Refactor to move accuracy metrics into model

capstone_traffic_light_classifier/main.py

@@ -23,11 +23,6 @@
     # Define model
     classifier = TrafficLightClassifier(x, targets, p, n_classes, learning_rate=1e-4)
 
-    # Define metrics
-    correct_predictions = tf.equal(tf.argmax(classifier.inference, axis=1),
-                                   tf.argmax(tf.one_hot(targets, depth=n_classes), axis=1))
-    accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32))
-
     with tf.Session() as sess:
 
         # Initialize all variables
@@ -60,8 +55,8 @@
             num_test_batches = 500
             for _ in range(num_test_batches):
                 x_batch, y_batch = dataset.load_batch(batch_size)
-                average_test_accuracy += sess.run(fetches=accuracy,
+                average_test_accuracy += sess.run(fetches=classifier.accuracy,
                                                   feed_dict={x: x_batch, targets: y_batch, p: 1.})
             average_test_accuracy /= num_test_batches
-            print('TEST accuracy: {:.03f}'.format(average_test_accuracy))
+            print('Training accuracy: {:.03f}'.format(average_test_accuracy))
             print('*' * 50)

capstone_traffic_light_classifier/traffic_light_classifier.py

@@ -19,11 +19,13 @@ def __init__(self, x, targets, keep_prob, n_classes, learning_rate):
         self._inference     = None
         self._loss          = None
         self._train_step    = None
+        self._accuracy      = None
         self._summaries     = None
 
         self.inference
         self.loss
         self.train_step
+        self.accuracy
         # self.summaries # todo add these
 
     @property
@@ -68,3 +70,12 @@ def train_step(self):
             with tf.variable_scope('training'):
                 self._train_step = tf.train.AdamOptimizer(learning_rate=1e-4).minimize(self.loss)
         return self._train_step
+
+    @property
+    def accuracy(self):
+        if self._accuracy is None:
+            with tf.variable_scope('accuracy'):
+                correct_predictions = tf.equal(tf.argmax(self.inference, axis=1),
+                                               tf.argmax(tf.one_hot(self.targets, depth=self.n_classes), axis=1))
+                self._accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32))
+        return self._accuracy