graphseg: segmentation of graph-based signals

What this package does

Performs estimation of a signal with regularization using the adjacency structure. This package has a natural application to spatial data: when we want to have a spatial segmentation of a region-dependent signal, you can apply this package to the adjacency graph of the spatial regions. More information in the associated paper.

This package uses a penalty similar to the graph fused lasso (Hoefling, 2010), except that the penalty on differences between adjacent vertices is given by the adaptive ridge (Liu and Li 2016; Frommlet and Nuel, 2016).


Here is a small example of segmentation of obesity prevalence on spatial data from the region of Utrecht, Netherlands (van de Kassteele et al, 2017). Each administrative region is a vertex of the graph and the edges are given by the presence of a shared border between two regions. This is illustrated by the figure:

Graphical abstract


# install.packages("remotes")

Bug report

If you have a problem or suggestion of improvement, please raise an issue.


This package is released under the GPLv3 License: see the LICENSE file or the online text. In short, you can use, modify, and distribute (including for commerical use) this package, with the notable obligations to use the GPLv3 license for your work and to provide a copy of the present source code.