Fixed confusion matrix

ragavvenkatesan · Mar 11, 2017 · 7245bb5 · 7245bb5
1 parent 9806a9a
commit 7245bb5
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 9 deletions.
diff --git a/yann/modules/resultor.py b/yann/modules/resultor.py
@@ -191,13 +191,13 @@ def _store_confusion_img (self, confusion, filename, verbose = 2):
         """
         if verbose >= 3:
             print ("... Saving the file down")
-        confusion = confusion / confusion.max() 
+        confusion = confusion / confusion.sum(axis = 0)
         fig = plt.figure()
         plt.matshow(confusion)
         plt.title('Confusion matrix')
-        plt.set_cmap('Greens')
+        plt.set_cmap('GnBu')
         plt.colorbar()
         plt.ylabel('True labels')
-        plt.xlabel('Predicated labels')
+        plt.xlabel('Predicted labels')
         plt.savefig(filename)
         plt.close('all')
diff --git a/yann/network.py b/yann/network.py
@@ -1935,13 +1935,15 @@ def _initialize_confusion (self, classifier = None, verbose = 2):
             print("... initializing confusion matrix function")   
 
         _classes = T.scalar('num_classes')
-        _predictions = self.inference_layers[classifier].predictions
-        _labels = self.y 
+        _predictions = self.inference_layers[classifier].predictions.dimshuffle(0, 'x')
+        _labels = self.y.dimshuffle(0, 'x')
+        _order = T.arange(self.num_classes_to_classify)
+
+        oneHot_labels = T.eq(_labels, _order).astype('int32')
+        oneHot_predictions = T.eq(_predictions, _order).astype('int32')
 
-        confusion = T.dot(  T.eq( _predictions.dimshuffle(0, 'x'), \
-                            T.arange(self.num_classes_to_classify).astype('int32')).T, \
-                            T.eq( _labels.dimshuffle(0,'x'), \
-                            T.arange(self.num_classes_to_classify).astype('int32')))
+        confusion = T.dot(oneHot_labels.T, oneHot_predictions)
+        # confusion_normalized = confusion / confusion.sum(axis = 0)     
 
         index = T.lscalar('index')
         self.mini_batch_confusion = theano.function(