confusion matrix diagnosis function

utils/diagnose_confusionmat.m

@@ -0,0 +1,55 @@
+function diagnose_confusionmat(gt, pred, labels)
+%DIAGNOSE_CONFUSIONMAT diagnose from classification results
+%   gt:: 
+%       Nx1 ground truth labels
+%   pred:: 
+%       Nx1 predicted labels
+%   labels::
+%       Nx1 cell, label names 
+
+labels = cellfun(@(s) strrep(s,'_',' '), labels, 'UniformOutput', false); 
+
+C = confusionmat(gt,pred);
+S = sum(C,2);
+C0 = C.*(1-eye(size(C)));
+S0 = sum(C0,2);
+
+fprintf('\n   DIAGNOSIS   \n===============\n\n');
+
+fprintf('There are %d testing instances coming from %d categories. \n', ...
+    sum(S), length(S));
+
+fprintf('There are %d wrongly classified instances, the overall \naccuracy is %.2f%%. \n\n',...
+    sum(S0), (1-sum(S0)/sum(S))*100);
+
+fprintf('   class   ::#test  #error     most-confused-to \n');
+fprintf('--------------------------------------------------\n');
+for c = 1: length(S), 
+    [y,i] = sort(C0(c,:),'descend');
+    fprintf('%10s :: %3d %2d(%4.1f%%)', ...
+        labels{c}(1:min(10,length(labels{c}))), S(c), S0(c), 100*S0(c)/S(c));
+    if y(1)>0, 
+        fprintf(' %10s:%3d(%5.1f%%)', ...
+            labels{i(1)}(1:min(10,length(labels{i(1)}))), y(1), 100*y(1)/S0(c));
+    end
+
+    fprintf('\n');
+end
+fprintf('\n');
+
+C2 = triu(C0+C0');
+[Y,I] = sort(C2(:),'descend');
+[I1, I2] = ind2sub(size(C),I);
+nPairs = min(20,max(find(Y,1,'last')));
+
+fprintf('Most confused pairs: \n\n');
+fprintf('            pair              #error(%%)  accum \n');
+fprintf('--------------------------------------------------\n');
+for p = 1:nPairs, 
+
+    fprintf('%12s <=>%12s %3d(%4.1f%%) %5.1f%%\n', ...
+        labels{I1(p)}(1:min(12,length(labels{I1(p)}))), ...
+        labels{I2(p)}(1:min(12,length(labels{I2(p)}))), ...
+        Y(p), 100*Y(p)/sum(S0), 100*sum(Y(1:p))/sum(S0));
+end
+fprintf('\n');