def tearDown(self):
for file in os.listdir(self.__model_dir):
os.remove(
os.path.join(self.__model_dir,
file)) if not file.endswith(".hdf5") else None
shutil.rmtree(self.__resource_tmp)
Undertest.reset()
def test_load_model_and_weights(self):
network_old = Undertest.get_network((2, 20), self.__hyperparameters)
weights_old = network_old._Network__model.get_weights()
network_new = Undertest.load_model_and_weights(
"{}/model.hdf5".format(self.__model_dir),
"{}/weights.hdf5".format(self.__weights_dir),
self.__hyperparameters)
weights_new = network_new._Network__model.get_weights()
self.assertFalse(np.array_equal(weights_old, weights_new))
def test_exception_on_creating_an_unknown_network(self):
self.__hyperparameters.network_type = "unknown"
try:
Undertest.get_network((2, 20), self.__hyperparameters)
except Exception as e:
self.assertTrue(isinstance(e, ValueError))
self.assertTrue("Unknown network type" in str(e))
else:
self.fail("Should have thrown exception")
def test_save_model_and_weights(self):
model_filepath = "{}/{}".format(self.__model_dir, "model.hdf5")
weights_filepath = "{}/{}".format(self.__weights_dir, "weights.hdf5")
Undertest.save_model_and_weights(
model_filepath,
weights_filepath,
"{}/{}".format(self.__resource_tmp, "model_combined.hdf5"),
)
self.assertEqual(
(2, 20),
Undertest.get_from_model(model_filepath,
self.__hyperparameters).input_shape)
def test_exception_on_resume_with_no_previous_training_log(self):
self.__hyperparameters.epochs = 2
network = Undertest.get_network((2, 20), self.__hyperparameters)
model_filepath = "{}/model.hdf5".format(self.__resource_tmp)
weights_filepath = "{}/weights.hdf5".format(self.__resource_tmp)
with open(self.__train_data, "rb") as file:
train_data = pickle.load(file)
with open(self.__labels, "rb") as file:
labels = pickle.load(file)
_, _ = network.fit_and_get_history(
train_data,
labels,
model_filepath,
weights_filepath,
self.__resource_tmp,
"{}/training_1.log".format(self.__resource_tmp),
False,
)
try:
network.fit_and_get_history(
train_data,
labels,
model_filepath,
weights_filepath,
self.__resource_tmp,
"{}/training_2.log".format(self.__resource_tmp),
True,
)
except Exception as e:
self.assertTrue(isinstance(e, AssertionError))
self.assertTrue(
"does not exist and is required by training resumption" in str(
e))
else:
self.fail("Should have thrown exception")
def test_exception_on_resume_with_no_extra_epochs(self):
self.__hyperparameters.epochs = 2
network = Undertest.get_network((2, 20), self.__hyperparameters)
model_filepath = "{}/model.hdf5".format(self.__resource_tmp)
weights_filepath = "{}/weights.hdf5".format(self.__resource_tmp)
with open(self.__train_data, "rb") as file:
train_data = pickle.load(file)
with open(self.__labels, "rb") as file:
labels = pickle.load(file)
_, _ = network.fit_and_get_history(
train_data,
labels,
model_filepath,
weights_filepath,
self.__resource_tmp,
"{}/training.log".format(self.__resource_tmp),
False,
)
try:
network.fit_and_get_history(
train_data,
labels,
model_filepath,
weights_filepath,
self.__resource_tmp,
"{}/training.log".format(self.__resource_tmp),
True,
)
except Exception as e:
self.assertTrue(isinstance(e, AssertionError))
self.assertEqual(
"The existing model has been trained for 2 epochs. Make sure the total epochs are larger than 2",
str(e))
else:
self.fail("Should have thrown exception")
def test_resume_and_get_history(self):
self.__hyperparameters.epochs = 2
network = Undertest.get_network((2, 20), self.__hyperparameters)
model_filepath = "{}/model.hdf5".format(self.__resource_tmp)
weights_filepath = "{}/weights.hdf5".format(self.__resource_tmp)
with open(self.__train_data, "rb") as file:
train_data = pickle.load(file)
with open(self.__labels, "rb") as file:
labels = pickle.load(file)
_, _ = network.fit_and_get_history(
train_data,
labels,
model_filepath,
weights_filepath,
self.__resource_tmp,
"{}/training.log".format(self.__resource_tmp),
False,
)
self.__hyperparameters.epochs = 3
val_loss, val_acc = network.fit_and_get_history(
train_data,
labels,
model_filepath,
weights_filepath,
self.__resource_tmp,
"{}/training.log".format(self.__resource_tmp),
True,
)
self.assertEqual(list, type(val_loss))
self.assertEqual(list, type(val_acc))
def test_get_predictions(self):
network = Undertest.get_from_model(
"{}/model.hdf5".format(self.__model_dir), self.__hyperparameters)
with open(self.__train_data, "rb") as file:
train_data = pickle.load(file)
self.assertEqual(
(11431, 1),
network.get_predictions(
train_data,
"{}/weights.hdf5".format(self.__weights_dir)).shape)
def test_simple_fit_with_generator(self):
self.__hyperparameters.epochs = 3
with h5py.File(self.__training_dump, "r") as hf:
val_loss, val_acc = Undertest.simple_fit_with_generator(
(2, 20),
hf["train_data"],
hf["labels"],
self.__hyperparameters,
)
self.assertEqual(list, type(val_loss))
self.assertEqual(list, type(val_acc))
self.assertTrue(len(val_loss) == self.__hyperparameters.epochs)
self.assertTrue(len(val_acc) == self.__hyperparameters.epochs)
def test_simple_fit(self):
with open(self.__train_data, "rb") as file:
train_data = pickle.load(file)
with open(self.__labels, "rb") as file:
labels = pickle.load(file)
val_loss, val_acc = Undertest.simple_fit(
(2, 20),
train_data,
labels,
self.__hyperparameters,
)
self.assertEqual(list, type(val_loss))
self.assertEqual(list, type(val_acc))
def test_fit_with_generator(self):
self.__hyperparameters.epochs = 3
network = Undertest.get_network((2, 20), self.__hyperparameters)
model_filepath = "{}/model.hdf5".format(self.__resource_tmp)
weights_filepath = "{}/weights.hdf5".format(self.__resource_tmp)
with h5py.File(self.__training_dump, "r") as hf:
val_loss, val_acc = network.fit_with_generator(
hf["train_data"],
hf["labels"],
model_filepath,
weights_filepath,
self.__resource_tmp,
"training.log",
False,
)
self.assertEqual(list, type(val_loss))
self.assertEqual(list, type(val_acc))
self.assertTrue(len(val_loss) == self.__hyperparameters.epochs)
self.assertTrue(len(val_acc) == self.__hyperparameters.epochs)
def test_throw_exception_on_fit_with_generator(self, mock_fit):
self.__hyperparameters.epochs = 3
network = Undertest.get_network((2, 20), self.__hyperparameters)
model_filepath = "{}/model.hdf5".format(self.__resource_tmp)
weights_filepath = "{}/weights.hdf5".format(self.__resource_tmp)
with h5py.File(self.__training_dump, "r") as hf:
try:
network.fit_with_generator(
hf["train_data"],
hf["labels"],
model_filepath,
weights_filepath,
self.__resource_tmp,
"training.log",
False,
)
except Exception as e:
self.assertTrue(mock_fit.called)
self.assertTrue(isinstance(e, TerminalException))
self.assertTrue("interrupted" in str(e))
else:
self.fail("Should have thrown exception")
def test_throw_exception_on_fit_and_get_history(self, mock_fit):
try:
network = Undertest.get_network((2, 20), self.__hyperparameters)
model_filepath = "{}/model.hdf5".format(self.__resource_tmp)
weights_filepath = "{}/weights.hdf5".format(self.__resource_tmp)
with open(self.__train_data, "rb") as file:
train_data = pickle.load(file)
with open(self.__labels, "rb") as file:
labels = pickle.load(file)
network.fit_and_get_history(
train_data,
labels,
model_filepath,
weights_filepath,
self.__resource_tmp,
"training.log",
False,
)
except Exception as e:
self.assertTrue(mock_fit.called)
self.assertTrue(isinstance(e, TerminalException))
self.assertTrue("interrupted" in str(e))
else:
self.fail("Should have thrown exception")
def test_summary(self):
network = Undertest.get_network((2, 20), self.__hyperparameters)
self.assertTrue(network.summary is None) # Why this is None
def test_create_conv_1d_network(self):
self.__hyperparameters.network_type = "conv_1d"
network = Undertest.get_network((2, 20), self.__hyperparameters)
self.assertEqual("conv_1d", network.n_type)
def test_create_bi_lstm_network(self):
self.__hyperparameters.network_type = "bi_lstm"
network = Undertest.get_network((2, 20), self.__hyperparameters)
self.assertEqual("bi_lstm", network.n_type)
def test_input_shape(self):
network = Undertest.get_network((2, 20), self.__hyperparameters)
self.assertEqual((2, 20), network.input_shape)
def test_get_from_model(self):
model_filepath = "{}/{}".format(self.__model_dir, "model.hdf5")
network = Undertest.get_from_model(model_filepath,
self.__hyperparameters)
self.assertEqual((2, 20), network.input_shape)
self.assertEqual("unknown", network.n_type)
def test_layers(self):
network = Undertest.get_network((2, 20), self.__hyperparameters)
self.assertEqual(16, len(network.layers))