#cvxclustr
Clustering is a fundamental problem in science and engineering. Many classic methods such as
* Alternating Method of Multipliers (ADMM)
* Alternating Minimization Algorithm (AMA)
for solving this convex formulation of the clustering problem. We seek the centroids
$\frac{1}{2} \sum_i || x_i - u_i||2^2 + \gamma \sum_l w{l} ||u_{l1} - u_{l2} ||$
Two penalty norms are currently supported: 1-norm and 2-norm.
You can install the stable version on [CRAN:] (http://cran.r-project.org/web/packages/cvxclustr/)
install.packages('cvxclustr', dependencies = TRUE)
The developmental version can be pulled from github using the R package [devtools] (http://cran.r-project.org/web/packages/devtools/index.html)
install_github('cvxclustr','echi')
or from the command line
git clone https://github.com/echi/cvxclustr.git
R CMD build cvxclustr
R CMD INSTALL cvxclustr_*.tar.gz
Details on the algorithms are in the paper [Splitting Methods for Convex Clustering] (http://arxiv.org/abs/1304.0499) by Chi and Lange.
Previous takes on this formulation of clustering:
Just Relax and Come Clustering! A Convexification of k-Means Clustering by Lindsten, Ohlsson, and Ljung.
Clusterpath: An Algorithm for Clustering using Convex Fusion Penalties by Hocking, Joulin, Bach, and Vert.