predict.relationalsom: Map new data to a fitted Relational Self-Organising Map
In yasomi: Yet Another Self Organising Map Implementation

Description Usage Arguments Details Value Author(s) Examples

Map new data to a fitted Self-Organising Map, i.e., compute the best matching unit for each given observation.

1
2
3

## S3 method for class 'relationalsom'
predict(object, newdata,
        with.secondwinner=FALSE,...)

`object`	an object of class `"relationalsom"`
`newdata`	an object of class `"crossdist"` compatible with the fitted som (see details)
`with.secondwinner`	switch to specify whether the second best matching unit should be computed and returned
`...`	not used

The newdata object must be of class "crossdist" as returned by dist and must contain the dissimilarities between the original data (used to fit the SOM) and the new data (for which the quantisation error is to be computed).

A list with components

`classif`	a vector of integer indicating to which unit each observation has been assigned
`error`	the quantisation error of the observations by the prototypes of this Self-Organising Map (see `error.quantisation.relationalsom` for a definition of this value)
`rdist`	a matrix of squared dissimilarities between the new data and the prototypes, as obtained by the relational formula (with possible negative values preserved)
`winners`	if `with.secondwinner=TRUE` a two column matrix with a line per observation containing the index of the best matching unit and of the second best matching unit (in this order)

Fabrice Rossi

data(iris)
# scaling
data <- scale(iris[1:4])

# use only part of the data
train <- sample(1:nrow(data),100)
data.train <- data[train,]
d <- dist(data.train)

# a small hexagonal grid
sg <- somgrid(xdim=5,ydim=5,topo="hex")

# fit the SOM
som <- batchsom(d,sg)

# map remaining data
results <- predict(som,dist(data.train,data[-train,]))
print(paste("Learning quantisation error:",error.quantisation(som)))
print(paste("Test quantisation error:",results$error))