dd_aic.m

模式识别工具箱,希望对大家有用!

function e = dd_aic(w,x)
%DD_AIC compute the Akaike Information Criterion for MoG
%
% e = dd_aic(w,x)
%
% Compute the Akaike Information Criterion of the Mixture of
% Gaussians. We assume we have a trained classifier w and data x.
%
% also see dd_error, dd_roc, dd_auc

% 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

[W,labl,map,d] = mapping(w)
if ~strcmp(map,'mog_dd')
end

p = w*x; p = +p(:,1);

switch map
  case 'gauss_dd'
    nrparam = d + d*(d+1)/2;  %mean and cov.matrix
    
  case 'parzen_dd'
    nrparam = 1;   %width parameter
    
  case 'mog_dd'
    c = size(W{1},1);
    [n,d] = size(x);

    % the number of parameters
    % for all covariance versions, the priors and the means are the same:
    nrparam = c + c*d;
    switch ctype
      case 'sphr'
        nrparam = nrparam + c;
      case 'diag'
        nrparam = nrparam + c*d;
      case 'full'
        nrparam = nrparam + c*d*(d+1)/2;
      otherwise
        error('Type of covariance matrix not recognized')
    end
  otherwise
    error('AIC cannot be computed for this mapping!');
end

% For the loglikelihood:
e = -2*sum(log()) + 2*nrparam;
%strangely this one does not seem to work!:
%e = -2*sum(log(sum(p,2))) + 2*nrparam/n;

return