knn_classify.m

function [new_accu, train_accu] = knn_classify(train_data, train_label, new_data, new_label, K)
new_data_size = length(new_label);
train_data_size = length(train_label);
% Standardise

mean_train_data = mean(train_data);
std_train_data = std(train_data);

standardise_new_data = bsxfun(@minus, new_data, mean_train_data );
standardise_new_data = bsxfun(@rdivide, standardise_new_data, std_train_data);
%(new_data-mean_train_data)./std_train_data
%standardise_train_data = (new_data-mean_train_data)./std_train_data

standardise_train_data = bsxfun(@minus, train_data, mean_train_data );
standardise_train_data = bsxfun(@rdivide, standardise_train_data, std_train_data);

test_label = [];
all_distances = {};
for i =1:new_data_size
    distances = [];
    sample = standardise_new_data(i,:);
    for j=1:train_data_size
        distances(end+1) = norm(sample-standardise_train_data(j,:));
    end
    %all_distances{i} = (distances);    
    [sorted idx] = sort(distances, 'ascend');
    indices= idx(1:K);
    pivot = sorted(K);
    k=K;
    if pivot == sorted(K+1)
        % search for all indices s
        while sorted(k+1) ~= pivot
            k=k+1;
            indices(end+1)=k;
        end
            
    end
    all_labels = [];
    for j=1:length(indices)
        all_labels(end+1) = train_label(indices(j));
    end
    test_label(end+1) = mode(all_labels);
    
end

%test_l = test_label;
new_accu=0;
train_accu = 0;

for  i=1:new_data_size
    if test_label(i) == new_label(i)
        new_accu = new_accu+1;
    end
end
new_accu = new_accu/new_data_size;

test_label = [];
all_distances = {};
for i =1:train_data_size
    distances = [];
    sample = standardise_train_data(i,:);
    for j=1:train_data_size
        if i ~= j
            distances(end+1) = norm(sample-standardise_train_data(j,:));
        end
    end
    all_distances{i} = distances;
end

for i=1:train_data_size
    distances = cell2mat(all_distances(i));
    [sorted idx] = sort(distances, 'ascend');
    indices= idx(1:K);
    pivot = sorted(K);
    k=K;
    if pivot == sorted(K+1)
        % search for all indices s
        while sorted(k+1) ~= pivot
            k=k+1;
            indices(end+1)=k;
        end
            
    end
    all_labels = [];
    for j=1:length(indices)
        all_labels(end+1) = train_label(indices(j));        
    end
    test_label(end+1) = mode(all_labels);
end
%train_l = test_label;
train_accu =0;
for  i=1:train_data_size
    if test_label(i) == train_label(i)
        train_accu = train_accu+1;
    end
end

train_accu = train_accu/(train_data_size);      


end

% k-nearest neighbor classifier
% Input:
%  train_data: N*D matrix, each row as a sample and each column as a
%  feature
%  train_label: N*1 vector, each row as a label
%  new_data: M*D matrix, each row as a sample and each column as a
%  feature
%  new_label: M*1 vector, each row as a label
%  K: number of nearest neighbors
%
% Output:
%  new_accu: accuracy of classifying new_data
%  train_accu: accuracy of classifying train_data (using leave-one-out
%  strategy)
%
% CSCI 567: Machine Learning, Fall 2015, Homework 1