mog_dd.m

模式识别工具箱,希望对大家有用!

%MOG_DD Mixture of Gaussians data description
%
%       W = mog_dd(A,fracrej,n)
%
% Training of a mixture of gaussians, with n gaussians.
%
% Algorithm was inspired from NetLab, Bishop but changed very much.
% beforehand a number of gaussians is chosen, using EM the size and
% the place of the gaussians is optimized.
%
%       W = mog_dd(A,fracrej,n,ctype)
%
% By setting ctype, the covariance structure of the covariance can be
% set. There are three possibilities:
%      ctype = 'sphr'  : diagonal cov. matrix with equal values
%      ctype = 'diag'  : diagonal cov. matrix
%      ctype = 'full'  : full cov. matrix
%
%
% required functions: mogEM and mogP

% Copyright: D. Tax, R.P.W. Duin, [e-mail]
% Faculty of Applied Physics, Delft University of Technology
% P.O. Box 5046, 2600 GA Delft, The Netherlands

function W = mog_dd(a,fracrej,n,ctype,reg,numiters)

if (nargin<6)
  numiters = 25;
end
if (nargin<5)
  reg = 0.1;
end
if (nargin<4)
  ctype = 'sphr';
end
if (nargin<3)
  n = 5;
end
if (nargin<2)
  fracrej = 0.05;
end
if (nargin<1)|isempty(a)
  W = mapping(mfilename,{fracrej,n,ctype,reg});
  return
end

if isa(fracrej,'double')           %training

  if ~isa(a,'dataset')              %train on training set
    error('MOG_DD: I need a dataset to train.');
  end
  a = target_class(a);     % only use the target class
  [nlab,lablist,m,k,c,prob,featlist] = dataset(a);

  % train it
  [means, covs, priors] = mogEM(+a, n, ctype, reg, numiters);

  % obtain the threshold:
  d = sum(mogP(+a,means,covs,priors),2);
  thr = threshold(d,fracrej);

  %and save all useful data:
  W.m = means;
  W.c = covs;
  W.p = priors;
  W.threshold = thr;
%  W = {means,covs,priors,thr};
  W = mapping(mfilename,W,str2mat('target','outlier'),k,c);

else                               %testing

  [W,classlist,type,k,c] = mapping(fracrej);  % unpack
  [nlab,lablist,m,k,c,p] = dataset(a);

  %compute:
%  out = sum(mogP(+a,W{1},W{2},W{3}),2);
%  newout = [out, ones(m,1)*W{4}];
  out = sum(mogP(+a,W.m,W.c,W.p),2);
  newout = [out, ones(m,1)*W.threshold];

  W = dataset(newout,getlab(a),classlist,p,lablist);

end

return