got rid of the fake_data flag, it's rubbish

nn_wtf/images_labels_data_set.py

@@ -5,26 +5,22 @@
 
 class ImagesLabelsDataSet:
 
-    def __init__(self, images, labels, fake_data=False, one_hot=False):
+    def __init__(self, images, labels):
         """Construct a DataSet. one_hot arg is used only if fake_data is true.
 
         Args:
           images: 4D numpy.ndarray of shape (num images, image height, image width, image depth)
           labels: 1D numpy.ndarray of shape (num images)
         """
 
-        if fake_data:
-            self._num_examples = 10000
-            self.one_hot = one_hot
-        else:
-            _check_constructor_arguments_valid(images, labels)
+        _check_constructor_arguments_valid(images, labels)
 
-            self._num_examples = images.shape[0]
+        self._num_examples = images.shape[0]
 
-            # Convert shape from [num examples, rows, columns, depth] to [num examples, rows*columns]
-            # TODO: assumes depth == 1
-            images = images.reshape(images.shape[0], images.shape[1] * images.shape[2])
-            images = normalize(images)
+        # Convert shape from [num examples, rows, columns, depth] to [num examples, rows*columns]
+        # TODO: assumes depth == 1
+        images = images.reshape(images.shape[0], images.shape[1] * images.shape[2])
+        images = normalize(images)
 
         self._images = images
         self._labels = labels
@@ -47,11 +43,8 @@ def num_examples(self):
     def epochs_completed(self):
         return self._epochs_completed
 
-    def next_batch(self, batch_size, fake_data=False):
+    def next_batch(self, batch_size):
         """Return the next `batch_size` examples from this data set."""
-        if fake_data:
-            return self._fake_batch(batch_size)
-
         return self._next_batch_in_epoch(batch_size)
 
     def _next_batch_in_epoch(self, batch_size):
@@ -75,14 +68,6 @@ def _shuffle_data(self):
         self._images = self._images[perm]
         self._labels = self._labels[perm]
 
-    def _fake_batch(self, batch_size):
-      fake_image = [1] * 784
-      if self.one_hot:
-        fake_label = [1] + [0] * 9
-      else:
-        fake_label = 0
-      return [fake_image for _ in range(batch_size)], [fake_label for _ in range(batch_size)]
-
 
 def normalize(ndarray):
     """Transform a ndarray that contains uint8 values to floats between 0. and 1.

nn_wtf/input_data.py

@@ -146,10 +146,3 @@ def read_data_sets(train_dir, one_hot=False):
         ImagesLabelsDataSet(test_images, test_labels)
     )
 
-
-def fake_data_sets(one_hot):
-    return DataSets(
-        ImagesLabelsDataSet([], [], fake_data=True, one_hot=one_hot),
-        ImagesLabelsDataSet([], [], fake_data=True, one_hot=one_hot),
-        ImagesLabelsDataSet([], [], fake_data=True, one_hot=one_hot)
-    )

nn_wtf/mnist_graph.py

@@ -32,7 +32,6 @@ def __init__(
             self.hidden += (hidden3,)
         self.batch_size = batch_size
         self.train_dir = ensure_is_dir(train_dir)
-        self.fake_data = False
 
         self._build_graph()
         self._setup_summaries()
@@ -45,15 +44,7 @@ def train(self, data_sets, max_steps, precision=None, steps_between_checks=100):
         for self.step in range(max_steps):
             start_time = time.time()
 
-            # Fill a feed dictionary with the actual set of images and labels for this particular
-            # training step.
-            feed_dict = self.fill_feed_dict(data_sets.train)
-
-            # Run one step of the model.  The return values are the activations from the `train_op`
-            # (which is discarded) and the `loss` Op. To inspect the values of your Ops or
-            # variables, you may include them in the list passed to session.run() and the value
-            # tensors will be returned in the tuple from the call.
-            _, loss_value = self.session.run([self.train_op, self.loss], feed_dict=feed_dict)
+            feed_dict, loss_value = self.run_training_steps(data_sets)
 
             duration = time.time() - start_time
 
@@ -70,6 +61,17 @@ def train(self, data_sets, max_steps, precision=None, steps_between_checks=100):
                 self.saver.save(self.session, save_path=self.train_dir, global_step=self.step)
                 self.print_evaluations(data_sets)
 
+    def run_training_steps(self, data_sets):
+        # Fill a feed dictionary with the actual set of images and labels for this particular
+        # training step.
+        feed_dict = self.fill_feed_dict(data_sets.train)
+        # Run one step of the model.  The return values are the activations from the `train_op`
+        # (which is discarded) and the `loss` Op. To inspect the values of your Ops or
+        # variables, you may include them in the list passed to session.run() and the value
+        # tensors will be returned in the tuple from the call.
+        _, loss_value = self.session.run([self.train_op, self.loss], feed_dict=feed_dict)
+        return feed_dict, loss_value
+
     def print_evaluations(self, data_sets):
         if self.verbose: print('Training Data Eval:')
         self.print_eval(data_sets.train)
@@ -106,7 +108,7 @@ def fill_feed_dict(self, data_set):
           feed_dict: The feed dictionary mapping from placeholders to values.
         """
         # Create the feed_dict for the placeholders filled with the next `batch size ` examples.
-        images_feed, labels_feed = data_set.next_batch(self.batch_size, self.fake_data)
+        images_feed, labels_feed = data_set.next_batch(self.batch_size)
         feed_dict = {
             self.images_placeholder: images_feed,
             self.labels_placeholder: labels_feed,

nn_wtf/tests/images_labels_data_set_test.py

@@ -17,10 +17,6 @@ class ImagesLabelsDataSetTest(unittest.TestCase):
     def test_init_without_fake_data_runs(self):
         _create_empty_data_set()
 
-    def test_init_with_fake_data_runs(self):
-        images = create_minimal_input_placeholder()
-        ImagesLabelsDataSet(images, images, fake_data=True)
-
     def test_init_with_different_label_size_fails(self):
         images = create_empty_image_data()
         labels = create_empty_label_dataof_size(NUM_TRAINING_SAMPLES+1)

nn_wtf/tests/predictor_test.py

@@ -0,0 +1,12 @@
+from nn_wtf.neural_network_graph import NeuralNetworkGraph
+
+from .util import MINIMAL_INPUT_SIZE, MINIMAL_OUTPUT_SIZE, MINIMAL_LAYER_GEOMETRY, MINIMAL_BATCH_SIZE
+from .util import create_minimal_input_placeholder
+
+import tensorflow as tf
+
+import unittest
+
+__author__ = 'Lene Preuss <lene.preuss@gmail.com>'
+
+class NeuralNetworkGraphTest(unittest.TestCase):