Fundamental Clustering Problems Suite

The package provides over sixty state-of-the-art clustering algorithms for unsupervised machine learning published in .

Table of contents

1. Description
2. Installation
3. Tutorial Examples
4. Manual
5. Use cases
6. Additional information
7. References

Description

The Fundamental Clustering Problems Suite (FCPS) summaries over sixty state-of-the-art clustering algorithms available in R language. An important advantage is that the input and output of clustering algorithms is simplified and consistent in order to enable users a swift execution of cluster analysis. By combining mirrored-density plots (MD plots) with statistical testing FCPS provides a tool to investigate the cluster tendency quickly prior to the cluster analysis itself . Common clustering challenges can be generated with arbitrary sample size . Additionally, FCPS sums 26 indicators with the goal to estimate the number of clusters up and provides an appropriate implementation of the clustering accuracy for more than two clusters . A subset of methods was used in a benchmarking of algorithms published in .

Installation

Installation using CRAN

Install automatically with all dependencies via

install.packages("FCPS",dependencies = T)

# Optionally, for the automatic installation
# of all suggested packages:
Suggested=c("kernlab", "cclust", "dbscan", "kohonen",
            "MCL", "ADPclust", "cluster", "DatabionicSwarm",
            "orclus", "subspace", "flexclust", "ABCanalysis",
            "apcluster", "pracma", "EMCluster", "pdfCluster", "parallelDist",
            "plotly", "ProjectionBasedClustering", "GeneralizedUmatrix",
            "mstknnclust", "densityClust", "parallel", "energy", "R.utils",
            "tclust", "Spectrum", "genie", "protoclust", "fastcluster", 
			"clusterability", "signal", "reshape2", "PPCI", "clustrd", "smacof",
			"rgl", "prclust", "dendextend",
            "moments", "prabclus", "VarSelLCM", "sparcl", "mixtools",
            "HDclassif", "clustvarsel", "knitr", "rmarkdown")

for(i in 1:length(Suggested)) {
  if (!requireNamespace(Suggested[i], quietly = TRUE)) {
    message(paste("Installing the package", Suggested[i]))
    install.packages(Suggested[i], dependencies = T)
  }
}

Installation using Github

Please note, that dependecies have to be installed manually.

remotes::install_github("Mthrun/FCPS")

Installation using R Studio

Please note, that dependecies have to be installed manually.

Tools -> Install Packages -> Repository (CRAN) -> FCPS

Tutorial Examples

The tutorial with several examples can be found on in the vignette on CRAN:

https://cran.r-project.org/web/packages/FCPS/vignettes/FCPS.html

Manual

The full manual for users or developers is available here: https://cran.r-project.org/web/packages/FCPS/FCPS.pdf

Use Cases

Cluster Analysis of High-dimensional Data

The package FCPS provides a clear and consistent access to state-of-the-art clustering algorithms:

library(FCPS)
data("Leukemia")
Data=Leukemia$Distance
Classification=Leukemia$Cls
ClusterNo=6
CA=ADPclustering(Leukemia$DistanceMatrix,ClusterNo)
Cls=ClusterRenameDescendingSize(CA$Cls)
ClusterPlotMDS(Data,Cls,main = ’Leukemia’,Plotter3D = ’plotly’)
ClusterAccuracy(Cls,Classification)
[1] 0.9963899

Generating Typical Challenges for Clustering Algorithms

Several clustering challenge can be generated with an arbitrary sample size:

set.seed(600)
library(FCPS)
DataList=ClusterChallenge("Chainlink", SampleSize = 750,
PlotIt=TRUE)
Data=DataList$Chainlink
Cls=DataList$Cls
> ClusterCount(Cls)
$CountPerCluster
$NumberOfClusters
$ClusterPercentages
[1] 377 373
[1] 2
[1] 50.26667 49.73333

Cluster-Tendency

For many applications, it is crucial to decide if a dataset possesses cluster structures:

library(FCPS)
set.seed(600)
DataList=ClusterChallenge("Chainlink",SampleSize = 750)
Data=DataList$Chainlink
Cls=DataList$Cls
library(ggplot2)
ClusterabilityMDplot(Data)+theme_bw()

Estimation of Number of Clusters

The “FCPS” package provides up to 26 indicators to determine the number of clusters:

library(FCPS)
set.seed(135)
DataList=ClusterChallenge("Chainlink",SampleSize = 900)
Data=DataList$Chainlink
Cls=DataList$Cls
Tree=HierarchicalClustering(Data,0,"SingleL")[[3]]
ClusterDendrogram(Tree,4,main="Single Linkage")
MaximumNumber=7
clsm <- matrix(data = 0, nrow = dim(Data)[1], ncol = MaximumNumber)
for (i in 2:(MaximumNumber+1)) {
clsm[,i-1] <- cutree(Tree,i)
}
out=ClusterNoEstimation(Data, ClsMatrix = clsm,
MaxClusterNo = MaximumNumber, PlotIt = TRUE)

Additional information

References

0. [Thrun/Stier, 2021] Thrun, M. C., & Stier, Q.: Fundamental Clustering Algorithms Suite SoftwareX, Vol. 13(C), pp. 100642. doi 10.1016/j.softx.2020.100642, 2021.
1. [Thrun, 2020] Thrun, M. C.: Improving the Sensitivity of Statistical Testing for Clusterability with Mirrored-Density Plot, in Archambault, D., Nabney, I. & Peltonen, J. (eds.), Machine Learning Methods in Visualisation for Big Data, DOI 10.2312/mlvis.20201102, The Eurographics Association, Norrköping , Sweden, May, 2020.
2. [Thrun/Ultsch, 2020a] Thrun, M. C., & Ultsch, A.: Clustering Benchmark Datasets Exploiting the Fundamental Clustering Problems, Data in Brief,Vol. 30(C), pp. 105501, DOI 10.1016/j.dib.2020.105501 , 2020.
3. [Thrun/Ultsch, 2021] Thrun, M. C., and Ultsch, A.: Swarm Intelligence for Self-Organized Clustering, Artificial Intelligence, Vol. 290, pp. 103237, \doi{10.1016/j.artint.2020.103237}, 2021.
4. [Thrun/Ultsch, 2020b] Thrun, M. C., & Ultsch, A. : Using Projection based Clustering to Find Distance and Density based Clusters in High-Dimensional Data, Journal of Classification, \doi{10.1007/s00357-020-09373-2}, Springer, 2020.