Type: Package
Title: Clustering a Data Set using Multi-SOM Algorithm
Version: 1.3
Author: Sarra Chair and Malika Charrad
Maintainer: Sarra Chair <sarra.chair@gmail.com>
Description: Implements two versions of the algorithm namely: stochastic and batch. The package determines also the best number of clusters and offers to the user the best clustering scheme from different results.
License: GPL-2
Depends: R (≥ 3.1.3), class
Imports: kohonen
URL: https://sites.google.com/site/malikacharrad/research/multisom-package
NeedsCompilation: no
Packaged: 2017-05-23 12:04:33 UTC; chaira
Repository: CRAN
Date/Publication: 2017-05-23 17:28:23 UTC

Self-Organizing Map: Batch version

Description

This function implements the batch version of the kohonen algorithm

Usage

BatchSOM(data,grid = somgrid(),min.radius=0.0001,
         max.radius=0.002,maxit=1000,
         init=c("random","sample","linear"),
         radius.type=c("gaussian","bubble","cutgauss","ep"))

Arguments

data

data to be used

grid

a grid for the representatives.The numbers of nodes should be approximately equal to 5*sqrt(n), which n denotes the number of sample.

min.radius

the minimum neighbourhood radius

max.radius

the maximum neighbourhood radius

maxit

the maximum number of iterations to be done

init

the method to be used to initialize the prototypes.The following are permitted: "random" uses random draws from N(0,1); "sample" uses a radom sample from the data; "linear" uses the linear grids upon the first two principle components direction.See package som.

radius.type

the neighborhood function type. The following are permitted: "gaussian" "bubble" "cutgauss" "ep"

Value

classif

a vector of integer indicating to which unit each observation has been assigned

codes

a matrix of code vectors

grid

the grid, an object of class "somgrid"

Author(s)

Sarra Chair and Malika Charrad

References

Kohonen, T. (1995) Self-Organizing Maps. Springer-Verlag.

Brian Ripley, William Venables (2015), class: Functions for Classification, URL https://cran.r-project.org/package=class.

Jun Yan (2010), som: Self-Organizing Map, URL https://cran.r-project.org/package=som.

Examples

data<-iris[,-c(5)]
BatchSOM(data,grid = somgrid(7,7,"hexagonal"),min.radius=0.0001,
              max.radius=0.002,maxit=1000,"random","gaussian")

MultiSOM for batch version

Description

This function implements the batch version of MultiSOM algorithm.

Usage

multisom.batch(data= NULL,xheight,xwidth,topo=c("rectangular",
           "hexagonal"),min.radius,max.radius,maxit=1000,
           init=c("random","sample","linear"),radius.type=
           c("gaussian","bubble","cutgauss","ep"),index="all")

Arguments

data

data to be used

xheight

the x-dimension of the map

xwidth

the y-dimension of the map

topo

the topology used to build the grid.The following are permitted: "hexagonal" "rectangular"

min.radius

the minimum neighbourhood radius

max.radius

the maximum neighbourhood radius

maxit

the maximum number of iterations to be done

init

the method to be used to initialize the prototypes.The following are permitted: "random" uses random draws from N(0,1); "sample" uses a radom sample from the data; "linear" uses the linear grids upon the first two principle components direction.

radius.type

the neighborhood function type. The following are permitted: "gaussian" "bubble" "cutgauss" "ep"

index

vector of the index to be calculated. This should be one of : "db", "dunn", "silhouette", "ptbiserial", "ch", "cindex", "ratkowsky", "mcclain", "gamma", "gplus", "tau", "ccc", "scott", "marriot", "trcovw", "tracew", "friedman", "rubin", "ball", "sdbw", "dindex", "hubert", "sv", "xie-beni", "hartigan", "ssi", "xu", "rayturi", "pbm", "banfeld", "all" (all indices will be used)

Details

Index Optimal number of clusters
1. "db" or "all" Minimum value of the index
(Davies and Bouldin 1979)
2. "dunn" or "all" Maximum value of the index
(Dunn 1974)
3. "silhouette" or "all" Maximum value of the index
(Rousseeuw 1987)
4. "ptbiserial" or "all" Maximum value of the index
(Milligan 1980, 1981)
5. "ch" or "all" Maximum value of the index
(Calinski and Harabasz 1974)
6. "cindex" or "all" Minimum value of the index
(Hubert and Levin 1976)
7. "ratkowsky" or "all" Maximum value of the index
(Ratkowsky and Lance 1978)
8. "mcclain" or "all" Minimum value of the index
(McClain and Rao 1975)
9. "gamma" or "all" Maximum value of the index
(Baker and Hubert 1975)
10. "gplus" or "all" Minimum value of the index
(Rohlf 1974) (Milligan 1981)
11. "tau" or "all" Maximum value of the index
(Rohlf 1974) (Milligan 1981)
12. "ccc" or "all" Maximum value of the index
(Sarle 1983)
13. "scott" or "all" Max. difference between hierarchy
(Scott and Symons 1971) levels of the index
14. "marriot" or "all" Max. value of second differences
(Marriot 1971) between levels of the index
15. "trcovw" or "all" Max. difference between hierarchy
(Milligan and Cooper 1985) levels of the index
16. "tracew" or "all" Max. value of absolute second
(Milligan and Cooper 1985) differences between levels of the index
17. "friedman" or "all" Max. difference between hierarchy
(Friedman and Rubin 1967) levels of the index
18. "rubin" or "all" Min. value of second differences
(Friedman and Rubin 1967) between levels of the index
19. "ball" or "all" Max. difference between hierarchy
(Ball and Hall 1965) levels of the index
20. "sdbw" or "all" Minimum value of the index
(Halkidi and Vazirgiannis 2001)
21. "dindex" or "all" Graphical method
(Lebart et al. 2000)
22. "hubert" or "all" Graphical method
(Hubert and Arabie 1985)
23. "sv" or "all" Maximum value of the index
(Zalik and Zalik, 2011)
24. "xie-beni" or "all" Minimum value of the index
(Xie and Beni 1991)
25. "hartigan" or "all" Maximum difference between
(Hartigan 1975) hierarchy levels of the index
26. "ssi" or "all" Maximum value of the index
(Dolnicar,Grabler and Mazanec 1999)
27. "xu" or "all" Max. value of second differences
(Xu 1997) between levels of the index
28. "rayturi" or "all" Minimum value of the index
(Ray and Turi 1999)
29. "pbm" or "all" Maximum value of the index
(Bandyopadhyay,Pakhira and Maulik 2004)
30. "banfeld" or "all" Minimum value of the index
(Banield and Raftery 1974)

Value

All.index.by.layer

Values of indices for each layer

Best.nc

Best number of clusters proposed by each index and the corresponding index value.

Best.partition

Partition that corresponds to the best number of clusters

Author(s)

Sarra Chair and Malika Charrad

References

Charrad M., Ghazzali N., Boiteau V., Niknafs A. (2014). "NbClust: An R Package for Determining the Relevant Number of Clusters in a Data Set.", "Journal of Statistical Software, 61(6), 1-36.", "URL http://www.jstatsoft.org/v61/i06/".

Khanchouch, I., Charrad, M., & Limam, M. (2014). A Comparative Study of Multi-SOM Algorithms for Determining the Optimal Number of Clusters. Journal of Statistical Software, 61(6), 1-36.

Examples


## A 4-dimensional example

set.seed(1)

data<-rbind(matrix(rnorm(100,sd=0.3),ncol=2),
         matrix(rnorm(100,mean=2,sd=0.3),ncol=2),
         matrix(rnorm(100,mean=4,sd=0.3),ncol=2),
         matrix(rnorm(100,mean=8,sd=0.3),ncol=2))

res<- multisom.batch(data,xheight= 8, xwidth= 8,"hexagonal",
                min.radius=0.00010,max.radius=0.002,
                maxit=1000,"random","gaussian","ch")

res$All.index.by.layer
res$Best.nc
res$Best.partition

Multisom for stochastic version

Description

This function implements the stochastic version of MultiSOM algorithm.

Usage

multisom.stochastic(data = NULL, xheight = 7, xwidth = 7,
                  topo = c("rectangular", "hexagonal"),
                  neighbouhood.fct =c("bubble","gaussian"),
                  dist.fcts = NULL, rlen = 100,alpha = c(0.05, 0.01),
                  radius = c(2, 1.5, 1.2, 1), index = "all")

Arguments

data

the data matrix of observations

xheight

the x-dimension of the map

xwidth

the y-dimension of the map

topo

the topology used to build the grid.The following are permitted: "hexagonal" "rectangular"

neighbouhood.fct

the neighbouhood function type. The following are permitted: "gaussian" "bubble"

dist.fcts

The metric used to determine the distance function. Possible choices are: "sumofsquares" "euclidean" "manhattan" "tanimoto"

rlen

the maximum number of iterations to be done

alpha

learning rate, a vector of two numbers indicating the amount of change. Default is to decline linearly from 0.05 to 0.01 over rlen updates.

radius

the radius of the neighbourhood, either given as a single number or a vector (start, stop). If it is given as a single number the radius will run from the given number to the negative value of that number; as soon as the neighbourhood gets smaller than one only the winning unit will be updated.

index

vector of the index to be calculated. This should be one of : "db", "dunn", "silhouette", "ptbiserial", "ch", "cindex", "ratkowsky", "mcclain", "gamma", "gplus", "tau", "ccc", "scott", "marriot", "trcovw", "tracew", "friedman", "rubin", "ball", "sdbw", "dindex", "hubert", "sv", "xie-beni", "hartigan", "ssi", "xu", "rayturi", "pbm", "banfeld", "all" (all indices will be used)

Value

All.index.by.layer

Values of indices for each layer.

Best.nc

Best number of clusters proposed by each index and the corresponding index value.

Best.partition

Partition that corresponds to the best number of clusters

Author(s)

Sarra Chair and Malika Charrad

Examples

## A real data example

data<-as.matrix(iris[,-c(5)])

res<-multisom.stochastic(data, xheight = 8, xwidth = 8,"hexagonal","gaussian",
                    dist.fcts = NULL, rlen = 100,alpha = c(0.05, 0.01),
                    radius = c(2, 1.5, 1.2, 1),c("db","ratkowsky","dunn"))

res$All.index.by.layer
res$Best.nc