DISNORM

Trainable mapping for dissimilarity matrix normalisation

    V = DISNORM(D,OPT)
    V = D*DISNORM([],OPT)
    V = D*DISNORM(OPT)
    F = E*V

Input
D	Dissimilarity matrix, double, prdataset or disdataset
E	Matrix to be normalized, e.g. D itself
OPT	'max' : maximum dissimilarity is set to 1 by global rescaling  'mean': average dissimilarity is set to 1 by global rescaling (default)

Output
V	Trained mapping
F	Normalized dissimilarity data

Description

Operation on dissimilarity matrices, like the computation of classifiers  in dissimilarity space, may depend on the scaling of the dissimilarities  (a single scalar for the entire matrix). This routine computes a scaling  for a giving matrix, e.g. a training set and applies it to other  matrices, e.g. the same training set or based on a test set.

There is a fixed version of this mapping: DNORM

See also

datasets, mappings, dnorm,