makegaussmixnd.m

function [points,labels] = makegaussmixnd (centers, stdevs, ppm, doplot)
 
% MAKEGAUSSMIXND  Produces points randomly sampled from a Gaussian Mixture Model in N-dimensions
%
% [points,labels] = makegaussmixNd (centers, stdevs, ppm, [doplot])
% 
% centers is a M x N matrix.
% stdevs is a M-long vector, or a single value
% numpts per mixture is an integer or an M-long vector
%    There are M N-dimensional mixture models. 
%    The m-th mixture is centered at (centers(m,1),...,centers(m,N))
%      and has standard deviation stdevs (if its an integer) or stdevs(m) in each dimension with a diagonal correlation matrix.
%      ppm (if ppm is an integer) or ppm (if it's a M-long vector) points are sampled from it.
% if doplot is provided, it should be a M-long character string with valid colors.
%
% The number of points generated is P = ppm*M or sum(ppm) 
%  depending on whether ppm is an integer or an m-long vector respectively.
% 
% points is a P x N matrix
% labels is a P-long vector
%  points(i,:) = (points(i,1),..,points(i,N)) is the i-th point and comes from the
%  labels(i)-th mixture model.
%
% Dinoj Surendran (dinoj@cs.uchicago.edu)
%  15 May 2004
%
% Example: [points,labels]=makegaussmixnd(4,[0 0; 0 5; 5 0; 5 5],1,500);
%  generates 2000 points, 
%    500 are from a gaussian distribution centered at (0,0) with stdeviation 1 
%    500 are from a gaussian distribution centered at (0,5) with stdeviation 1
%    500 are from a gaussian distribution centered at (5,0) with stdeviation 1
%    500 are from a gaussian distribution centered at (5,5) with stdeviation 1
%
% Note: It uses the mvnrnd function that appears in MAtlab 6.5 (but
% not 6.0). 

if ((nargin < 3) | (nargin > 4))
   error ('This function needs 3 (or 4) arguments! You have sinned. \nType "help makegaussmixnd" and say seventeen Hail Marys.');
end

if (~isnumeric(centers) | (length(size(centers)) ~= 2))
  error (sprintf ('The second argument should be a matrix with %d rows\n',M));
end

M = size(centers,1);
N = size(centers,2);

if (~isnumeric(stdevs) | (length(size(stdevs)) ~= 2) | ((length(stdevs) ~= M) & (length(stdevs)~=1)) | (0 < length(find(stdevs<=0))))
  error (sprintf ('The third argument should be a vector with %d elements, all positive real values\n',M));
end

if (~isnumeric(ppm) | (length(size(ppm)) ~= 2) | ((size(ppm,1)~=1) & (size(ppm,2)~=1)) | ((length(ppm) ~= 1) & (length(ppm) ~= M)))
  error (sprintf('The fourth argument should be a positive integer or a vector with %d positive integers\n',M));
end

if (1==length(ppm))
  ppm = repmat(ppm,1,M);
end
if (1==length(stdevs))
  stdevs = repmat(stdevs,1,M);
end

N = size(centers,2);        % dimension of the points

points = [];
labels = [];

for m=1:M,
   labels = [labels repmat(m,1,ppm(m))];
end
P = length(labels);

for pp = 1 : P                                  % I hope matlab distinguishes between upper and lower case in windows
  means (pp,:) = centers (labels (pp), :);
end
devs  = diag (repmat (stdevs(labels(pp)), 1, N));
points = mvnrnd (means,devs);                     % assumes Matlab
                                                  % 6.5 at least

if (nargin == 4) & ischar(doplot) & (M == length(doplot))
  plotcol (points, ppm, doplot);
end