%GENDATV Generation of a very large dataset
%
% [A,U] = GENDATV(M,C,K,R,T)
% A = GENDATV(M,U)
%
% INPUT
% M Number of objects to be generated (def. 10^4)
% C Number of classes (def. 10^2)
% K Number of features (def 10)
% R std. dev of class distributions (def 1)
% S Std. dev. of class priors (def 0)
% U Dataset with class means, class priors and R
%
% OUTPUT
% A C-class dataset with M objects and K features
% U Dataset with class means, class priors and R
%
% DESCRIPTION
% This routine generates an arbitrarily large dataset of C classes by first
% generating the class means from a standard K-dimensional normal
% distribution. The class priors are genetated from a N(1,S) distribution,
% After that their sum is normalized to one. The classes themselves have
% spherical normal distributions with an equal standard deviation R for all
% features and all classes.
%
% In U the class means, the class priors and R are returned, such that
% similar data from the same distribution can be generated.
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% DATASETS, PRDATASETS
% Copyright: R.P.W. Duin, r.p.w.duin@37steps.com
function [a,u] = gendatv(varargin)
[m,c,k,r,s] = setdefaults(varargin,10^4,10^2,10,1,0);
if isdataset(c)
u = c;
r = getuser(u,'R');
k = size(u,2);
else
u = gauss(c,zeros(1,k));
u = setlabels(u,[1:c]');
if s==0
u = setprior(u,0);
else
p = 1+randn(1,c)*s;
p(p<0) = 0;
p = p/sum(p);
u = setprior(u,p);
end
u = setuser(u,r,'R');
end
n = genclass(m,getprior(u));
a = gauss(n,u,r*eye(k));
return