som.tune: Parameter tuning for Self-Organising Maps

In yasomi: Yet Another Self Organising Map Implementation

Description

This function tunes some parameters of a Self-Organising Map by optimising a specified error measure. The prior structure is not optimised by this function.

Usage

som.tune(data, somgrid, control = som.tunecontrol(somgrid),
         weights, verbose = FALSE, internalVerbose = FALSE)
## S3 method for class 'somtune'
print(x,...)

Arguments

data	the data to which the SOM will be fitted. This can be, e.g., a matrix or data frame of observations (which should be scaled), or a distance matrix
somgrid	an object of class 'somgrid'
control	a control object which specifies the parameter to optimize
weights	optional weights for the data points
verbose	switch for tracing the tuning process
internalVerbose	switch for tracing each call to batchsom
x	an object of class "somtune"
...	not used

Details

The optimisation is done via a simple grid search that fit a Self-Organising Map for each possible combination of the parameters to tune. The prior structure itself is excluded from the optimisation as most quantisation related error measures decrease with the size of the grid.

Value

An object of class "somtune", a list with components

best.som	the best SOM according to the chosen error criterion
errors	the error for each configuration
quantisation	the quantisation error for each configuration
isquant	TRUE if the best SOM was chosen according to the quantisation error
control	the control object used by this call
dimensions	a list of strings with the names of the parameters that were varied by the function
init,assignement,radii,annealing,kernel	5 vectors containing the parameters used by each tested configuration
best.index	the index of the configuration used by the best SOM

Note

this function is modelled after David Mayer's tune function.

Author(s)

Fabrice Rossi

See Also

som.tunecontrol and plot.somtune

Examples

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

# a medium hexagonal grid
sg <- somgrid(xdim=15,ydim=15,topo="hex")

# choose a good SOM via Kaski and Lagus' error measure
tune.results <- som.tune(data,sg,
                         som.tunecontrol(sg,criterion=error.kaskilagus))

# print the results
print(tune.results)

# umatrix of the best SOM
umatrix(tune.results$best.som)

# give more weight to the first specie (i.e. setosa)
tune.results <-  som.tune(data,sg,
                          som.tunecontrol(sg,criterion=error.kaskilagus),
                          weights=c(rep(3,50),rep(1,100)))

# print the results
print(tune.results)

# umatrix of the best SOM
umatrix(tune.results$best.som)

yasomi documentation built on May 2, 2019, 5:59 p.m.