Change Point Detection Toolbox

Change point detection algorithms for segmenting time series data.

Overview

Segmentation performance of eight different algorithms is tested in this repo.

Dependencies

HRVToolbox for real data analysis part

Configuration

• Install R Studio to implement BCP method. Add 'bcp' package which is located here: https://cran.r-project.org/web/packages/bcp/index.html

Add also stingr package: install_package("stringr")

• Install NAG MATLAB toolbox from following address: https://www.nag.com/content/downloads-nag-toolbox-matlab-versions

• The executable "rrgenV3" is compiled for macOS. For other operating systems, follow these steps:

1. cd into rrGenerator folder

2. Delete rrgenV3 since this executable is for mac

3. Compile rrgenV3.c into executable:

   gcc -Wall rrgenV3.c -lm -o rrgenV3

4. Test the executable by calling rrgenV3 from the terminal, using the number 1 as a seed, and setting max elements to 200:
   ./rrgenV3 1 200 % For mac computer
   :rrgenV3 1 200 % For windows computer
   

You should see output printed to the terminal.

Algorithms

• Modified Bayesian Online Changepoint Detection (Modified BOCD): bocpd_revised.m
• Original Bayesian Online Changepoint Detection (Original BOCD): bocpd_original.m
• Bayesian Blocks (BBLOCKS): bblocks.m
• Pruned Exact Linear Time (PELT1) from NAG MATLAB toolbox
• Matlab function (PELT2) 'findchangepts'
• Binary Segmentation (BiS) from NAG MATLAB toolbox
• Recursive Mean Difference Maximization (RMDM) : rmdm.m Includes 2 versions, rmdm.m_v1 is written by Maxim Osipov and rmdm.m is the modified version.
• Bayesian Analysis of Changepoint Problems (BCP) : bcp_testRRGEN.R

Real data analysis

Physionet capslpdb database is located here: https://www.physionet.org/pn6/capslpdb/

• Download hypnograms from above address.

• Download EKG files in csv format.
  ex: rdsamp -r capslpdb/n5.edf -H -f 0 -ps -c >n5.csv

• Arrange files in same format as CPD/Toolbox/testData/physionet_capslpdb. EKG file and hypnogram of same subject should be named the same. ex.: n1.txt

Possible errors & how to troubleshoot them

• /bin/bash: Rscript: command not found
  Add where R is located to your MATLAB environment
  setenv('PATH', [getenv('PATH'),':','/usr/local/bin']);

• In artificial data analysis, no data is generated Make sure your rrgenV3 executable is suitable for your computer type. If not, delete this and re compile as decribed above.

Useful Links

Related to setting up parameters of NAG functions:
PELT: http://www.nag.com/numeric/mb/nagdoc_mb/manual_25_1/html/g13/g13naf.html
BiS: http://www.nag.com/numeric/mb/nagdoc_mb/manual_25_1/html/g13/g13ndf.html