do.similarity.projection: Function that computes sets of similarity indices using...

In mosclust: Model Order Selection for Clustering

Function that computes sets of similarity indices using randomized maps.

Description

This function may use different clustering algorithms and different similarity measures to compute similarity indices. Random projections techniques are applied to perturb the data. More precisely pairs of data sets are projected into lower dimensional subspaces using randomized maps, and then are clustered and the resulting clusterings are compared using similarity indices between pairs of clusterings (e.g. Rand Index, Jaccard or Fowlkes and Mallows indices). These indices are computed multiple times for different number of clusters.

Usage

do.similarity.projection(X, c = 2, nprojections = 100, dim = 2, pmethod = "PMO", 
scale = TRUE, seed = 100, s = sFM, alg.clust.sim = Hierarchical.sim.projection, 
distance = "euclidean", hmethod = "ward.D")

Arguments

X	matrix of data (variables are rows, examples columns)
c	if it is a vector of length 1, number of clusters from 2 to c are considered; otherwise are considered the number of clusters stored in the vector c.
nprojections	number of pairs of projected data
dim	dimension of the projected data
pmethod	pmethod : projection method. It must be one of the following: "RS" (random subspace projection) "PMO" (Plus Minus One random projection) (default) "Norm" (normal random projection) "Achlioptas" (Achlioptas random projection)
scale	if TRUE randomized projections are scaled (default)
seed	numerical seed for the random generator
s	similarity function to be used. It may be one of the following: - sFM (Fowlkes and Mallows) - sJaccard (Jaccard) - sM (matching coefficient) (default Fowlkes and Mallows)
alg.clust.sim	method that computes the similarity indices using subsampling techniques and a specific clustering algorithm. It may be one of the following: - Hierarchical.sim.resampling (hierarchical clustering algorithm, default) - Kmeans.sim.resampling (c - mean algorithm) - PAM.sim.resampling (Prediction Around Medoid algorithm) - Fuzzy.kmeans.sim.resampling (Fuzzy c-mean)
distance	it must be one of the two: "euclidean" (default) or "pearson" (that is 1 - Pearson correlation)
hmethod	the agglomeration method to be used. This parameter is used only by the hierarchical clustering algorithm. This should be one of the following: "ward.D", "single", "complete", "average", "mcquitty", "median" or "centroid", according of the hclust method of the package stats.

Value

a matrix that stores the similarity between pairs of clustering across multiple number of clusters and multiple clusterings performed on subsamples of the original data. Number of rows equal to the length of c (number of clusters); number of columns equal to nsub, that is the number of subsamples considered for each number of clusters.

Author(s)

Giorgio Valentini [e-mail]

References

A.Bertoni, G. Valentini, Randomized maps for assessing the reliability of patients clusters in DNA microarray data analyses, Artificial Intelligence in Medicine 37(2):85-109 2006

See Also

do.similarity.resampling, do.similarity.noise

Examples

library("clusterv")
# Data set generation
M <- generate.sample6 (n=20, m=10, dim=600, d=3, s=0.2);
# computing similarity indices with the fuzzy c-mean algorithm
Sp.fuzzy.kmeans.sample6 <- do.similarity.projection(M, c=8, nprojections=30, 
   dim=JL.predict.dim(120,0.2), pmethod="PMO", alg.clust.sim=Fuzzy.kmeans.sim.projection);
# computing similarity indices using the c-mean algorithm
Sp.kmeans.sample6 <- do.similarity.projection(M, c=8, nprojections=30, 
   dim=JL.predict.dim(120,0.2), pmethod="PMO", alg.clust.sim=Kmeans.sim.projection);
# computing similarity indices using the hierarchical clustering algorithm
Sp.HC.sample6 <- do.similarity.projection(M, c=8, nprojections=30, 
   dim=JL.predict.dim(120,0.2), pmethod="PMO", alg.clust.sim=Hierarchical.sim.projection);

mosclust documentation built on June 8, 2025, 11:23 a.m.