Trainable classifier based on Parzen density estimation
[W,H] = PARZENDC(A,H)
For each of the classes in the dataset A, a Parzen density is estimated using PARZENML. For each class, a feature normalisation on variance is included in the procedure. As a result, the Parzen density estimate uses different smoothing parameters for each class and each feature.
If a set of smoothing parameters H is specified, no learning is performed, only the classifier W is produced. H should have the size of [C x K] if A has C classes and K features. If the size of H is [1 x K] or [C x 1], or [1 x 1], then identical values are assumed for all the classes and/or features.
The densities for the points of a dataset B can be found by D = B*W. D is an [M x C] dataset, if B has M objects.
prex_density, for, densities, and, prex_parzen, for, differences, between,
PARZENC, PARZENDC and PARZENM.