knn.m

一个关于k最近临法分类的matlab程序

function [eachClass, nearestSampleIndex, knnmat] = ...
	knn(sampledata, testdata, k)
% KNN	K-nearest neighbor rule for classification
%	Usage:
%	[EACH_CLASS, NEAREST_SAMPLE_INDEX] = KNN(SAMPLE, INPUT, K)
%
%	SAMPLE: Sample data set (The last column is the desired class of each
%		sample vector. The values of the last column are assumed to
%		be integers ranging from 1 to N.)
%	INPUT: Test input matrix (each row is an input vector)
%	K: the "k" in "k nearest neighbor"
%	EACH_CLASS: A vector denoting the KNN output class of each input vector
%	NEAREST_SAMPLE_INDEX: Index of SAMPLE that are closest to INPUT

%	Roger Jang, 9703xx, 990613, 991215 

if nargin < 3, k = 1;, end

featureNum = size(sampledata,2)-1;
sampleInput = sampledata(:, 1:featureNum);
sampleOutput = sampledata(:, featureNum+1);
classLabel = countele(sampleOutput);	% possible output class
classNum = length(classLabel);
testNum = size(testdata, 1);		% no. of test input vectors
testInput = testdata(:, 1:featureNum);	% strip out class info, if any

% Euclidean distance matrix between sampleInput and testInput
distmat = vecdist(sampleInput, testInput);

% knnmat(i,j) = class of i-th nearest point of j-th test input vector
% (The size of knnmat is k by testNum.)
[junk, nearestSampleIndex] = sort(distmat, 1);
% The following "reshape" is necessary if k == 1.
knnmat = reshape(sampleOutput(nearestSampleIndex(1:k,:)), k, testNum);

% class_count(i,j) = number of class-i points in j-th test input's neighborhood
% (The size of class_count is classNum by testNum.)
class_count = zeros(classNum, testNum);
for i = 1:testNum,
	[sortedElement, elementCount] = countele(knnmat(:,i));
	class_count(sortedElement, i) = elementCount;
end

[junk, eachClass] = max(class_count, [], 1);
eachClass = eachClass';