Version: 1.7.6
Date: 2022-09-20
Title: Somoclu
Imports: kohonen
Description: Somoclu is a massively parallel implementation of self-organizing maps. It exploits multicore CPUs and it can be accelerated by CUDA. The topology of the map can be planar or toroid and the grid of neurons can be rectangular or hexagonal . Details refer to (Peter Wittek, et al (2017)) <doi:10.18637/jss.v078.i09>.
URL: https://peterwittek.github.io/somoclu/
BugReports: https://github.com/peterwittek/somoclu/issues
License: GPL-3
Packaged: 2022-09-25 05:08:58 UTC; sgao
Author: Peter Wittek [aut], Shichao Gao [cre]
Maintainer: Shichao Gao <xgdgsc@gmail.com>
NeedsCompilation: yes
Repository: CRAN
SystemRequirements: C++11
LinkingTo: Rcpp
LazyData: true
Date/Publication: 2022-09-26 07:10:07 UTC

convert Somoclu train result to kohonen class for plotting

Description

A function call to convert Somoclu train result to kohonen class for plotting.

Usage

Rsomoclu.kohonen(input_data, result, n.hood = NULL, toroidal = FALSE)

Arguments

input_data

input data, matrix format

result

The result returned by Rsomoclu.train

n.hood

Same as in koohonen, the shape of the neighbourhood, either "circular" or "square". The latter is the default for rectangular maps, the former for hexagonal maps.

toroidal

if TRUE, the edges of the map are joined. Note that in a hexagonal toroidal map, the number of rows must be even.

Value

An object of class kohonen for plotting.

See Also

https://www.r-bloggers.com/2014/02/self-organising-maps-for-customer-segmentation-using-r/

Examples

library('Rsomoclu')
library('kohonen')
data("rgbs", package = "Rsomoclu")
input_data <- rgbs
input_data <- data.matrix(input_data)
nSomX <- 20
nSomY <- 20
nEpoch <- 10
radius0 <- 0
radiusN <- 0
radiusCooling <- "linear"
scale0 <- 0
scaleN <- 0.01
scaleCooling <- "linear"
kernelType <- 0
mapType <- "planar"
gridType <- "rectangular"
compactSupport <- FALSE
codebook <- NULL
neighborhood <- "gaussian"
stdCoeff <- 0.5
res <- Rsomoclu.train(input_data, nEpoch, nSomX, nSomY,
                      radius0, radiusN,
                      radiusCooling, scale0, scaleN,
                      scaleCooling,
                      kernelType, mapType, gridType, compactSupport, 
                      neighborhood, stdCoeff, codebook)
## Convert to kohonen object for plotting
sommap = Rsomoclu.kohonen(input_data, res)
## Show 'codebook'
plot(sommap, type="codes", main = "Codes")
## Show 'component planes'
plot(sommap, type = "property", property = sommap$codes[[1]][,1],
     main = colnames(sommap$codes)[1])
## Show 'U-Matrix'
plot(sommap, type="dist.neighbours")

Train function for Somoclu

Description

A function call to Somoclu to train the Self Organizing Map.

Usage

Rsomoclu.train(input_data, nEpoch, nSomX, nSomY,
                     radius0, radiusN,
                     radiusCooling, scale0, scaleN,
                     scaleCooling,
                     kernelType, mapType, gridType, compactSupport,
                     neighborhood, stdCoeff, codebook, vectDistance)

Arguments

input_data

input data, matrix format

nEpoch

Maximum number of epochs

nSomX

Number of columns in map (size of SOM in direction x)

nSomY

Number of rows in map (size of SOM in direction y)

radius0

Start radius (default: half of the map in direction min(x,y))

radiusN

End radius (default: 1)

radiusCooling

Radius cooling strategy: linear or exponential (default: linear)

scale0

Starting learning rate (default: 1.0)

scaleN

Finishing learning rate (default: 0.01)

scaleCooling

Learning rate cooling strategy: linear or exponential (default: linear)

kernelType

Kernel type 0: Dense CPU 1: Dense GPU 2: Sparse CPU (default: 0)

mapType

Map type: planar or toroid (default: "planar")

gridType

Grid type: square or hexagonal (default: "rectangular")

compactSupport

Compact support for Gaussian neighborhood, (default:TRUE)

neighborhood

Neighborhood function: gaussian or bubble (default: "gaussian")

codebook

initial codebook, (default:NULL)

stdCoeff

The coefficient in the Gaussian neighborhood function exp(-||x-y||^2/(2*(coeff*radius)^2)), (default:0.5)

vectDistance

the vector distance function "norm-3", "norm-6", "norm-2"(same as default) "norm-inf", is supported with kerneltype = 0 only , (default:euclidean)

Value

a list including elements

codebook

the codebook

globalBmus

global Best Matching Unit matrix

uMatrix

uMatrix

Author(s)

Peter Wittek, Shichao Gao

References

Peter Wittek, Shi Chao Gao, Ik Soo Lim, Li Zhao (2017). Somoclu: An Efficient Parallel Library for Self-Organizing Maps. Journal of Statistical Software, 78(9), 1-21. doi:10.18637/jss.v078.i09.

Examples

library('Rsomoclu')
data("rgbs", package = "Rsomoclu")
input_data <- rgbs
input_data <- data.matrix(input_data)
nSomX <- 10
nSomY <- 10
nEpoch <- 10
radius0 <- 0
radiusN <- 0
radiusCooling <- "linear"
scale0 <- 0
scaleN <- 0.01
scaleCooling <- "linear"
kernelType <- 0
mapType <- "planar"
gridType <- "rectangular"
compactSupport <- FALSE
codebook <- NULL
neighborhood <- "gaussian"
stdCoeff <- 0.5
vectDistance <- "euclidean"
res <- Rsomoclu.train(input_data, nEpoch, nSomX, nSomY,
                      radius0, radiusN,
                      radiusCooling, scale0, scaleN,
                      scaleCooling,
                      kernelType, mapType, gridType, compactSupport, neighborhood,
                      stdCoeff, codebook, vectDistance)
res$codebook
res$globalBmus
res$uMatrix
library('kohonen')
sommap = Rsomoclu.kohonen(input_data, res)

tiny rgbs data

Description

tiny rgbs data for testing

Usage

rgbs

Format

matrix in plain text form