parzen

 PARZEN Estimates joint pdf using parzen kernels

        pdf = parzen(m,p,C,normalize)

         m: matrix (RxQ1) with points where joint pdf has to be evaluated
         p: data matrix (RxQ2) with points of distribution
         C: kernel bandwidth (covariance matrices) can be scalar, vector (1xQ2) 
            of width factors for each point of m or an n-array (RxRxQ2) with a whole
            covariance matrix for each point of m
 normalize: 1 (default) for minmax normalization, 0 in other case

 DESCRIPTION
   Estimates the joint pdf of data points stored in m, and
   evaluates it for points of p using a Parzen kernel estimation with
   gaussian RBF's of widths given by C

 EXAMPLES
   % PDF scatter plot
   figure;
   p = getpoints;
   u = rectop(linspace(-10,10,50),linspace(-10,10,50),'idx');
   pdf = parzen(u,p,0.1);
   planes(u,pdf);
   hold on;
   plotp(p);
   hold off;


   % Visualization of PDF map
   figure;
   p = [randn(3,100)-3 randn(3,100)+3];
   som = inisom(p,{20,20},{'x','y','z'});
   som = bsom(p,som,10,3);
   figure(1);
   pdf = parzen(som.w',p,0.1);
   figure(1); 
   planes(som.pos,pdf,{'low dim manifold pdf map'})