def test_automl_early_stopping_callback(tmpdir):
# TODO: fix this unit test
# Given
hp_repository = InMemoryHyperparamsRepository(cache_folder=str(tmpdir))
n_epochs = 60
auto_ml = AutoML(
pipeline=Pipeline([
FitTransformCallbackStep().set_name('callback'),
MultiplyByN(2).set_hyperparams_space(
HyperparameterSpace({'multiply_by': FixedHyperparameter(2)})),
NumpyReshape(new_shape=(-1, 1)),
linear_model.LinearRegression()
]),
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
validation_splitter=ValidationSplitter(0.20),
scoring_callback=ScoringCallback(mean_squared_error,
higher_score_is_better=False),
callbacks=[
MetricCallback('mse',
metric_function=mean_squared_error,
higher_score_is_better=False),
],
n_trials=1,
refit_trial=True,
epochs=n_epochs,
hyperparams_repository=hp_repository)
# When
data_inputs = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
expected_outputs = data_inputs * 2
auto_ml = auto_ml.fit(data_inputs=data_inputs,
expected_outputs=expected_outputs)
# Then
p = auto_ml.get_best_model()
def test_automl_should_shallow_copy_data_before_each_epoch():
# see issue #332 https://github.com/Neuraxio/Neuraxle/issues/332
data_inputs = np.random.randint(0, 100, (100, 3))
expected_outputs = np.random.randint(0, 3, 100)
from sklearn.preprocessing import StandardScaler
p = Pipeline([
SKLearnWrapper(StandardScaler()),
SKLearnWrapper(LinearSVC(),
HyperparameterSpace({'C': RandInt(0, 10000)})),
])
auto_ml = AutoML(p,
validation_splitter=ValidationSplitter(0.20),
refit_trial=True,
n_trials=10,
epochs=10,
cache_folder_when_no_handle='cache',
scoring_callback=ScoringCallback(
mean_squared_error, higher_score_is_better=False),
callbacks=[
MetricCallback('mse',
metric_function=mean_squared_error,
higher_score_is_better=False)
],
hyperparams_repository=InMemoryHyperparamsRepository(
cache_folder='cache'),
continue_loop_on_error=False)
random_search = auto_ml.fit(data_inputs, expected_outputs)
best_model = random_search.get_best_model()
assert isinstance(best_model, Pipeline)
def _test_trial_scores(
expected_output_mult, pipeline,
hyperparams_optimizer: BaseHyperparameterSelectionStrategy,
tmpdir: str):
hp_repository: InMemoryHyperparamsRepository = InMemoryHyperparamsRepository(
cache_folder=str(tmpdir))
n_epochs = 1
n_trials = 20
auto_ml: AutoML = AutoML(
pipeline=pipeline,
hyperparams_optimizer=hyperparams_optimizer,
validation_splitter=ValidationSplitter(0.5),
scoring_callback=ScoringCallback(mean_squared_error,
higher_score_is_better=False),
callbacks=[
MetricCallback('mse',
metric_function=mean_squared_error,
higher_score_is_better=False),
],
n_trials=n_trials,
refit_trial=True,
epochs=n_epochs,
hyperparams_repository=hp_repository)
# When
data_inputs = np.array([0, 0])
expected_outputs = expected_output_mult * np.ones_like(data_inputs)
auto_ml.fit(data_inputs=data_inputs, expected_outputs=expected_outputs)
# Then
trials: Trials = hp_repository.load_all_trials(status=TRIAL_STATUS.SUCCESS)
validation_scores = [t.get_validation_score() for t in trials]
return validation_scores
def test_logger_automl(self, tmpdir):
# Given
context = ExecutionContext()
self.tmpdir = str(tmpdir)
hp_repository = HyperparamsJSONRepository(cache_folder=self.tmpdir)
n_epochs = 2
n_trials = 4
auto_ml = AutoML(
pipeline=Pipeline([
MultiplyByN(2).set_hyperparams_space(
HyperparameterSpace(
{'multiply_by': FixedHyperparameter(2)})),
NumpyReshape(new_shape=(-1, 1)),
LoggingStep()
]),
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(
),
validation_splitter=ValidationSplitter(0.20),
scoring_callback=ScoringCallback(mean_squared_error,
higher_score_is_better=False),
n_trials=n_trials,
refit_trial=True,
epochs=n_epochs,
hyperparams_repository=hp_repository,
continue_loop_on_error=False)
# When
data_container = DataContainer(
data_inputs=np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]),
expected_outputs=np.array([10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0]))
auto_ml.handle_fit(data_container, context)
# Then
file_paths = [
os.path.join(hp_repository.cache_folder, f"trial_{i}.log")
for i in range(n_trials)
]
assert len(file_paths) == n_trials
for f in file_paths:
assert os.path.exists(f)
for f in file_paths:
with open(f, 'r') as f:
log = f.readlines()
assert len(log) == 36
def test_automl_savebestmodel_callback(tmpdir):
# Given
hp_repository = HyperparamsJSONRepository(cache_folder=str('caching'))
validation_splitter = ValidationSplitter(0.20)
auto_ml = AutoML(
pipeline=Pipeline([
MultiplyByN(2).set_hyperparams_space(HyperparameterSpace({
'multiply_by': FixedHyperparameter(2)
})),
NumpyReshape(new_shape=(-1, 1)),
linear_model.LinearRegression()
]),
validation_splitter=validation_splitter,
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
scoring_callback=ScoringCallback(mean_squared_error, higher_score_is_better=False),
callbacks=[
BestModelCheckpoint()
],
n_trials=1,
epochs=10,
refit_trial=False,
print_func=print,
hyperparams_repository=hp_repository,
continue_loop_on_error=False
)
data_inputs = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
expected_outputs = data_inputs * 4
# When
auto_ml.fit(data_inputs=data_inputs, expected_outputs=expected_outputs)
#Then
trials: Trials = hp_repository.load_all_trials()
best_trial = trials.get_best_trial()
best_trial_score = best_trial.get_validation_score()
best_trial.cache_folder = hp_repository.cache_folder
best_model = best_trial.get_model('best')
_, _, valid_inputs, valid_outputs = ValidationSplitter(0.20).split(data_inputs, expected_outputs)
predicted_output = best_model.predict(valid_inputs)
score = mean_squared_error(valid_outputs, predicted_output)
assert best_trial_score == score
def _test_within_auto_ml_loop(tmpdir, pipeline):
X_train = np.random.random((25, 50)).astype(np.float32)
Y_train = np.random.random((25, )).astype(np.float32)
validation_splitter = KFoldCrossValidationSplitter(3)
scoring_callback = ScoringCallback(median_absolute_error,
higher_score_is_better=False)
auto_ml = AutoML(
pipeline=pipeline,
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
validation_splitter=validation_splitter,
scoring_callback=scoring_callback,
n_trials=2,
epochs=1,
hyperparams_repository=HyperparamsJSONRepository(cache_folder=tmpdir),
refit_trial=True,
continue_loop_on_error=False)
auto_ml.fit(X_train, Y_train)
def test_automl_with_kfold(tmpdir):
# Given
hp_repository = HyperparamsJSONRepository(cache_folder=str('caching'))
auto_ml = AutoML(
pipeline=Pipeline([
MultiplyByN(2).set_hyperparams_space(
HyperparameterSpace({'multiply_by': FixedHyperparameter(2)})),
NumpyReshape(new_shape=(-1, 1)),
linear_model.LinearRegression()
]),
validation_splitter=ValidationSplitter(0.20),
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
scoring_callback=ScoringCallback(mean_squared_error,
higher_score_is_better=False),
callbacks=[
MetricCallback('mse',
metric_function=mean_squared_error,
higher_score_is_better=False),
],
n_trials=1,
epochs=10,
refit_trial=True,
print_func=print,
hyperparams_repository=hp_repository,
continue_loop_on_error=False)
data_inputs = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
expected_outputs = data_inputs * 4
# When
auto_ml.fit(data_inputs=data_inputs, expected_outputs=expected_outputs)
# Then
p = auto_ml.get_best_model()
outputs = p.transform(data_inputs)
mse = mean_squared_error(expected_outputs, outputs)
assert mse < 1000
def test_automl_early_stopping_callback(tmpdir):
# Given
hp_repository = InMemoryHyperparamsRepository(cache_folder=str(tmpdir))
n_epochs = 10
max_epochs_without_improvement = 3
auto_ml = AutoML(
pipeline=Pipeline([
MultiplyByN(2).set_hyperparams_space(
HyperparameterSpace({'multiply_by': FixedHyperparameter(2)})),
NumpyReshape(new_shape=(-1, 1)),
]),
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
validation_splitter=ValidationSplitter(0.20),
scoring_callback=ScoringCallback(mean_squared_error,
higher_score_is_better=False),
callbacks=[
MetricCallback('mse',
metric_function=mean_squared_error,
higher_score_is_better=False),
EarlyStoppingCallback(max_epochs_without_improvement)
],
n_trials=1,
refit_trial=True,
epochs=n_epochs,
hyperparams_repository=hp_repository,
continue_loop_on_error=False)
# When
data_inputs = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
expected_outputs = data_inputs * 2
auto_ml.fit(data_inputs=data_inputs, expected_outputs=expected_outputs)
# Then
trial = hp_repository.trials[0]
assert len(trial.validation_splits) == 1
validation_scores = trial.validation_splits[0].get_validation_scores()
nepochs_executed = len(validation_scores)
assert nepochs_executed == max_epochs_without_improvement + 1
def _make_autoML_loop(tmpdir, p: Pipeline):
hp_repository = InMemoryHyperparamsRepository(cache_folder=str(tmpdir) + "_hp")
n_epochs = 1
return AutoML(
pipeline=p,
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
validation_splitter=ValidationSplitter(0.20),
scoring_callback=ScoringCallback(mean_squared_error, higher_score_is_better=False),
n_trials=1,
refit_trial=True,
epochs=n_epochs,
hyperparams_repository=hp_repository,
cache_folder_when_no_handle=str(tmpdir),
continue_loop_on_error=False
)
def main(tmpdir, sleep_time: float = 0.001, n_iter: int = 10):
DATA_INPUTS = np.array(range(100))
EXPECTED_OUTPUTS = np.array(range(100, 200))
HYPERPARAMETER_SPACE = HyperparameterSpace({
'multiplication_1__multiply_by': RandInt(1, 2),
'multiplication_2__multiply_by': RandInt(1, 2),
'multiplication_3__multiply_by': RandInt(1, 2),
})
print('Classic Pipeline:')
classic_pipeline_folder = os.path.join(str(tmpdir), 'classic')
pipeline = Pipeline([
('multiplication_1', MultiplyByN()),
('sleep_1', ForEachDataInput(Sleep(sleep_time))),
('multiplication_2', MultiplyByN()),
('sleep_2', ForEachDataInput(Sleep(sleep_time))),
('multiplication_3', MultiplyByN()),
], cache_folder=classic_pipeline_folder).set_hyperparams_space(HYPERPARAMETER_SPACE)
time_a = time.time()
auto_ml = AutoML(
pipeline,
refit_trial=True,
n_trials=n_iter,
cache_folder_when_no_handle=classic_pipeline_folder,
validation_splitter=ValidationSplitter(0.2),
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
scoring_callback=ScoringCallback(mean_squared_error, higher_score_is_better=False),
callbacks=[
MetricCallback('mse', metric_function=mean_squared_error, higher_score_is_better=False)
],
)
auto_ml = auto_ml.fit(DATA_INPUTS, EXPECTED_OUTPUTS)
outputs = auto_ml.get_best_model().predict(DATA_INPUTS)
time_b = time.time()
actual_score = mean_squared_error(EXPECTED_OUTPUTS, outputs)
print('{0} seconds'.format(time_b - time_a))
print('output: {0}'.format(outputs))
print('smallest mse: {0}'.format(actual_score))
print('best hyperparams: {0}'.format(pipeline.get_hyperparams()))
assert isinstance(actual_score, float)
print('Resumable Pipeline:')
resumable_pipeline_folder = os.path.join(str(tmpdir), 'resumable')
pipeline = ResumablePipeline([
('multiplication_1', MultiplyByN()),
('ForEach(sleep_1)', ForEachDataInput(Sleep(sleep_time))),
('checkpoint1', ExpandDim(DefaultCheckpoint())),
('multiplication_2', MultiplyByN()),
('sleep_2', ForEachDataInput(Sleep(sleep_time))),
('checkpoint2', ExpandDim(DefaultCheckpoint())),
('multiplication_3', MultiplyByN())
], cache_folder=resumable_pipeline_folder).set_hyperparams_space(HYPERPARAMETER_SPACE)
time_a = time.time()
auto_ml = AutoML(
pipeline,
refit_trial=True,
n_trials=n_iter,
cache_folder_when_no_handle=resumable_pipeline_folder,
validation_splitter=ValidationSplitter(0.2),
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
scoring_callback=ScoringCallback(mean_squared_error, higher_score_is_better=False),
callbacks=[
MetricCallback('mse', metric_function=mean_squared_error, higher_score_is_better=False)
]
)
auto_ml = auto_ml.fit(DATA_INPUTS, EXPECTED_OUTPUTS)
outputs = auto_ml.get_best_model().predict(DATA_INPUTS)
time_b = time.time()
pipeline.flush_all_cache()
actual_score = mean_squared_error(EXPECTED_OUTPUTS, outputs)
print('{0} seconds'.format(time_b - time_a))
print('output: {0}'.format(outputs))
print('smallest mse: {0}'.format(actual_score))
print('best hyperparams: {0}'.format(pipeline.get_hyperparams()))
assert isinstance(actual_score, float)
def main():
def accuracy(data_inputs, expected_outputs):
return np.mean(
np.argmax(np.array(data_inputs), axis=1) == np.argmax(
np.array(expected_outputs), axis=1))
# load the dataset
df = read_csv('data/winequality-white.csv', sep=';')
data_inputs = df.values
data_inputs[:, -1] = data_inputs[:, -1] - 1
n_features = data_inputs.shape[1] - 1
n_classes = 10
p = Pipeline([
TrainOnlyWrapper(DataShuffler()),
ColumnTransformerInputOutput(
input_columns=[(
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10], ToNumpy(np.float32)
)],
output_columns=[(11, Identity())]
),
OutputTransformerWrapper(PlotDistribution(column=-1)),
MiniBatchSequentialPipeline([
Tensorflow2ModelStep(
create_model=create_model,
create_loss=create_loss,
create_optimizer=create_optimizer
) \
.set_hyperparams(HyperparameterSamples({
'n_dense_layers': 2,
'input_dim': n_features,
'optimizer': 'adam',
'activation': 'relu',
'kernel_initializer': 'he_uniform',
'learning_rate': 0.01,
'hidden_dim': 20,
'n_classes': 3
})).set_hyperparams_space(HyperparameterSpace({
'n_dense_layers': RandInt(2, 4),
'hidden_dim_layer_multiplier': Uniform(0.30, 1),
'input_dim': FixedHyperparameter(n_features),
'optimizer': Choice([
OPTIMIZERS.ADAM.value,
OPTIMIZERS.SGD.value,
OPTIMIZERS.ADAGRAD.value
]),
'activation': Choice([
ACTIVATIONS.RELU.value,
ACTIVATIONS.TANH.value,
ACTIVATIONS.SIGMOID.value,
ACTIVATIONS.ELU.value,
]),
'kernel_initializer': Choice([
KERNEL_INITIALIZERS.GLOROT_NORMAL.value,
KERNEL_INITIALIZERS.GLOROT_UNIFORM.value,
KERNEL_INITIALIZERS.HE_UNIFORM.value
]),
'learning_rate': LogUniform(0.005, 0.01),
'hidden_dim': RandInt(3, 80),
'n_classes': FixedHyperparameter(n_classes)
}))
], batch_size=33),
OutputTransformerWrapper(Pipeline([
ExpandDim(),
OneHotEncoder(nb_columns=n_classes, name='classes')
]))
])
auto_ml = AutoML(
pipeline=p,
hyperparams_repository=InMemoryHyperparamsRepository(
cache_folder='trials'),
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
validation_splitter=ValidationSplitter(test_size=0.30),
scoring_callback=ScoringCallback(accuracy,
higher_score_is_better=True),
callbacks=[
MetricCallback(
name='classification_report_imbalanced_metric',
metric_function=classificaiton_report_imbalanced_metric,
higher_score_is_better=True),
MetricCallback(name='f1',
metric_function=f1_score_weighted,
higher_score_is_better=True),
MetricCallback(name='recall',
metric_function=recall_score_weighted,
higher_score_is_better=True),
MetricCallback(name='precision',
metric_function=precision_score_weighted,
higher_score_is_better=True),
EarlyStoppingCallback(max_epochs_without_improvement=3)
],
n_trials=200,
refit_trial=True,
epochs=75)
auto_ml = auto_ml.fit(data_inputs=data_inputs)
def main():
# Define classification models, and hyperparams.
# See also HyperparameterSpace documentation : https://www.neuraxle.org/stable/api/neuraxle.hyperparams.space.html#neuraxle.hyperparams.space.HyperparameterSpace
decision_tree_classifier = SKLearnWrapper(
DecisionTreeClassifier(),
HyperparameterSpace({
'criterion': Choice(['gini', 'entropy']),
'splitter': Choice(['best', 'random']),
'min_samples_leaf': RandInt(2, 5),
'min_samples_split': RandInt(2, 4)
}))
extra_tree_classifier = SKLearnWrapper(
ExtraTreeClassifier(),
HyperparameterSpace({
'criterion': Choice(['gini', 'entropy']),
'splitter': Choice(['best', 'random']),
'min_samples_leaf': RandInt(2, 5),
'min_samples_split': RandInt(2, 4)
}))
ridge_classifier = Pipeline([
OutputTransformerWrapper(NumpyRavel()),
SKLearnWrapper(
RidgeClassifier(),
HyperparameterSpace({
'alpha': Choice([0.0, 1.0, 10.0, 100.0]),
'fit_intercept': Boolean(),
'normalize': Boolean()
}))
]).set_name('RidgeClassifier')
logistic_regression = Pipeline([
OutputTransformerWrapper(NumpyRavel()),
SKLearnWrapper(
LogisticRegression(),
HyperparameterSpace({
'C': LogUniform(0.01, 10.0),
'fit_intercept': Boolean(),
'penalty': Choice(['none', 'l2']),
'max_iter': RandInt(20, 200)
}))
]).set_name('LogisticRegression')
random_forest_classifier = Pipeline([
OutputTransformerWrapper(NumpyRavel()),
SKLearnWrapper(
RandomForestClassifier(),
HyperparameterSpace({
'n_estimators': RandInt(50, 600),
'criterion': Choice(['gini', 'entropy']),
'min_samples_leaf': RandInt(2, 5),
'min_samples_split': RandInt(2, 4),
'bootstrap': Boolean()
}))
]).set_name('RandomForestClassifier')
# Define a classification pipeline that lets the AutoML loop choose one of the classifier.
# See also ChooseOneStepOf documentation : https://www.neuraxle.org/stable/api/neuraxle.steps.flow.html#neuraxle.steps.flow.ChooseOneStepOf
pipeline = Pipeline([
ChooseOneStepOf([
decision_tree_classifier, extra_tree_classifier, ridge_classifier,
logistic_regression, random_forest_classifier
])
])
# Create the AutoML loop object.
# See also AutoML documentation : https://www.neuraxle.org/stable/api/neuraxle.metaopt.auto_ml.html#neuraxle.metaopt.auto_ml.AutoML
auto_ml = AutoML(
pipeline=pipeline,
hyperparams_optimizer=RandomSearchHyperparameterSelectionStrategy(),
validation_splitter=ValidationSplitter(test_size=0.20),
scoring_callback=ScoringCallback(accuracy_score,
higher_score_is_better=True),
n_trials=7,
epochs=1,
hyperparams_repository=HyperparamsJSONRepository(cache_folder='cache'),
refit_trial=True,
continue_loop_on_error=False)
# Load data, and launch AutoML loop !
X_train, y_train, X_test, y_test = generate_classification_data()
auto_ml = auto_ml.fit(X_train, y_train)
# Get the model from the best trial, and make predictions using predict.
# See also predict documentation : https://www.neuraxle.org/stable/api/neuraxle.base.html#neuraxle.base.BaseStep.predict
best_pipeline = auto_ml.get_best_model()
y_pred = best_pipeline.predict(X_test)
accuracy = accuracy_score(y_true=y_test, y_pred=y_pred)
print("Test accuracy score:", accuracy)
shutil.rmtree('cache')
def main(tmpdir):
boston = load_boston()
X, y = shuffle(boston.data, boston.target, random_state=13)
X = X.astype(np.float32)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, shuffle=False)
# Note that the hyperparameter spaces are defined here during the pipeline definition, but it could be already set
# within the classes ar their definition if using custom classes, or also it could be defined after declaring the
# pipeline using a flat dict or a nested dict.
p = Pipeline([
AddFeatures([
SKLearnWrapper(
PCA(n_components=2),
HyperparameterSpace({"n_components": RandInt(1, 3)})
),
SKLearnWrapper(
FastICA(n_components=2),
HyperparameterSpace({"n_components": RandInt(1, 3)})
),
]),
ModelStacking([
SKLearnWrapper(
GradientBoostingRegressor(),
HyperparameterSpace({
"n_estimators": RandInt(50, 300), "max_depth": RandInt(1, 4),
"learning_rate": LogUniform(0.07, 0.7)
})
),
SKLearnWrapper(
KMeans(),
HyperparameterSpace({"n_clusters": RandInt(5, 10)})
),
],
joiner=NumpyTranspose(),
judge=SKLearnWrapper(
Ridge(),
HyperparameterSpace({"alpha": LogUniform(0.7, 1.4), "fit_intercept": Boolean()})
),
)
])
print("Meta-fitting on train:")
auto_ml = AutoML(
p,
validation_splitter=ValidationSplitter(0.20),
refit_trial=True,
n_trials=10,
epochs=1, # 1 epoc here due to using sklearn models that just fit once.
cache_folder_when_no_handle=str(tmpdir),
scoring_callback=ScoringCallback(mean_squared_error, higher_score_is_better=False),
callbacks=[MetricCallback('mse', metric_function=mean_squared_error, higher_score_is_better=False)],
hyperparams_repository=InMemoryHyperparamsRepository(cache_folder=str(tmpdir))
)
random_search = auto_ml.fit(X_train, y_train)
p = random_search.get_best_model()
print("")
print("Transforming train and test:")
y_train_predicted = p.predict(X_train)
y_test_predicted = p.predict(X_test)
print("")
print("Evaluating transformed train:")
score_transform = r2_score(y_train_predicted, y_train)
print('R2 regression score:', score_transform)
print("")
print("Evaluating transformed test:")
score_test = r2_score(y_test_predicted, y_test)
print('R2 regression score:', score_test)
