def test_default_config(self):
random_csv = create_random_csv()
target = 'D'
config = Config.make_default_config(
csv_path=random_csv,
target_name=target,
target_type=TargetType.BINARY_CLASSIFICATION,
train_ratio=0.9)
self.assertIsNotNone(config)
remove_random_csv()
def test_activation_for_regression_is_sigmoid(self):
config = self.make_config_for_type(TargetType.REGRESSION)
previous_layer = self.make_sample_layer()
layer = config.model_assembler.make_final_layer(previous_layer)
self.check_layer_properties(layer=layer,
activation="Sigmoid",
outputs=1,
operation='dense')
remove_random_csv()
def test_activation_for_multiclass_is_softmax(self):
config = self.make_config_for_type(
TargetType.MULTICLASS_CLASSIFICATION)
previous_layer = self.make_sample_layer()
layer = config.model_assembler.make_final_layer(previous_layer)
self.check_layer_properties(layer=layer,
activation="Softmax",
outputs=config.unique_classes,
operation='dense')
remove_random_csv()
def test_model_params_for_regression_classification(self):
config = self.make_config_for_type(TargetType.REGRESSION)
previous_layer = self.make_sample_layer()
layer = config.model_assembler.make_final_layer(previous_layer)
model = Model(inputs=previous_layer, outputs=layer)
model = config.model_assembler.compile_model(model)
self.check_model_parameters(model=model,
optimizer=keras.optimizers.Adam,
loss="mean_absolute_error",
metrics=[])
remove_random_csv()
def test_model_params_for_binary_classification(self):
config = self.make_config_for_type(TargetType.BINARY_CLASSIFICATION)
previous_layer = self.make_sample_layer()
layer = config.model_assembler.make_final_layer(previous_layer)
model = Model(inputs=previous_layer, outputs=layer)
model = config.model_assembler.compile_model(model)
self.check_model_parameters(model=model,
optimizer=keras.optimizers.Adam,
loss="binary_crossentropy",
metrics=["accuracy"])
remove_random_csv()
def test_custom_config(self):
random_csv = create_random_csv()
target = 'D'
processor = CustomProcessor()
assembler = CustomAssembler()
config = Config.make_custom_config(csv_path=random_csv,
target_name=target,
train_ratio=0.9,
target_processor=processor,
model_assembler=assembler)
self.assertIsNotNone(config)
remove_random_csv()
def test_create_and_remove_random_csv(self):
rows = 5
cols = 5
columns = 'ABCDE'
csv_path = create_random_csv(rows, cols, columns)
exists_csv = os.path.exists(csv_path)
self.assertEqual(exists_csv, True)
df = pd.read_csv(csv_path)
self.__check_dataframe_data(df, rows, cols, columns)
remove_random_csv()
exists_csv = os.path.exists(csv_path)
self.assertEqual(exists_csv, False)
def test_model_embedding_size(self):
random_csv = create_random_csv()
target = 'D'
config = Config.make_default_config(
csv_path=random_csv,
target_name=target,
target_type=TargetType.BINARY_CLASSIFICATION,
train_ratio=0.9)
network = EmbeddingNetwork(config)
for layer in network.model.layers:
if isinstance(layer, Embedding):
embedding_size = int(layer.embeddings.initial_value.shape[1])
self.assertEqual(
get_embedding_size(config.df[layer.name].nunique()),
embedding_size)
remove_random_csv()
def tearDown(self):
remove_random_csv()