getConfusionMatrixLatexSumbul.m

% This function just generates the latex code for the confusion
% matrix in the article.

clear all, close all

rng('shuffle')


r = RGCclass(0);
r.lazyLoad('knownSumbul');

% featuresUsed = { 'dendriticDensity', ...
%                  'somaArea', ...
%                  'totalDendriticLength', ...
%                  'meanTerminalSegmentLength'};

% featSet = 'Best3Set';
% featSet = 'A12Set';
% featSet = 'ManualUncorrSet';
featSet = 'Sumbul4';

switch(featSet)

 case 'Sumbul4'

   featuresUsed = {'biStratificationDistance', ...
                   'densityOfBranchPoints', ...
                   'stratificationDepth', ...
                   'totalDendriticLength'};
   
   %%% The features below this line were for Rana dataset  
   
 case 'Best3Set'
   % Best 3 feature set for classification
   featuresUsed = { 'fractalDimensionBoxCounting', ...
                    'somaArea', ...
                    'meanTerminalSegmentLength' };

  case 'A12Set'
    % A good 12 feature set for classification??? Dont remember
    featuresUsed = {'branchAssymetry', ...
                    'dendriticDensity', ...
                    'dendriticDiameter', ...
                    'dendriticField', ...
                    'densityOfBranchPoints', ...
                    'fractalDimensionBoxCounting', ...
                    'meanBranchAngle', ...
                    'meanTerminalSegmentLength', ...
                    'numBranchPoints', ...
                    'numSegments', ...
                    'somaArea', ...
                    'totalDendriticLength' }

  case 'ManualUncorrSet'
    % Trying to pick uncorrelated features, using at correlation matrix
    featuresUsed = { 'numBranchPoints', ...
                     'meanSegmentLength', ...
                     'dendriticField', ...
                     'totalDendriticLength', ...
                     'somaArea', ...
                     'biStratificationDistance', ...
                     'meanBranchAngle', ...
                     'meanSegmentTortuosity', ...
                     'stratificationDepth' };

  case 'Best5'
    % Best 5 feature set, from exhaustive search
    featuresUsed = { 'densityOfBranchPoints', ...
                     'dendriticField', ...
                     'fractalDimensionBoxCounting', ...
                     'meanTerminalSegmentLength', ...
                     'somaArea' };
  
  
  case 'Best6'
    % Best 6 feature set, from exhaustive search
    % Redid it with more iterations (20 vs 5), and then DBP changed to NBP
    featuresUsed = { 'numberOfBranchPoints', ...
                     'dendriticField', ...
                     'fractalDimensionBoxCounting', ...
                     'meanSegmentTortuosity', ...
                     'meanTerminalSegmentLength', ...
                     'somaArea' };

  otherwise
    fprintf('Unknown featSet: %s', featSet)
    keyboard
end

r.setFeatureMat(featuresUsed);

[CM,CMsd] = r.confusionMatrixCrossValidation(5,1000);

disp('\nConfusion matrix with cross validation')

% strHeader = 'C = \\left[ \\begin{array}{ccccc}\n';
strLine = ['\\textbf{%s} & $%.2f \\pm %.2f$  &  $%.2f \\pm %.2f$  &  $%.2f \\pm %.2f$  &  $%.2f \\pm %.2f$   &  $%.2f \\pm %.2f$\\\\\n'];
strEnd = '\\end{array} \\right]\n';

strHeader = ['\\begin{table}\\begin{tabular}{llllll}\n' ...
             '& \\bf{%s} & \\bf{%s} & \\bf{%s} & \\bf{%s} & \\bf{%s}\\\\\n'...
            '\\hline\n'];
strEnd = '\\end{tabular}\\end{table}\n'; 

fprintf(strHeader, ...
        r.RGCtypeName{find(r.RGCtypeID == 1,1)}, ...
        r.RGCtypeName{find(r.RGCtypeID == 2,1)}, ...
        r.RGCtypeName{find(r.RGCtypeID == 3,1)}, ...
        r.RGCtypeName{find(r.RGCtypeID == 4,1)}, ...
        r.RGCtypeName{find(r.RGCtypeID == 5,1)})
        
for i = 1:5
  
  fprintf(strLine, ...
          r.RGCtypeName{find(r.RGCtypeID == i,1)}, ...
          CM(i,1),CMsd(i,1), ...
          CM(i,2),CMsd(i,2), ...
          CM(i,3),CMsd(i,3), ...
          CM(i,4),CMsd(i,4), ...
          CM(i,5),CMsd(i,5))
          
end

fprintf(strEnd)

truePos = diag(CM)'./sum(CM);

for i = 1:5
  fprintf('%s: %.1f %%\n', r.RGCtypeName{find(r.RGCtypeID == i,1)}, ...
          truePos(i)*100)
end


disp('\nConfusion matrix generated with leave one out:')

CM2 = r.confusionMatrixLeaveOneOut();
makeLatexTableHeaders(r.RGCuniqueNames, r.RGCuniqueNames, CM2, ...
                      '%d','Predicted Class','True Class')