The Complete SPRLIB & ANNLIB

som_goodness

- compute the goodness-of-fit between a trained SOM and a dataset

SYNOPSIS

void som_goodness (net, dset, ndim, nx, ny, spcost, result)

ARGUMENTS

NET *net (Trained) SOM network
DATASET *dset Dataset, for which the distance (similarity or goodness) to the SOM has to be determined
int ndim Number of dimensions of the SOM (1-D or 2-D)
int nx Map size in the first direction
int ny Map size in the second direction (should be 1 for 1-D SOMs)
double **spcost The shortest-path distance matrix of the SOM
double *result Result: the goodness-of-fit of the SOM to this dataset

RETURNS

The goodness-of-fit, in the result parameter.

FUNCTION

Computes the goodness-of-fit between a (trained) SOM and a dataset, according to "Comparing Self-Organizing Maps" - S. Kaski, K. Lagus, in: Proc. ICANN96, Lecture Notes in Computer Science 1112, Springer, 1996, pp. 809 - 814. This measure takes, besides quantization error, also a penalty for distortions in topology preservation into account.

NOTE

The function assumes that the matrix of shortest-path distances between a node and every other node in the map spcost is already allocated and determined (using make_spcost_matrix).

SEE ALSO

make_spcost_matrix , eval_set_on_net

This document was generated using api2html on Thu Mar 5 09:00:00 MET DST 1998