README

The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.

Locating distributional changes in piece-wise stationary time series with long-range dependencies

Implementations of the change-point estimation algorithms proposed in the following two papers.

Python

Installing

The Python package is precompiled for Windows (32/64 bit), MacOS (64 bit) and many Linux (32/64 bit) distributions. If you want to use the code on system for which there are no precompiled binaries like Raspberry Pis or Android phones head to the Compiling section in the Appendix.

How to use

To find changepoint positions in a sequence (given as either a python list or a numpy array), with a minimum distance of 0.03 * len(sequence) between changepoints which have been generated by 2 different processes simply call:

import PyChest
estimates = PyChest.find_changepoints(sequence, 0.03, 2)

estimates will now be a python list containing a list of changepoints in increasing order. The list can be empty.

If you do not know the number of generating processes you can call the list_estimator instead like following

import PyChest
estimates = PyChest.list_estimator(sequence, 0.03)

Now estimates is a list of changepoint estimates at least 0.03 * len(sequence) apart in decreasing order of a “score” associated to each estimate by the algorithm, which reflects the quality of the estimate; the list can be empty.

Installing

The R implementation uses the Python package which is precompiled for Windows (32/64 bit), MacOS (64 bit) and many Linux (32/64 bit) distributions. If you want to use the code on system for which there are no precompiled binaries like Raspberry Pis or Android phones head to the Compiling section in the Appendix.

You will need to have Python installed on your system in order to run the R-package.

devtools::install_github("azalk/GoChest")
library("RChest")
install_PyChest()

install_PyChest is used to install PyChest onto your local Python environment, call it to update the PyChest version R uses. You should not need to do this every time you use the R-package.

How to use

To find changepoint positions in a sequence (given as either a Matrix/Array or a Multi-element vector), with a minimum distance of 0.03 * length(sequence) between changepoints which have been generated by 2 different processes simply call:

library("RChest")
estimates <- find_changepoints(sequence, 0.03, 2L)

Now estimates will now be a Multi-element vector containing a list of changepoints at least 0.03 * length(sequence) apart in increasing order. The list can be empty.

If you do not know the number of generating processes you can call the list_estimator instead like following

library("RChest")
estimates <- list_estimator(sequence, 0.03)

Now estimates is a list of changepoint estimates at least 0.03 * length(sequence) apart in decreasing order of a “score” associated to each estimate by the algorithm, which reflects the quality of the estimate; the list can be empty.

Installing

Go will now complain that the directory layout is unexpected. That is because of the duality of it being a Python and Go package and can safely be ignored.

How to use

To find changepoint positions in a sequence (given as a slice), with a minimum distance of 0.03 * len(sequence) between changepoints which have been generated by 2 different processes simply call:

import (
    "github.com/azalk/GoChest/GoChest"
)
func main() {
    sequence := make([]float64, 0)

    // fill sequence somehow

    estimates = GoChest.FindChangepoints(sequence, 0.03, 2)
}

estimates will now be a slice containing a list of changepoints at least 0.03 * len(sequence) apart in increasing order. The slice can be empty.

If you do not know the number of generating processes you can call the ListEstimator instead like following

import (
    "github.com/azalk/GoChest/GoChest"
)

func main() {
    sequence := make([]float64, 0)

    // fill sequence somehow

    estimates = GoChest.ListEstimator(sequence, 0.03)
}

How to cite

Appendix

Compiling the Go code

You should not need to compile the Go code yourself if you just want to use the package on Windows, MacOS or a common Unix distribution, you can find instructions how to install the code here: Python.

If you want to compile the Go code yourself you need to install Go first, then clone and enter the repository by typing the following code into a command line:

Next you will need to actually compile the code, for Windows enter the following line in the console to do that

Your compiled c-shared library is now situated in the PyChestBuild folder. To complete the installation with simply type:

The package will now prefer your compiled library over the precompiled libraries.

Compiling for R

If you want to use the R-package with your own compiled binary, first install the R-Package by typing the following into an R-command-line:

devtools::install_github("azalk/GoChest")

Next, compile the Go code by following the Compiling the Go code chapter. Finally, install the Python package into the virtual environment used by reticulate.

Do not run the init_RChest command as that might override your self-compiled binary, if the Python or Go code updates you will need to recompile the code.

Cross Compiling

Cross compiling, for example to make the package available on Android phones, is a little more complicated. In order to build the code you will need to enable cgo specifically and provide a c-compiler for your target platform. Your final command should look something like this:

Which would compile for windows/amd64 architecture given that the CC=ValidCCompilerForThePlatform is replaced by an actual c-compiler for windows/amd64.

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.