error.kaskilagus.somnum: Kaski and Lagus' error measure for Self-Organising Maps on...
In yasomi: Yet Another Self Organising Map Implementation

Description Usage Arguments Value Author(s) References See Also Examples

Compute an error measure of a fitted Self-Organising Maps defined by S. Kaski and K. Lagus. For vector data, the error combines for each observation its quantisation error and a graph based distance in the prior structure compute with the Euclidean distance between the prototypes (in the data space).

1 2	## S3 method for class 'somnum' error.kaskilagus(som, newdata,...)

`som`	an object of class `"somnum"`
`newdata`	an optional matrix or a data frame of observations
`...`	not used

If newdata is not given, the function returns the error made by the fitted som on the data used to fit it. Those data must have been saved in the som object (this is the default behavior of batchsom). When newdata is specified, the function returns the error made by the fitted som on the corresponding data.

Fabrice Rossi

Kaski, S. and Lagus, K. (1996) Comparing self-organizing maps, in: C. von der Malsburg, W. von Seelen, J. Vorbrüggen, B. Sendhoff (eds.), Proceedings of International Conference on Artificial Neural Networks (ICANN'96, Bochum, Germany), vol. 1112 of Lecture Notes in Computer Science, Springer, pp. 809–814.

error.quantisation.somnum, som.tunecontrol

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
st <- som.tune(data,sg,som.tunecontrol(sg,criterion=error.kaskilagus))
som <- st$best.som

# and display the prototypes
plot(som)