%
% Examples 9.15 and 9.17
% 
% epage 392 (cluster/
% 
% Written by:
% -- 
% John L. Weatherwax                2008-02-20
% 
% email: wax@alum.mit.edu
% 
% Please send comments and especially bug reports to the
% above email address.
% 
%-----

clear all; 
close all; 
clc; 

addpath('../../Code/CSTool');
load bank;

% lump everything into one data matrix: 
X = [forge; genuine]; 

X = zscore(X);

%--
% Apply Hierarchical clustering (epage 382): 
%--
if( 0 ) 
  ye = pdist(X,'euclid'); 
  %ye = pdist(X,'cityblock');
  z = linkage(ye,'single');
  %z = linkage(ye,'complete'); 
  %z = linkage(ye,'average'); 
  %z = linkage(ye,'ward'); 
  [H,T] = dendrogram(z); 
  xlabel('datum instance'); 
  title( 'algomative clustering with single linkage' );
  saveas( gcf, 'prob_9_10_hierarchical_linkage', 'eps' ); 

  figure; hist(T,length(unique(T)));
  title( 'the distribution of data points under each dendrogram leaf' ); 
  xlabel( 'dendrogram label' ); ylabel( 'number of data points in each label' ); 
  saveas( gcf, 'prob_9_10_den_hist', 'eps' ); 

  fprintf('the cophenet coefficient for this clustering is = %10.5f\n',cophenet(z,ye)); 
end

%--
% Apply k-means clustering ()
%--
if( 1 ) 
  k = 2;           % <- assume we have two clusters 
  [cid,nr,centers] = cskmeans(X,k); 

  % plot the two clusters overlayed with truth: 
  figure; 
  tm=plot( X(1:100,1), X(1:100,2), 'sk' ); hold on;    % <- this is the truth 
  tw=plot( X(101:200,1), X(101:200,2), 'dk' ); 

  forge_cluster = cid(1);                              % <- this is what MATLAB called the forged cluster ... 
  if( forge_cluster==1 )                               % <- so this must be what MATLAB called the womans cluster ... 
    genine_cluster=2; 
  else
    genine_cluster=1; 
  end

  dim1=1; dim2=2; 
  am=plot( X(find(cid==forge_cluster),dim1), X(find(cid==forge_cluster),dim2), 'sr', 'markersize', 10 ); 
  aw=plot( X(find(cid==genine_cluster),dim1), X(find(cid==genine_cluster),dim2), 'dr', 'markersize', 10 ); 
  legend( [tm, tw, am, aw], { 'true forge', 'true genine', 'approx forge', 'approx genine' }, 'location', 'best' ); 
  saveas( gcf, 'prob_9_10_k_means', 'eps' ); 
  
  % compute how accurate this clustering is: 
  %
  n_forge_correct   = sum(cid(1:100)==forge_cluster);
  n_genuine_correct = sum(cid(101:200)==genine_cluster);
  pcc = (n_forge_correct + n_genuine_correct)/200; 
  fprintf( 'precent of correct clustering = %10.5f\n', pcc ); 
end