source: prextra/kannc.m

%KANNC 1-Nearest Neighbor Classifier using ANN Matlab Wrapper
%
%   W = KANNC(A,K,OPTION)
%
% INPUT
%   A      Dataset
%   K      Number of desired neighbours
%   OPTION Options for ANNQUERY
%
% OUTPUT
%   W      NN classifier using the ANN Query package
%
% DESCRIPTION  
% This is the nearest neighbor implementation for PRTools using the ANN
% Matlab Wrapper package. It should be in the path. If needed download it
% http://webscripts.softpedia.com/scriptDownload/ANN-MATLAB-Wrapper-Download-33976.html
%
% SEE ALSO
% MAPPINGS, DATASETS, ANNQUERY, KNNC

% Copyright: R.P.W. Duin, r.p.w.duin@prtools.org
% Faculty EWI, Delft University of Technology
% P.O. Box 5031, 2600 GA Delft, The Netherlands


function out = kannc(a,knn,opt)

        prtrace(mfilename);
        
        annquerycheck;
        name = 'KANNC';
        % No input data, return an untrained classifier.
        if nargin < 3, opt = []; end
        if nargin < 2, knn = 1; end
        if (nargin == 0) | (isempty(a))
                out = mapping(mfilename,'untrained',{knn,opt});
                out = setname(out,name);
        elseif isdataset(a) & ~ismapping(knn) % training
                islabtype(a,'crisp');
                isvaldfile(a,1,2); % at least 1 object per class, 2 classes
                a = testdatasize(a);
                a = testdatasize(a,'objects');
                a = seldat(a);    % get labeled objects only
                [m,k,c] = getsize(a);
                v.data = a;
                v.knn  = knn;
                v.opt  = opt;
                out = mapping(mfilename,'trained',v,getlablist(a),k,c);
                out = setname(out,name);
                out = setbatch(out,0);
        elseif nargin == 2 & ismapping(knn)  % execution
                % to avoid confusion, rename trained mapping
                w = knn;
                v = +w;   % get datafield
                [k,c] = size(w);
                nlab = getnlab(v.data);
                J = annquery(+v.data',(+a)',v.knn)';
                n = size(J,1); % no of test objects
                J = nlab(J);
                if size(J,2) == 1
                        out = zeros(c,n);
                        out([1:c:n*c]+J'-1) = ones(1,n);
                        out = out';
                else
                        out = hist(J',[1:c])';
                end
                % Use Bayes estimators for posteriors
                out = (out+ones(size(out,1),c))/(v.knn+c);
                out = setdat(a,out,w);
        else
                error('Illegal input')
        end