www.pudn.com > standardmodelrelease.zip > CLSnn.m

function [Model,looerrors] = CLSnn(X,y,sPARAMS);
%function [Model,looerrors] = CLSnn(X,y,sPARAMS);
%
%Builds a NN classifier
%X contains the data-points as COLUMNS, i.e., X is nfeatures \times nexamples
%y is a column vector of all the labels. y is nexamples \times 1
%sPARAMS is a structure of parameters:
%sPARAMS.k is the k for knn
%sPARAMS.deg determines the p-norm to be used as distance
%Model contains the parameters of the nn classifier 

if nargin<3
  sPARAMS.k = 1;
end

if ~isfield(sPARAMS,'deg')
  sPARAMS.deg = 2;
end

Model.k = sPARAMS.k;
Model.deg = sPARAMS.deg;
Model.trainX = X;
Model.trainy = y;

if isfield(sPARAMS,'numindeces')
  Model.numindeces = sPARAMS.numindeces;
else
  Model.numindeces = inf;
end

if nargout>1
  deg = Model.deg;
  len1 = size(X,2);
  len2 = size(Model.trainX,2);
  if deg==2
    X2 = sum((X).^2,1);
    Z2 = sum((Model.trainX).^2,1);
    distance = (repmat(Z2,len1,1)+repmat(X2',1,len2)-2*X'*Model.trainX)';
  else
    distance = zeros(len2,len1);
    for i = 1:len1,
      for j = 1:len2,
	distance(j,i) = sum(abs(X(:,i)-Model.trainX(:,j)).^deg);
      end
    end
  end
  
  maxdistance = max(distance(:));
  distance = distance + eye(len1)*maxdistance;
  
  [sorted,index] = sort(distance);
  yy = Model.trainy(index);
  if Model.k>1
    weights = mean(yy(1:Model.k,:),1)';
    disp('kNN weights::just sign no voting');
    labels = sign(weights);
  else
    labels = yy(1,:)';
    weights = 1./(sorted(1,:)'+eps);
  end
  looerrors = mean(labels==y);
end