som: Creation of a som object

In harrysouthworth/kohonen: Self-organizing Maps for Data Classification

Description

Create and initialize a som object.

Usage

som(formula, data, grid = grid(9, 9, type = "hexagonal"), neighborhood = "gaussian", weights.min = -1, weights.max = 1)

Arguments

formula	a formula object, no variables may appear on the left (response) side.
data	a data frame in which to interpret the variables named in the formula.
grid	a grid specifying the structure of the projection space.
neighborhood	a character, possible values are 'uniform' and 'gaussian'.
weights.min	numeric value, minimal value to used during the creation of the initial weights.
weights.max	numeric value, maximal value to used during the creation of the initial weights.

Details

Scale data to mean = 0 and std = 1 is recommanded. If the model contains factors, they will be transformed with the contr.treatment function.

Value

object of class {som}: a list with several components:

data	model.matrix of the model
withinss	within sums of squares for each row
weights	weights of the network
grid	argument {grid}
size	vector of length {nrow(weights)}. The ith value gives the number of data points in cluster i.
neighborhood	argument {neighborhood}
groups	vector of length {nrow(data)}. The ith value gives the cluster of the data points i
metagroups	vector of length {nrow(data)}. The ith value gives the meta-cluster of the data points i

Note

side effects Meta-groups are all equals to 1, the same meta-group. Warning NA values are not allowed.

Author(s)

David Gohel

References

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

See Also

grid learn learnBatch plot.som summary.som predict.som biplot.som getWeights setWeights

Examples

library(MASS)
lcrabs <- log(crabs[, 4:8])

lcrabs.som <- som ( formula = ~ . , data = lcrabs
	, neighborhood = "uniform"
	, grid = grid ( xdim = 10 , ydim = 10 , type = "hexagonal" ) 
	, weights.min = min (lcrabs), weights.max = max (lcrabs)
	)
lcrabs.som

harrysouthworth/kohonen documentation built on May 17, 2019, 3:03 p.m.