def test_find_best_architecture(self):
""" Find_best_architecture should return a single model, parameters, type and valid knn accuracy."""
np.random.seed(123)
num_timesteps = 100
num_channels = 2
num_samples_train = 5
num_samples_val = 3
X_train = np.random.rand(num_samples_train, num_timesteps,
num_channels)
y_train = to_categorical(np.array([0, 0, 1, 1, 1]))
X_val = np.random.rand(num_samples_val, num_timesteps, num_channels)
y_val = to_categorical(np.array([0, 1, 1]))
best_model, best_params, best_model_type, knn_acc = find_architecture.find_best_architecture(
X_train,
y_train,
X_val,
y_val,
verbose=False,
subset_size=10,
number_of_models=1,
nr_epochs=1)
assert hasattr(best_model, 'fit')
self.assertIsNotNone(best_params)
self.assertIsNotNone(best_model_type)
assert 1 >= knn_acc >= 0
def test_find_best_architecture_with_class_weights(self):
"""Model should not ignore tiny class with huge class weight. Note that this test is non-deterministic,
even though a seed was set. Note2 that this test is very slow, taking up 40% of all mcfly test time."""
tf.random.set_seed(1234) # Needs tensorflow API v2
X_train, y_train = _create_2_class_labeled_dataset(
1, 999) # very unbalanced
X_val, y_val = _create_2_class_labeled_dataset(1, 99)
X_test, y_test = _create_2_class_labeled_dataset(10, 10)
class_weight = {0: 2, 1: 0.002}
best_model, _, _, _ = find_architecture.find_best_architecture(
X_train,
y_train,
X_val,
y_val,
verbose=False,
subset_size=1000,
number_of_models=5,
nr_epochs=1,
model_type='CNN',
class_weight=class_weight)
probabilities = best_model.predict_proba(X_test)
predicted = probabilities.argmax(axis=1)
np.testing.assert_array_equal(predicted, y_test.argmax(axis=1))
def create_dummy_model():
""" Function to aid the tests on saving and loading a model"""
np.random.seed(123)
num_timesteps = 100
num_channels = 2
num_samples_train = 5
num_samples_val = 3
X_train = np.random.rand(
num_samples_train,
num_timesteps,
num_channels)
y_train = to_categorical(np.array([0, 0, 1, 1, 1]))
X_val = np.random.rand(num_samples_val, num_timesteps, num_channels)
y_val = to_categorical(np.array([0, 1, 1]))
best_model, best_params, best_model_type, knn_acc = find_architecture.find_best_architecture(
X_train, y_train, X_val, y_val, verbose=False, subset_size=10,
number_of_models=1, nr_epochs=1)
return(best_model)
def create_dummy_model():
""" Function to aid the tests on saving and loading a model"""
np.random.seed(123)
num_timesteps = 100
num_channels = 2
num_samples_train = 5
num_samples_val = 3
X_train = np.random.rand(
num_samples_train,
num_timesteps,
num_channels)
y_train = to_categorical(np.array([0, 0, 1, 1, 1]))
X_val = np.random.rand(num_samples_val, num_timesteps, num_channels)
y_val = to_categorical(np.array([0, 1, 1]))
best_model, best_params, best_model_type, knn_acc = find_architecture.find_best_architecture(
X_train, y_train, X_val, y_val, verbose=False, subset_size=10,
number_of_models=1, nr_epochs=1)
return(best_model)
def test_find_best_architecture(self):
""" Find_best_architecture should return a single model, parameters, type and valid knn accuracy."""
np.random.seed(123)
num_timesteps = 100
num_channels = 2
num_samples_train = 5
num_samples_val = 3
X_train = np.random.rand(
num_samples_train,
num_timesteps,
num_channels)
y_train = to_categorical(np.array([0, 0, 1, 1, 1]))
X_val = np.random.rand(num_samples_val, num_timesteps, num_channels)
y_val = to_categorical(np.array([0, 1, 1]))
best_model, best_params, best_model_type, knn_acc = find_architecture.find_best_architecture(
X_train, y_train, X_val, y_val, verbose=False, subset_size=10,
number_of_models=1, nr_epochs=1)
assert hasattr(best_model, 'fit')
self.assertIsNotNone(best_params)
self.assertIsNotNone(best_model_type)
assert 1 >= knn_acc >= 0