an ensemble of partitions or hierarchies, or something
coercible to that (see cl_ensemble).
k
an integer giving the number of classes to be used in the
partition.
method
a character string or a function, as for argument
method of function cl_dissimilarity.
solver
a character string indicating the $k$-medoids solver
to be employed. May be abbreviated. If "pam" (default), the
Partitioning Around Medoids (Kaufman & Rousseeuw (1990), Chapter 2)
heuristic
Value
An object of class "cl_pam" representing the obtained
secondary partition, which is a list with the following
components.
clusterthe class ids of the partition.
medoid_idsthe indices of the medoids.
prototypesa cluster ensemble with the $k$ prototypes
(medoids).
criterionthe value of the criterion function of the
partition.
descriptiona character string indicating the dissimilarity
method employed.
Details
An optimal $k$-medoids partition of the given cluster ensemble is
defined as a partition of the objects $x_i$ (the elements of the
ensemble) into $k$ classes $C_1, \ldots, C_k$ such that the
criterion function
$L = \sum_{l=1}^k \min_{j \in C_l} \sum_{i \in C_l} d(x_i, x_j)$
is minimized.
Such secondary partitions (e.g., Gordon & Vichi, 1998) are obtained by
computing the dissimilarities $d$ of the objects in the ensemble
for the given dissimilarity method, and applying a dissimilarity-based
$k$-medoids solver to $d$.
References
L. Kaufman and P. J. Rousseeuw (1990).
Finding Groups in Data: An Introduction to Cluster Analysis.
Wiley, New York.
A. D. Gordon and M. Vichi (1998).
Partitions of partitions.
Journal of Classification, 15, 265--285.
See Also
cl_pclust for more general prototype-based partitions of
clusterings.
data("Kinship82")
party <- cl_pam(Kinship82, 3, "symdiff")
## Compare results with tables 5 and 6 in Gordon & Vichi (1998).party
lapply(cl_prototypes(party), cl_classes)
table(cl_class_ids(party))