added tests for normalize()

lene · Feb 28, 2016 · 44c217f · 44c217f
1 parent 4bc12e9
commit 44c217f
Show file tree

Hide file tree

Showing 2 changed files with 36 additions and 6 deletions.
diff --git a/nn_wtf/images_labels_data_set.py b/nn_wtf/images_labels_data_set.py
@@ -2,7 +2,7 @@
 
 import numpy
 
-__author__ = 'Lene Preuss <lp@sinnwerkstatt.com>'
+__author__ = 'Lene Preuss <lene.preuss@gmail.com>'
 
 
 class ImagesLabelsDataSet(DataSetBase):
@@ -33,11 +33,16 @@ def normalize(ndarray):
     :param ndarray:
     :return:
     """
-    ndarray = ndarray.astype(numpy.float32)
-    return numpy.multiply(ndarray, 1.0 / 255.0)
+    assert isinstance(ndarray, numpy.ndarray)
+    assert ndarray.dtype == numpy.uint8
+
+    return numpy.multiply(ndarray.astype(numpy.float32), 1.0/255.0)
 
 
 def invert(ndarray):
+    assert isinstance(ndarray, numpy.ndarray)
+    assert ndarray.dtype == numpy.float32
+
     return numpy.subtract(1.0, ndarray)
 
 

diff --git a/nn_wtf/tests/images_labels_data_set_test.py b/nn_wtf/tests/images_labels_data_set_test.py
@@ -1,11 +1,11 @@
-from nn_wtf.images_labels_data_set import ImagesLabelsDataSet
+from nn_wtf.images_labels_data_set import ImagesLabelsDataSet, normalize
 from .util import create_minimal_input_placeholder
 
 import numpy
 
 import unittest
 
-__author__ = 'Lene Preuss <lp@sinnwerkstatt.com>'
+__author__ = 'Lene Preuss <lene.preuss@gmail.com>'
 # pylint: disable=missing-docstring
 
 NUM_TRAINING_SAMPLES = 20
@@ -58,6 +58,21 @@ def test_next_batch_sets_epochs_completed(self):
         _, _ = data_set.next_batch(batch_size)
         self.assertEqual(1, data_set.epochs_completed)
 
+    def test_normalize_dtype(self):
+        data = create_empty_image_data()
+        normalized = normalize(data)
+        self.assertEqual(normalized.dtype, numpy.float32)
+
+    def test_normalize_range(self):
+        data = create_random_image_data(0, 255)
+        normalized = normalize(data)
+        self.assertLessEqual(normalized.max(), 1.)
+        self.assertGreaterEqual(normalized.min(), 0.)
+
+    def test_normalize_bad_input(self):
+        data = create_random_image_data(0, 255).astype(numpy.float32)
+        with self.assertRaises(AssertionError):
+            normalize(data)
 
 def _create_empty_data_set():
     images = create_empty_image_data()
@@ -66,7 +81,17 @@ def _create_empty_data_set():
 
 
 def create_empty_image_data():
-    buf = [0] * NUM_TRAINING_SAMPLES * IMAGE_WIDTH * IMAGE_HEIGHT
+    return image_data_from_list([0] * NUM_TRAINING_SAMPLES * IMAGE_WIDTH * IMAGE_HEIGHT)
+
+
+def create_random_image_data(min_val, max_val):
+    from random import randrange
+    return image_data_from_list(
+        [randrange(min_val, max_val+1) for _ in range(NUM_TRAINING_SAMPLES * IMAGE_WIDTH * IMAGE_HEIGHT)]
+    )
+
+
+def image_data_from_list(buf):
     data = numpy.fromiter(buf, dtype=numpy.uint8)
     return data.reshape(NUM_TRAINING_SAMPLES, IMAGE_WIDTH, IMAGE_HEIGHT, 1)