def test_compile_with_conflicting_data_format_2(self):
for k in [keras, tf.keras]:
with self.subTest(k=k):
with tf.Graph().as_default(), tf.compat.v1.Session(
).as_default() as session:
input_x = k.layers.Input(shape=(28, 28, 3), name='l0')
input_y_1 = k.layers.Conv2D(filters=4,
kernel_size=(3, 3),
data_format='channels_last',
name='l1')(input_x)
input_y_2 = k.layers.Conv2D(filters=4,
kernel_size=(3, 3),
data_format='channels_first',
name='l2')(input_x)
model = KerasModel(model=k.Model(
inputs=[input_x], outputs=[input_y_1, input_y_2]),
session=session)
compiled = compiler.compile_source(
source=model,
config=Config(input_nodes=[NodeSpec(layer_name='l0')]))
self.assertEqual(len(compiled.inputs), 1)
self.assertIs(compiled.inputs[0].tensor, model.model.input)
self.assertIsNone(compiled.inputs[0].data_format)
self.assertEqual(len(compiled.outputs), 2)
self.assertIs(compiled.outputs[0], model.model.outputs[0])
self.assertIs(compiled.outputs[1], model.model.outputs[1])
self.assertIs(compiled.session, session)
def test_compile_simple_with_data_format(self):
for k in [keras, tf.keras]:
for (keras_data_format, data_format) in [
('channels_first', DataFormat.CHANNELS_FIRST),
('channels_last', DataFormat.CHANNELS_LAST)
]:
with self.subTest(k=k, data_format=keras_data_format):
with tf.Graph().as_default(), tf.compat.v1.Session(
).as_default() as session:
model = KerasModel(model=k.Sequential([
k.layers.Conv2D(filters=4,
kernel_size=(3, 3),
data_format=keras_data_format,
input_shape=(28, 28, 3))
]),
session=session)
compiled = compiler.compile_source(source=model,
config=Config())
self.assertEqual(len(compiled.inputs), 1)
self.assertIs(compiled.inputs[0].tensor, model.model.input)
self.assertEqual(compiled.inputs[0].data_format,
data_format)
self.assertEqual(len(compiled.outputs), 1)
self.assertIs(compiled.outputs[0], model.model.output)
self.assertIs(compiled.session, session)
def test_compile_with_input_layer_data_format(self):
for k in [keras, tf.keras]:
for (keras_data_format, data_format) in [
('channels_first', DataFormat.CHANNELS_FIRST),
('channels_last', DataFormat.CHANNELS_LAST)
]:
with self.subTest(k=k, data_format=keras_data_format):
with tf.Graph().as_default(), tf.compat.v1.Session(
).as_default() as session:
input_x = k.layers.Input(shape=(28, 28, 3), name='l0')
output_y = k.layers.Conv2D(
filters=4,
kernel_size=(3, 3),
data_format=keras_data_format,
name='l1')(input_x)
model = KerasModel(model=k.Model(inputs=[input_x],
outputs=[output_y]),
session=session)
compiled = compiler.compile_source(
source=model,
config=Config(input_nodes=[NodeSpec(layer_name='l0')]))
self.assertEqual(len(compiled.inputs), 1)
self.assertIs(compiled.inputs[0].tensor, model.model.input)
self.assertEqual(compiled.inputs[0].data_format,
data_format)
self.assertEqual(len(compiled.outputs), 1)
self.assertIs(compiled.outputs[0], model.model.output)
self.assertIs(compiled.session, session)
def test_compile_with_output_name(self):
for k in [keras, tf.keras]:
with self.subTest(k=k):
with tf.Graph().as_default(), tf.compat.v1.Session(
).as_default() as session:
layer_1 = k.layers.Dense(units=8,
name='l1',
input_shape=[16])
layer_2 = k.layers.Dense(units=4, name='l2')
layer_3 = k.layers.Dense(units=2, name='l3')
model = KerasModel(model=k.Sequential(
[layer_1, layer_2, layer_3]),
session=session)
compiled = compiler.compile_source(
source=model,
config=Config(output_nodes=[NodeSpec(layer_name='l2')]))
self.assertEqual(len(compiled.inputs), 1)
self.assertIs(compiled.inputs[0].tensor, model.model.input)
self.assertIsNone(compiled.inputs[0].data_format)
self.assertEqual(len(compiled.outputs), 1)
self.assertIs(compiled.outputs[0], layer_2.output)
self.assertIs(compiled.session, session)
def test_from_env_full(self):
config = Config.from_env({'INPUT_LAYER_NAMES': 'abc,def:3',
'OUTPUT_LAYER_NAMES': 'ghi:2,jkl'})
self.assertEqual(config, Config(input_nodes=[NodeSpec(layer_name='abc'),
NodeSpec(layer_name='def', node_index=3)],
output_nodes=[NodeSpec(layer_name='ghi', node_index=2),
NodeSpec(layer_name='jkl')]))
def test_from_json_full(self):
config = Config.from_json({'input_layer_names': ['abc', 'def:3'],
'output_layer_names': ['ghi:2', 'jkl']})
self.assertEqual(config, Config(input_nodes=[NodeSpec(layer_name='abc'),
NodeSpec(layer_name='def', node_index=3)],
output_nodes=[NodeSpec(layer_name='ghi', node_index=2),
NodeSpec(layer_name='jkl')]))
def test_from_env_input_names_only(self):
config = Config.from_env({'input_layer_names': 'abc,def:3'})
self.assertEqual(
config,
Config(input_nodes=[
NodeSpec(layer_name='abc'),
NodeSpec(layer_name='def', node_index=3)
],
output_nodes=None))
def test_from_json_output_names_only(self):
config = Config.from_json({'output_layer_names': ['abc', 'def:3']})
self.assertEqual(
config,
Config(input_nodes=None,
output_nodes=[
NodeSpec(layer_name='abc'),
NodeSpec(layer_name='def', node_index=3)
]))
def test_from_env_output_names_only(self):
config = Config.from_env({'OUTPUT_LAYER_NAMES': 'abc,def:3'})
self.assertEqual(
config,
Config(input_nodes=None,
output_nodes=[
NodeSpec(layer_name='abc'),
NodeSpec(layer_name='def', node_index=3)
]))
def test_compile_simple(self):
with tf.Graph().as_default(), tf.compat.v1.Session().as_default() as session:
model = KerasModel(model=keras.Sequential([keras.layers.Dense(units=4, input_shape=[8])]), session=session)
compiled = compiler.compile_source(source=model, config=Config())
self.assertEqual(len(compiled.inputs), 1)
self.assertIs(compiled.inputs[0].tensor, model.model.input)
self.assertIsNone(compiled.inputs[0].data_format)
self.assertEqual(len(compiled.outputs), 1)
self.assertIs(compiled.outputs[0], model.model.output)
self.assertIs(compiled.session, session)
def test_compile_with_input_name_to_input_layer(self):
with tf.Graph().as_default(), tf.compat.v1.Session().as_default() as session:
input_tensor = keras.layers.Input(shape=[16], name='l0')
output_tensor = keras.layers.Dense(units=8)(input_tensor)
model = KerasModel(model=keras.Model(inputs=[input_tensor], outputs=[output_tensor]), session=session)
compiled = compiler.compile_source(source=model, config=Config(input_nodes=[NodeSpec(layer_name='l0')]))
self.assertEqual(len(compiled.inputs), 1)
self.assertIs(compiled.inputs[0].tensor, model.model.input)
self.assertIsNone(compiled.inputs[0].data_format)
self.assertEqual(len(compiled.outputs), 1)
self.assertIs(compiled.outputs[0], model.model.output)
self.assertIs(compiled.session, session)
def test_compile_with_output_name(self):
with tf.Graph().as_default(), tf.compat.v1.Session().as_default() as session:
origin_model = keras.Sequential()
origin_model.add(keras.layers.Dense(units=8, name='l1', input_shape=(16,)))
origin_model.add(keras.layers.Dense(units=4, name='l2'))
origin_model.add(keras.layers.Dense(units=2, name='l3'))
model = KerasModel(model=origin_model, session=session)
compiled = compiler.compile_source(source=model, config=Config(output_nodes=[NodeSpec(layer_name='l2')]))
self.assertEqual(len(compiled.inputs), 1)
self.assertIs(compiled.inputs[0].tensor, model.model.input)
self.assertIsNone(compiled.inputs[0].data_format)
self.assertEqual(len(compiled.outputs), 1)
self.assertIs(compiled.outputs[0], origin_model.layers[1].output)
self.assertIs(compiled.session, session)
def test_from_env_minimal(self):
config = Config.from_env({})
self.assertEqual(config, Config(input_nodes=None, output_nodes=None))
