of the likelihoods of all the segments. This package provides a handful of algorithms with different time complexities and assumption compromises so the user is able to choose the best one for the problem at hand. The implementation of the segmentation algorithms in this package are based on the paper by Bruno M. de Castro, Florencia Leonardi (2018) . The Berlin weather sample dataset was provided by Deutscher Wetterdienst < https://dwd.de/>. You can find all the references in the Acknowledgments section of this package's repository via the URL below.

segmentr

Given a likelihood provided by the user, this package applies it to a given matrix dataset in order to find change points in the data that maximize the sum of the likelihoods of all the segments.

This package provides a handful of algorithms with different time complexities and assumption compromises so the user is able to choose the best one for the problem at hand.

Installation

Install the package from CRAN:

install.packages("segmentr")

Example

Sample code using the package to identify change points in the segments’ averages:

require(segmentr)
#> Loading required package: segmentr

make_segment <- function(n, p) matrix(rbinom(100 * n, 1, p), nrow = 100)
data <- cbind(make_segment(5, 0.1), make_segment(10, 0.9), make_segment(2, 0.1))
mean_lik <- function(X) abs(mean(X) - 0.5) * ncol(X)^2
segment(data, likelihood = mean_lik, algorithm = "hieralg")
#> Segments (total of 3):
#> 
#> 1:5
#> 6:15
#> 16:17

For an in depth step-by-step, please check vignette("segmentr").

Acknowledgments

This package is part of a Master’s degree research thesis at IME-USP, with Florencia Leonardi as thesis adviser.

The algorithms in this package are based on a paper by Bruno M. de Castro and Florencia Leonardi.

The berlin sample dataset was provided by © Deutscher Wetterdienst and put together with the rdwd package by Berry Boessenkool. Check make_berlin.R for the script that builds the dataset.