anticlust - Subset Partitioning via Anticlustering
The method of anticlustering partitions a pool of elements
into groups (i.e., anticlusters) with the goal of maximizing
between-group similarity or within-group heterogeneity. The
anticlustering approach thereby reverses the logic of cluster
analysis that strives for high within-group homogeneity and
clear separation between groups. Computationally,
anticlustering is accomplished by maximizing instead of
minimizing a clustering objective function, such as the
intra-cluster variance (used in k-means clustering) or the sum
of pairwise distances within clusters. The main function
anticlustering() gives access to optimal and heuristic
anticlustering methods described in Papenberg and Klau (2021;
<doi:10.1037/met0000301>), Brusco et al. (2020;
<doi:10.1111/bmsp.12186>), Papenberg (2024;
<doi:10.1111/bmsp.12315>), Papenberg, Wang, et al. (2025;
<doi:10.1016/j.crmeth.2025.101137>), Papenberg, Breuer, et al.
(2025; <doi:10.1017/psy.2025.10052>), and Yang et al. (2022;
<doi:10.1016/j.ejor.2022.02.003>). The optimal algorithms
require that an integer linear programming solver is installed.
This package will install 'lpSolve'
(<https://cran.r-project.org/package=lpSolve>) as a default
solver, but it is also possible to use the package 'Rglpk'
(<https://cran.r-project.org/package=Rglpk>), which requires
the GNU linear programming kit
(<https://www.gnu.org/software/glpk/glpk.html>), the package
'Rsymphony' (<https://cran.r-project.org/package=Rsymphony>),
which requires the SYMPHONY ILP solver
(<https://github.com/coin-or/SYMPHONY>), or the commercial
solver Gurobi, which provides its own R package that is not
available via CRAN (<https://www.gurobi.com/downloads/>).
'Rglpk', 'Rsymphony', 'gurobi' and their system dependencies
have to be manually installed by the user because they are only
suggested dependencies. Full access to the bicriterion
anticlustering method proposed by Brusco et al. (2020) is given
via the function bicriterion_anticlustering(), while
kplus_anticlustering() implements the full functionality of the
k-plus anticlustering approach proposed by Papenberg (2024).
Some other functions are available to solve classical
clustering problems. The function balanced_clustering() applies
a cluster analysis under size constraints, i.e., creates
equal-sized clusters. The function matching() can be used for
(unrestricted, bipartite, or K-partite) matching. The function
wce() can be used optimally solve the (weighted) cluster
editing problem, also known as correlation clustering, clique
partitioning problem or transitivity clustering.
