close all	% Close all figure windows
clear all	% Clear all variables in memory
%loadiris;
loadaba;

% ====== Data normalization to have zero mean and unity variance r.v.
new_feature = normal(feature);
data = [new_feature output];

misclassify = zeros(feature_n, 1);
all_in_n = size(data, 2)-1;
in_n = 3;
model_n = prod(1:all_in_n)/(prod(1:in_n)*prod(1:(all_in_n-in_n)));
index = zeros(model_n, in_n);
misclassify = zeros(model_n, 1);
K = 1;

% ====== Construct KNNR with different input variables
fprintf('\nConstruct %d KNNR models, each with %d inputs selected from %d candidates...\n\n',...
	model_n, in_n, all_in_n);

% ====== You need to replace the following lines to generate 
% "index" and "misclassify" via exhaustive search
index = [ 
     1     6     7;	% Inputs 1, 6, and 7 are selected
     2     5     6;	% Inputs 2, 5, and 6 are selected
     1     3     4;	% Inputs 1, 3, and 4 are selected
     3     4     6;	% Inputs 3, 4, and 6 are selected
     2     4     5;	% Inputs 2, 4, and 5 are selected
     4     6     7;	% Inputs 4, 6, and 7 are selected
     1     4     5;
     1     3     8;
     4     5     8;
     2     6     8;
     2     4     6;
     1     6     8;
     1     5     6;
     1     4     7;
     2     6     7;
     1     3     6;
     3     6     8;
     1     5     7;
     1     3     7;
     1     2     6;
     5     6     8;
     4     5     6;
     3     5     8;
     2     7     8;
     2     5     8;
     1     5     8;
     1     4     8;
     3     4     5;
     2     3     6;
     2     3     4;
     1     7     8;
     1     3     5;
     2     4     7;
     2     3     5;
     1     2     5;
     1     2     3;
     6     7     8;
     5     6     7;
     4     5     7;
     3     4     7;
     2     3     7;
     1     2     7;
     1     2     4;
     4     7     8;
     4     6     8;
     3     5     6;
     3     4     8;
     1     4     6;
     1     2     8;
     5     7     8;
     3     7     8;
     2     5     7;
     3     6     7;
     3     5     7;
     2     3     8;
     2     4     8];
misclassify = floor(50*rand(model_n,1));
% ====== Don't change anything below this line!

% ====== Reordering according to misclassification numbers
[a b] = sort(misclassify);
b = flipud(b);		% List according to decreasing trn error
misclassify = misclassify(b);
index = index(b, :);

% ====== Display the results
figTitle = 'Exhaustive Input Selection';
figure('Name', figTitle, 'NumberTitle', 'off');
x = (1:model_n)';
error_rate = misclassify/size(data,1)*100;
[a, b] = min(error_rate);
fprintf('Minimal error rate = %.1f%%.\n', a);
fprintf('Selected inputs: ');
for p = 1:in_n,
	fprintf(' %s', deblank(input_name(index(size(index,1),p), :)));
end
fprintf('\n');

subplot(211);
plot(x, error_rate, '-', x, error_rate, 'co');
tmp = x(:, ones(1, 3))';
X = tmp(:);
tmp = [zeros(model_n, 1) error_rate nan*ones(model_n, 1)]';
Y = tmp(:);
hold on; plot(X, Y, 'g'); hold off;
axis([-inf inf 0 100]);
axis([1 model_n -inf inf]);
set(gca, 'xticklabels', []);
ylabel('LOU Error Rates (%)');
grid on
h = findobj(gcf, 'type', 'line');
set(h, 'linewidth', 2)

% ====== Add text of input variables
for k = 1:model_n,
	text(x(k), 0, ...
	[deblank(input_name(index(k,1), :)) ' ' ...
	 deblank(input_name(index(k,2), :)) ' ' ...
	 deblank(input_name(index(k,3), :)) ' ']);
end
h = findobj(gcf, 'type', 'text');
set(h, 'rot', 90, 'fontsize', 9, 'hori', 'right');
title(['Exhaustive Search for ', int2str(in_n), ' Inputs']);