source: prextra/kdnnc.m @ 100

%KDNNC 1-Nearest Neighbor Classifier using KDTree
%
%   W = KNNC(A,TYPE)
%
% INPUT
%   A      Dataset
%   TYPE  'fast' (default) or 'good'
%
% OUTPUT
%   W      NN classifier using KDTree for finding nearest neighbors
%
% DESCRIPTION  
% This is a KD-Tree nearest neighbor implementation for PRTools using the
% KD-Tree package by Pramod Vemulapalli, see 
% http://www.mathworks.com/matlabcentral/fileexchange/26649-kdtree-implementation-in-matlab
% It should be in the Matlab search path.
%
% SEE ALSO
% MAPPINGS, DATASETS, 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 = kdnnc(a,par)

        prtrace(mfilename);
        
        kdtreecheck;
        name = 'KDTree NN Classifier';
        % No input data, return an untrained classifier.
        if nargin < 2, par = 'good'; end
        if (nargin == 0) | (isempty(a))
                out = mapping(mfilename,'untrained',par);
                out = setname(out,name);
        elseif isdataset(a) & isstr(par) % training
                if isempty(strmatch(par,char('fast','good')))
                        error('KDTree type should be either ''fast'' or ''good''')
                end
                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);
                data.kdtree = kd_buildtree(+a,0);
                data.nlab = getnlab(a);
                data.type = par;
                out = mapping(mfilename,'trained',data,getlablist(a),k,c);
                out = setname(out,name);
        elseif nargin == 2 & ismapping(par)  % execution
                kdtree = getdata(par,'kdtree');
                nlab = getdata(par,'nlab');
                if strcmp(getdata(par,'type'),'fast')
                        kdtype = @kd_closestpointfast;
                else
                        kdtype = @kd_closestpointgood;
                end
                a_data = +a;
                % PRTools want posteriors. In this case these are just 0/1 's
                out = zeros(size(a,1),size(getlab(par),1));
                m = size(a,1);
                s = sprintf('Testing %i objects: ',m);
                prwaitbar(m,s);
                for i=1:m;
                        prwaitbar(m,i,[s int2str(i)]);
                        j = kdtype(kdtree,a_data(i,:));
                        out(i,nlab(j)) = 1;
                end
                prwaitbar(0);
                out = setdat(a,out);
        end