GAUSSM

Trainable mapping, mixture of Gaussians (MoG) density estimate

   W = GAUSSM(A,K,R,S,M)
   W = A*GAUSSM([],K,R,S,M)
   W = A*GAUSSM(K,R,S,M)

Input
A	Dataset
K	Number of Gaussians per class
R,S,M	Regularisation parameters, 0 <= R,S <= 1, see QDC

Output
W	Mixture of Gaussians density estimate

Description

Estimation of a PDF by the dataset A by a mixture of Gaussians procedure.  Use is made of EMCLUST(A,QDC,K). Unlabeled objects are neglected, unless  A is entirely unlabeled or double. Then all objects are used. If A is a  multi-class crisp labeled dataset the densities are estimated class by  class and then weighted and combined according their prior probabilities.  Use +A instead of A to obtain a single set of Gaussians. In all cases,  just single density estimator W is returned.

Note that it is necessary to set the label type of A to soft labels  (A = LABTYPE(A,'soft') in order to use the traditional EM algorithm  based on posterior probabilities instead of using crisp labels.

The mapping W may be applied to a new dataset B using DENSITY = B*W.

   W = A*GAUSSM

uses a single Gaussian per class (K=1) and no regularisation. If  regulariisation is desired, also K should be supplied.

Example(s)

 a = gendatb;
 w = a*gaussm(2);
 scatterd(a)
 plotm(w)

See also

datasets, mappings, qdc, mogc, emclust, plotm, testc,