add basic test for layers; bump version for last planned release supporting tensorflow 1

Author: mhuen

test/test_layers.py

@@ -0,0 +1,98 @@
+#!/usr/local/bin/python
+from __future__ import division, print_function
+
+import unittest
+import numpy as np
+import tensorflow as tf
+from itertools import product
+
+from tfscripts import layers
+
+
+class TestConvModule(unittest.TestCase):
+
+    def setUp(self):
+        self.random_state = np.random.RandomState(42)
+
+    def test_new_conv_nd_layers(self):
+        """Test new_conv_nd_layers
+        """
+        data = tf.constant(self.random_state.normal(size=[3, 4, 4, 4, 2]),
+                           dtype=tf.float32)
+
+        shapes = [[3, 3, 3, 2, 3], [3, 3, 3, 3, 7], [3, 3, 3, 7, 1]]
+        weights_list = [tf.constant(self.random_state.normal(size=shape),
+                                    dtype=tf.float32) for shape in shapes]
+
+        shapes = [[3], [7], [1]]
+        biases_list = [tf.constant(self.random_state.normal(size=shape),
+                                   dtype=tf.float32) for shape in shapes]
+
+        pooling_strides_list = [[1, 1, 1, 1, 1],
+                                [1, 2, 2, 2, 1],
+                                [1, 2, 2, 2, 1]]
+        pooling_ksize_list = [[1, 1, 1, 1, 1],
+                              [1, 2, 2, 2, 1],
+                              [1, 2, 2, 2, 1]]
+        pooling_type_list = [None, 'max', 'max']
+        activation_list = ['elu', 'relu', '']
+        filter_size_list = [[3, 3, 3], [2, 0, 3], [3, 3, 3]]
+        num_filters_list = [3, 7, 1]
+        method_list = ['convolution', 'hex_convolution', 'convolution']
+
+        layer, weights, biases = layers.new_conv_nd_layers(
+                                data,
+                                filter_size_list=filter_size_list,
+                                num_filters_list=num_filters_list,
+                                pooling_type_list=pooling_type_list,
+                                pooling_strides_list=pooling_strides_list,
+                                pooling_ksize_list=pooling_ksize_list,
+                                activation_list=activation_list,
+                                method_list=method_list,
+                                weights_list=weights_list,
+                                biases_list=biases_list,
+                                verbose=False,
+                                )
+
+        result_true = [[[[[1.5821583]]]],
+                       [[[[1.8506659]]]],
+                       [[[[1.6045672]]]]]
+
+        with tf.Session() as sess:
+            result = sess.run(layer[-1])
+
+        self.assertTrue(np.allclose(result_true, result))
+
+    def test_new_fc_layers(self):
+        """Test new_fc_layers
+        """
+        data = tf.constant(self.random_state.normal(size=[3, 7]),
+                           dtype=tf.float32)
+
+        shapes = [[7, 3], [3, 7], [7, 1]]
+        weights_list = [tf.constant(self.random_state.normal(size=shape),
+                                    dtype=tf.float32) for shape in shapes]
+
+        shapes = [[3], [7], [1]]
+        biases_list = [tf.constant(self.random_state.normal(size=shape),
+                                   dtype=tf.float32) for shape in shapes]
+
+        activation_list = ['elu', 'relu', '']
+        fc_sizes = [3, 7, 1]
+
+        layer, weights, biases = layers.new_fc_layers(
+                                data,
+                                fc_sizes=fc_sizes,
+                                activation_list=activation_list,
+                                weights_list=weights_list,
+                                biases_list=biases_list,
+                                verbose=False,
+                                )
+        result_true = [[0.06657974],
+                       [-0.3617597],
+                       [0.00241985]]
+
+        with tf.Session() as sess:
+            result = sess.run(layer[-1])
+
+        self.assertTrue(np.allclose(result_true, result))

tfscripts/__about__.py

@@ -1,4 +1,4 @@
-__version__ = "0.1.0-dev"
+__version__ = "0.2.0"
 __author__ = "Mirco Huennefeld"
 __author_email__ = "mirco.huennefeld@tu-dortmund.de"
 __description__ = "Collection of TF functions and helpful additions"