source: distools/parzend_map.m @ 125

%PARZEN_MAP Map a dissimilarity dataset on a Parzen densities based classifier
% 
%       F = PARZEN_MAP(D,W)
%
% INPUT
%   D   Dissimilarity dataset
%   W   Trained Parzen dissimilarity classifier mapping
%
% OUTPUT
%   F   Mapped dataset
%
% DESCRIPTION 
% 
% SEE ALSO
% MAPPINGS, DATASETS, PARZENDDC

% Copyright: R.P.W. Duin, r.p.w.duin@prtools.org
% Faculty EWI, Delft University of Technology
% P.O. Box 5031, 2600 GA Delft, The Netherlands

function f = parzend_map(d,w)
        prtrace(mfilename);

        % If no mapping is supplied, train one.
        if (nargin < 2)
                w = parzenddc(d); 
        end

        pars = getdata(w);
        I = getdata(w,'objects');
        p = getdata(w,'prior');
        nlab = getdata(w,'nlab');
        h = getdata(w,'smoothpar');
        q = getdata(w,'weights');
        [k,c] = getsize(w);
        [m,k] = size(d);
        e = +d(:,I);
        
        f = zeros(m,c);
        for j=1:c
                J = find(nlab==j);
                f(:,j) = p(j)*mean(repmat(q(J)',m,1).*exp(-(e(:,J).^2)./repmat((2.*h(J).*h(J))',m,1)),2);
        end
        
        f = setdata(d,f,getlabels(w));
        
return