| % PCDA Performs Principal Component Discriminant Analysis |
| % |
| % USAGE: |
| % strike=pcda(test,train,class,ndis,plflag,prflag); |
| % |
| % INPUTS: |
| % test : an (m x p) array of test data to classify |
| %m is the number of test data to classify |
| %p is the number of variates |
| % train: an (n x p) array of data used for training |
| %n is the number of data of the training set |
| % class: an (n x 1) vector with the class of the train dataset |
| %as an integer ranging from 1 to the number of classes |
| % ndis: optional integer number specifying the procedure used for |
| %the density estimation by Kernel regression for each class: |
| %n(k) <= ndis ==> estimate of the density function for each sample |
| %computed from the whole training dataset by Kernel regression |
| %n(k) > ndis ==> estimate of the density function computed once from |
| %the training set on an equispaced grid with approximately |
| %ndis points by Kernel regression; then for each |
| %test data, density is computed by linear interpolation. |
| %This option speeds up computations |
| %ndis < 0 or missing always implies the first option |
| %(n(k) is the number of training data belonging to class k) |
| % plflag: optional flag for producing plots: |
| %plflag = 0 ==> no plot |
| %plflag = 1 ==> summary plots |
| %plflag = 2 ==> detailed plots |
| %plflag < 0 or missing ==> no plot |
| % prflag: optional flag for prior: |
| %prflag = 0 ==> uniform prior |
| %prflag = 1 ==> class frequencies estimated on the training set |
| % |
| % |
| % OUTPUTS |
| % strike: an (m x 1) vector containing the class of the test dataset as |
| %assigned by the method |
| % |
| % Example: |
| % train = mvnrnd( 1 1, eye(2), 100); … |
| %mvnrnd(−1 −1, 2*eye(2), 100)]; |
| % class = repmat(1,100,1); repmat(2,100,1)]; |
| % test = unifrnd(−5, 5, 100, 2); |
| % strike = pcda(test,train,class); |
| % |
| % Copyright 2002–2008 Umberto Amato, Istituto per le Applicazioni del Calcolo |
| % Mauro Picone CNR, Napoli (Italy) and Anestis Antoniadis, Laboratoire de |
| % Modelisation et Calcul IMAG, Universite J. Fourier, Grenoble (France) |
