kmed (version 0.3.0)

rankkmed: Rank k-medoid algorithm

Description

This function runs the rank k-medoids algorithm proposed by Zadegan et. al. (2013).

Usage

rankkmed(distdata, ncluster, m = 3, iterate = 10, init = NULL)

Arguments

distdata

A distance matrix (n x n) or dist object.

ncluster

A number of clusters.

m

A number of objects to compute hostility (see Details).

iterate

A number of iterations for the clustering algorithm.

init

A vector of initial objects as the cluster medoids (see Details).

Value

Function returns a list of components:

cluster is the clustering memberships result.

medoid is the id medoids.

minimum_distance is the distance of all objects to their cluster medoid.

Details

This algorithm is claimed to cope with the local optima problem of the simple and fast-kmedoids algorithm (fastkmed). The m argument is defined by the user and has to be \(1 < m \leq n\). The m is a hostility measure computed by $$m_i = \sum_{X_j \in Y} r_{ij}$$ where \(x_j\) is the object j, Y is the set of objects as many as m, and \(r_{ij}\) is the rank distance, i.e. sorted distance, between object i and j.

init can be provided with a vector of id objects. The length of the vector has to be equal to the number of clusters. However, assigning a vector in the init argument, the algorithm is no longer the rank k-medoids algorithm.

References

Zadegan, S.M.R, Mirzaie M, and Sadoughi, F. 2013. Ranked k-medoids: A fast and accurate rank-based partitioning algorithm for clustering large datasets. Knowledge-Based Systems 39, 133-143.

Examples

# NOT RUN {
num <- as.matrix(iris[,1:4])
mrwdist <- distNumeric(num, num, method = "mrw")
result <- fastkmed(mrwdist, ncluster = 3, iterate = 50)
table(result$cluster, iris[,5])


# }