SOM {class}R Documentation

Self-Organizing Maps: Online Algorithm


Kohonen's Self-Organizing Maps are a crude form of multidimensional scaling.


SOM(data, grid = somgrid(), rlen = 10000, alpha, radii, init)


data a matrix or data frame of observations, scaled so that Euclidean distance is appropriate.
grid A grid for the representatives: see somgrid.
rlen the number of updates: used only in the defaults for alpha and radii.
alpha the amount of change: one update is done for each element of alpha. Default is to decline linearly from 0.05 to 0 over rlen updates.
radii the radii of the neighbourhood to be used for each update: must be the same length as alpha. Default is to decline linearly from 4 to 1 over rlen updates.
init the initial representatives. If missing, chosen (without replacement) randomly from data.


alpha and radii can also be lists, in which case each component is used in turn, allowing two- or more phase training.


an object of class "SOM" with components

grid the grid, an object of class "somgrid".
codes a matrix of representatives.


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

Kohonen, T., Hynninen, J., Kangas, J. and Laaksonen, J. (1996) SOM PAK: The self-organizing map program package. Laboratory of Computer and Information Science, Helsinki University of Technology, Technical Report A31.

Ripley, B. D. (1996) Pattern Recognition and Neural Networks. Cambridge.

Venables, W. N. and Ripley, B. D. (2002) Modern Applied Statistics with S. Fourth edition. Springer.

See Also

somgrid, batchSOM


data(crabs, package = "MASS")

lcrabs <- log(crabs[, 4:8])
crabs.grp <- factor(c("B", "b", "O", "o")[rep(1:4, rep(50,4))])
gr <- somgrid(topo = "hexagonal")
crabs.som <- SOM(lcrabs, gr)

## 2-phase training
crabs.som2 <- SOM(lcrabs, gr,
    alpha = list(seq(0.05, 0, len = 1e4), seq(0.02, 0, len = 1e5)),
    radii = list(seq(8, 1, len = 1e4), seq(4, 1, len = 1e5)))

[Package class version 7.2-14 Index]