def test_failed_to_repeat_network(self):
network = layers.join(
layers.Input(10),
layers.Relu(5),
)
network.create_variables()
relu = network.layers[1]
error_message = "input shape is incompatible with the output shape"
with self.assertRaisesRegexp(LayerConnectionError, error_message):
layers.repeat(relu, n=4)
def test_repeat_network(self):
block = layers.join(
layers.Convolution((3, 3, 32)),
layers.Relu(),
layers.BatchNorm(),
)
network = layers.repeat(block, n=5)
self.assertEqual(len(network), 15)
self.assertShapesEqual(network.output_shape, (None, None, None, 32))
def test_repeat_with_name_patterns(self):
network = layers.repeat(layers.Relu(10, name='rl{}'), n=4)
layer_names = [layer.name for layer in network.layers]
self.assertSequenceEqual(layer_names, ['rl1', 'rl2', 'rl3', 'rl4'])
def test_repeat_once(self):
input_layer = layers.Relu(10)
output_layer = layers.repeat(input_layer, n=1)
self.assertIs(output_layer, input_layer)
def test_wrong_number_of_repeats(self):
error_message = "parameter should be a positive integer"
for wrong_value in (0, 1.5, 9. / 3.):
with self.assertRaisesRegexp(ValueError, error_message):
layers.repeat(layers.Relu(10), n=wrong_value)
def test_repeat_layer(self):
network = layers.repeat(layers.Relu(10), n=5)
self.assertEqual(len(network), 5)
self.assertShapesEqual(network.output_shape, (None, 10))