%BAYESC Bayes classifier based on given density estimates
%
% W = BAYESC(WA,WB, ... ,P,LABLIST)
% D = X*W
%
% INPUT
% WA, WB, ... Trained mappings for supplying class density estimates
% P Vector with class prior probabilities
% Default: equal priors
% LABLIST List of class names (labels)
% X Testset
%
% OUTPUT
% W Bayes classifier.
% D Classification matrix
%
% DESCRIPTION
% The trained mappings WA,WB, ... should supply proper densities estimates
% D for a dataset X by D = X*WA, etcetera. E.g. they should be trained by
% commands like GAUSSM(A), PARZENM(A), KNNM(A). Consequently, they should
% have a size of K x 1 (assuming that X and A are K-dimensional). Also
% sizes of K x N are supported, assuming a combined density estimate for N
% classes simultaneously. BAYESC weighs the class densitites by the class
% priors in P and names the classes by LABLIST. If LABLIST is not supplied,
% the labels stored in the mappings are used.
%
% REFERENCES
% 1. R.O. Duda, P.E. Hart, and D.G. Stork, Pattern classification, 2nd edition,
% John Wiley and Sons, New York, 2001.
% 2. A. Webb, Statistical Pattern Recognition, John Wiley & Sons, New York, 2002.
%
% SEE ALSO (<a href="http://37steps.com/prtools">PRTools Guide</a>)
% DATASETS, MAPPINGS, GAUSSM, PARZENM, KNNM
% Copyright: R.P.W. Duin, duin@ph.tn.tudelft.nl
% Faculty of Applied Sciences, Delft University of Technology
% P.O. Box 5046, 2600 GA Delft, The Netherlands
function w = bayesc(varargin)
n = nargin;
p = [];
lablist = [];
if (nargin > 1)
if nargin > 2 & (~ismapping(varargin{end-1}))
p = varargin{end-1};
lablist = varargin{end};
n = n-2;
elseif (~ismapping(varargin{end}))
p = varargin{end};
n = n-1;
end
end
if (nargin < 1 | isempty(varargin{1}))
% Definition
w = prmapping(mfilename,'combiner',{p,lablist});
w = setname(w,'Bayes Classifier');
elseif ismapping(varargin{1})
% Construction of the trained Bayes Classifier
w = [];
k = size(varargin{1},1);
wsize = 0;
for j=1:n
v = varargin{j};
if ~ismapping(v) | getout_conv(v) ~= 0
error('Density estimating mapping expected and not found')
end
if size(v,1) ~= k
error('Mappings / density estimators should be defined for the same dimensionality')
end
w = [w v];
wsize = wsize + size(v,2);
end
w = setsize(w,[k,wsize]);
c = size(w,2);
if isempty(p), p = ones(1,c)/c; end
if length(p) ~= c
error('Vector with prior probabilities has wrong length')
end
if isempty(lablist)
lablist = getlabels(w);
end
if size(lablist,1) ~= c
error('Label list has wrong size')
end
w = w*affine(p(:)');
w = setlabels(w,lablist);
w = setname(w,'Bayes Classifier');
else
error('Wrong input')
end