def test_no_fully_connected_layers(self):
with self.test_session() as sess:
x = tf.placeholder(tf.float32, shape=[None, 3])
out = graph.fully_connected_layers([], x)
self.assertAllEqual(3, out.get_shape()[-1])
x_val = np.array([[3., 2., 1.], [5., 6., 87.]])
sess.run(tf.global_variables_initializer())
out_val = sess.run(out, feed_dict={x: x_val})
self.assertEquals((2, 3), out_val.shape)
self.assertAllClose(x_val, out_val)
def test_fully_connected_layers(self):
with self.test_session() as sess:
x = tf.placeholder(tf.float32, shape=[None, 3])
out = graph.fully_connected_layers([10, 20, 100, 1], x)
self.assertAllEqual(1, out.get_shape()[-1])
x_val = np.array([[3., 2., 1.], [5., 6., 87.]])
sess.run(tf.initialize_all_variables())
out_val = sess.run(out, feed_dict={x: x_val})
self.assertEquals((2, 1), out_val.shape)
self.assertAllClose([[1.597199],[50.472717]], out_val)
def test_fully_connected_doesnt_use_hidden_dim_as_layer_name(self):
tf.set_random_seed(0)
with self.test_session() as sess:
x = tf.placeholder(tf.float32, shape=[None, 3])
out = graph.fully_connected_layers([10, 10, 10, 10, 20, 1], x)
self.assertAllEqual(1, out.get_shape()[-1])
sess.run(tf.global_variables_initializer())
sess.run(out, feed_dict={
x: np.array([[1., 2., 3.], [4., 5., 6.]])})
x_val = np.array([[-3., 2., 1.], [5., 6., 87.]])
out_val = sess.run(out, feed_dict={x: x_val})
def test_fully_connected_doesnt_use_hidden_dim_as_layer_name(self):
tf.set_random_seed(0)
with self.test_session() as sess:
x = tf.placeholder(tf.float32, shape=[None, 3])
out = graph.fully_connected_layers([10, 10, 10, 10, 20, 1], x)
self.assertAllEqual(1, out.get_shape()[-1])
sess.run(tf.global_variables_initializer())
sess.run(out,
feed_dict={x: np.array([[1., 2., 3.], [4., 5., 6.]])})
x_val = np.array([[-3., 2., 1.], [5., 6., 87.]])
out_val = sess.run(out, feed_dict={x: x_val})
def test_fully_connected_layers(self):
tf.set_random_seed(0)
with self.test_session() as sess:
x = tf.placeholder(tf.float32, shape=[None, 3])
out = graph.fully_connected_layers([10, 20, 100, 1], x)
self.assertAllEqual(1, out.get_shape()[-1])
sess.run(tf.global_variables_initializer())
sess.run(out,
feed_dict={x: np.array([[1., 2., 3.], [4., 5., 6.]])})
x_val = np.array([[-3., 2., 1.], [5., 6., 87.]])
out_val = sess.run(out, feed_dict={x: x_val})
self.assertEqual((2, 1), out_val.shape)
self.assertAllClose([[0.0], [0.245194]], out_val)
def test_no_fully_connected_layers(self):
tf.set_random_seed(0)
with self.test_session() as sess:
x = tf.placeholder(tf.float32, shape=[None, 3])
out = graph.fully_connected_layers([], x)
self.assertAllEqual(3, out.get_shape()[-1])
sess.run(tf.global_variables_initializer())
sess.run(out, feed_dict={
x: np.array([[1., 2., 3.], [4., 5., 6.]])})
x_val = np.array([[3., 2., 1.], [5., 6., 87.]])
out_val = sess.run(out, feed_dict={x: x_val})
self.assertEqual((2, 3), out_val.shape)
self.assertAllClose(x_val, out_val)