kohonen (version 1.1.1)

bdk: Supervised version of Kohonen's self-organising maps

Description

Supervised version of self-organising maps for mapping high-dimensional spectra or patterns to 2D: the Bi-Directional Kohonen map. This is an alternating training of the X-space and the Y-space of the map, where for updating the X-space more weight is given to the features in Y, and vice versa. Weights start by default with values of (0.75, 0.25) and during training go to (0.5, 0.5). Prediction is done only using the X-space.

Usage

bdk(data, Y, grid=somgrid(), rlen = 100, alpha = c(0.05, 0.01),
    radius = quantile(nhbrdist, 0.67), xweight = 0.75,
    toroidal = FALSE, keep.data = TRUE)

Arguments

data

a matrix, with each row representing an object.

Y

property that is to be modelled. In case of classification, Y is a matrix of zeros, with exactly one '1' in each row indicating the class. For prediction of continuous properties, Y is a vector. A combination is possible, too, but one then sho

grid

a grid for the representatives: see somgrid.

rlen

the number of times the complete data set will be presented to the network.

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 starting radius of the neighbourhood to be used for each update: the decrease is linear over rlen updates and reaches the value of 1 after one-third of the iterations. After that, only the winning units are updated.

xweight

the initial weight given to the X map in the calculation of distances for updating Y, and to the Y map for updating X. This will linearly go to 0.5 during training. Defaults to 0.75.

toroidal

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

keep.data

save data in return value.

Value

• an object of class "kohonen" with components
• grid
  the grid, an object of class "somgrid".
• changes
  matrix containing two columns of mean average deviations from code vectors. Column 1 contains deviations used for updating Y; column 2 for updating X.
• codes
  matrix of code vectors.
• codeYs
  matrix of Y values associated with each unit.
• toroidal
  whether a toroidal map is used.
• data
  data matrix, only if keep.data is TRUE.
• Y
  Y, only if keep.data is TRUE.
• type
  the type of som, here "bdk"

References

W.J. Melssen, R. Wehrens, and L.M.C. Buydens. Chemom. Intell. Lab. Syst., 83, 99-113 (2006).

See Also

som, xyf, plot.kohonen, predict.kohonen

Examples

### Simulated example
library(MASS)

csize <- 15
c1 <- mvrnorm(csize, mu=c(5,3,4),
              Sigma=matrix(c(2, 0,0,0, 2, 1, 0, 1, 2),3,3))
c2 <- mvrnorm(csize, mu=c(5.5, 3.5, 4.5),
              Sigma=matrix(c(2, 0,0,0, 2, 1, 0, 1, 2),3,3))
c3 <- mvrnorm(csize, mu=c(0,0,0),
              Sigma=matrix(c(2, 0,0,0, 2, 1, 0, 1, 2),3,3))

X <- rbind(c1, c2, c3)
classes <- c(rep(1, csize), rep(2, csize), rep(3, csize))

bdkmap <- bdk(X, classvec2classmat(classes), somgrid(4, 4, "hexagonal"))
plot(bdkmap, "prediction", palette=rainbow)


### Wine example
data(wines)
set.seed(7)

training <- sample(length(wine.classes), 120)
Xtraining <- scale(wines[training,])
bdk.wines <- bdk(Xtraining, classvec2classmat(wine.classes[training]),
                 grid = somgrid(5, 5, "hexagonal"))
Xtest <- scale(wines[-training,],
               center = attr(Xtraining, "scaled:center"),
               scale = attr(Xtraining, "scaled:scale"))

bdk.prediction <- predict(bdk.wines, newdata=Xtest)
table(wine.classes[-training], bdk.prediction$classif)