Classifier normalisation for ML posterior probabilities
W = CNORMC(W,A)
The mapping W is scaled such that the likelihood of the posterior probabilities of the samples in A, estimated by A*W*SIGM, are maximised. This is particularly suitable for two-class discriminants. To obtain consistent classifiers in PRTools, it is necessary to call CNORMC in the construction of all classifiers that output distances instead of densities or posterior probability estimates.
If A has soft labels or target labels, W is returned without change.