def create_test_case_invalid_step_choosen():
a_callback = TapeCallbackFunction()
b_callback = TapeCallbackFunction()
return NeuraxleTestCase(pipeline=Pipeline([
ChooseOneOrManyStepsOf([
('a',
TransformCallbackStep(a_callback,
transform_function=lambda di: di * 2)),
('b',
TransformCallbackStep(b_callback,
transform_function=lambda di: di * 2))
]),
]),
callbacks=[a_callback, b_callback],
expected_callbacks_data=[DATA_INPUTS, DATA_INPUTS],
hyperparams={
'ChooseOneOrManyStepsOf__c__enabled': True,
'ChooseOneOrManyStepsOf__b__enabled': False
},
hyperparams_space={
'ChooseOneOrManyStepsOf__a__enabled':
Boolean(),
'ChooseOneOrManyStepsOf__b__enabled':
Boolean()
},
expected_processed_outputs=np.array(
[0, 2, 4, 6, 8, 10, 12, 14, 16, 18]))
def create_test_case_fit_multiple_steps_choosen():
a_callback = TapeCallbackFunction()
b_callback = TapeCallbackFunction()
c_callback = TapeCallbackFunction()
d_callback = TapeCallbackFunction()
return NeuraxleTestCase(
pipeline=Pipeline([
ChooseOneOrManyStepsOf([
('a', FitTransformCallbackStep(a_callback, c_callback, transform_function=lambda di: di * 2)),
('b', FitTransformCallbackStep(b_callback, d_callback, transform_function=lambda di: di * 2))
]),
]),
callbacks=[a_callback, c_callback, b_callback, d_callback],
expected_callbacks_data=[
[],
(DATA_INPUTS, EXPECTED_OUTPUTS),
[],
(DATA_INPUTS, EXPECTED_OUTPUTS)
],
hyperparams={
'ChooseOneOrManyStepsOf__a__enabled': True,
'ChooseOneOrManyStepsOf__b__enabled': True
},
hyperparams_space={
'ChooseOneOrManyStepsOf__a__enabled': Boolean(),
'ChooseOneOrManyStepsOf__b__enabled': Boolean()
},
expected_processed_outputs=np.array([0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 0, 2, 4, 6, 8, 10, 12, 14, 16, 18])
)
def test_model_stacking_fit_transform():
model_stacking = Pipeline([
ModelStacking(
[
SKLearnWrapper(
GradientBoostingRegressor(),
HyperparameterSpace({
"n_estimators": RandInt(50, 600),
"max_depth": RandInt(1, 10),
"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()
})),
)
])
expected_outputs_shape = (379, 1)
data_inputs_shape = (379, 13)
data_inputs = _create_data(data_inputs_shape)
expected_outputs = _create_data(expected_outputs_shape)
model_stacking, outputs = model_stacking.fit_transform(
data_inputs, expected_outputs)
assert outputs.shape == expected_outputs_shape
def __init__(self, brothers):
super().__init__(brothers,
SKLearnWrapper(
Ridge(),
HyperparameterSpace({
"alpha": LogUniform(0.1, 10.0),
"fit_intercept": Boolean()
})),
joiner=NumpyTranspose())
def __init__(self, wrapped: BaseTransformer, enabled: bool = True, nullified_return_value=None,
cache_folder_when_no_handle=None, use_hyperparameter_space=True, nullify_hyperparams=True):
hyperparameter_space = HyperparameterSpace({
OPTIONAL_ENABLED_HYPERPARAM: Boolean()
}) if use_hyperparameter_space else {}
MetaStep.__init__(
self,
hyperparams=HyperparameterSamples({
OPTIONAL_ENABLED_HYPERPARAM: enabled
}),
hyperparams_space=hyperparameter_space,
wrapped=wrapped
)
ForceHandleOnlyMixin.__init__(self, cache_folder_when_no_handle)
if nullified_return_value is None:
nullified_return_value = []
self.nullified_return_value = nullified_return_value
self.nullify_hyperparams = nullify_hyperparams
from neuraxle.base import MetaStepMixin, BaseStep, NonFittableMixin, NonTransformableMixin
from neuraxle.hyperparams.distributions import RandInt, Boolean
from neuraxle.hyperparams.space import HyperparameterSpace, HyperparameterSamples
from neuraxle.steps.loop import StepClonerForEachDataInput
from testing.test_pipeline import SomeStep
SOME_STEP_HP_KEY = 'somestep_hyperparam'
RAND_INT_SOME_STEP = RandInt(-10, 0)
RAND_INT_STEP_CLONER = RandInt(0, 10)
META_STEP_HP = 'metastep_hyperparam'
SOME_STEP_HP = "SomeStep__somestep_hyperparam"
META_STEP_HP_VALUE = 1
SOME_STEP_HP_VALUE = 2
HYPE_SPACE = HyperparameterSpace({"a__test": Boolean()})
HYPE_SAMPLE = HyperparameterSamples({"a__test": True})
class SomeMetaStepMixin(NonTransformableMixin, NonFittableMixin, MetaStepMixin,
BaseStep):
pass
class SomeStepInverseTransform(SomeStep):
def fit_transform(self, data_inputs, expected_outputs=None):
return self, 'fit_transform'
def inverse_transform(self, processed_outputs):
return 'inverse_transform'
SKLearnWrapper(
GradientBoostingRegressor(),
HyperparameterSpace({
"n_estimators": RandInt(50, 600),
"max_depth": RandInt(1, 10),
"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:")
p = p.meta_fit(X_train,
y_train,
metastep=RandomSearch(n_iter=10,
higher_score_is_better=True,
validation_technique=KFoldCrossValidation(
scoring_function=r2_score,
k_fold=10)))
# Here is an alternative way to do it, more "pipeliney":
# p = RandomSearch(
# n_iter=15,
def main():
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, 600),
"max_depth": RandInt(1, 10),
"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:")
p = p.meta_fit(X_train,
y_train,
metastep=RandomSearch(
n_iter=10,
higher_score_is_better=True,
validation_technique=KFoldCrossValidationWrapper(
scoring_function=r2_score, k_fold=10)))
# Here is an alternative way to do it, more "pipeliney":
# p = RandomSearch(
# p,
# n_iter=15,
# higher_score_is_better=True,
# validation_technique=KFoldCrossValidation(scoring_function=r2_score, k_fold=3)
# ).fit(X_train, y_train)
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)
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')
HYPERPARAMETERS_SPACE = HyperparameterSpace({
'learning_rate':
LogUniform(0.0001, 0.1),
'l2_weight_reg':
LogUniform(0.0001, 0.1),
'momentum':
LogUniform(0.01, 1.0),
'hidden_size':
Quantized(LogUniform(16, 512)),
'num_layers':
RandInt(1, 4),
'num_lstm_layers':
RandInt(1, 2),
'use_xavier_init':
Boolean(),
'use_max_pool_else_avg_pool':
Boolean(),
'dropout_drop_proba':
LogUniform(0.3, 0.7)
})
HYPERPARAMETERS = HyperparameterSamples({
'learning_rate': 0.1,
'l2_weight_reg': 0.001,
'hidden_size': 32,
'num_layers': 3,
'num_lstm_layers': 1,
'use_xavier_init': True,
'use_max_pool_else_avg_pool': True,
'dropout_drop_proba': 0.5,
from neuraxle.hyperparams.distributions import RandInt, Boolean
from neuraxle.hyperparams.space import HyperparameterSpace, HyperparameterSamples
from neuraxle.steps.loop import StepClonerForEachDataInput
from testing.test_pipeline import SomeStep
SOME_STEP_HP_KEY = 'somestep_hyperparam'
RAND_INT_SOME_STEP = RandInt(-10, 0)
RAND_INT_STEP_CLONER = RandInt(0, 10)
META_STEP_HP = 'metastep_hyperparam'
SOME_STEP_HP = "SomeStep__somestep_hyperparam"
META_STEP_HP_VALUE = 1
SOME_STEP_HP_VALUE = 2
HYPE_SPACE = HyperparameterSpace({
"a__test": Boolean()
})
HYPE_SAMPLE = HyperparameterSamples({
"a__test": True
})
class SomeMetaStepMixin(NonTransformableMixin, NonFittableMixin, MetaStepMixin, BaseStep):
pass
class SomeStepInverseTransform(SomeStep):
def fit_transform(self, data_inputs, expected_outputs=None):
return self, 'fit_transform'
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)
def test_deep_learning_pipeline():
# Given
boston = load_boston()
data_inputs, expected_outputs = shuffle(boston.data,
boston.target,
random_state=13)
expected_outputs = expected_outputs.astype(np.float32)
data_inputs = data_inputs.astype(np.float32)
pipeline = 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, 600),
"max_depth": RandInt(1, 10),
"learning_rate": LogUniform(0.07, 0.7)
})),
SKLearnWrapper(
KMeans(n_clusters=7),
HyperparameterSpace({"n_clusters": RandInt(5, 10)})),
],
joiner=NumpyTranspose(),
judge=SKLearnWrapper(
Ridge(),
HyperparameterSpace({
"alpha": LogUniform(0.7, 1.4),
"fit_intercept": Boolean()
})),
)
])
p = DeepLearningPipeline(
pipeline,
validation_size=VALIDATION_SIZE,
batch_size=BATCH_SIZE,
batch_metrics={'mse': to_numpy_metric_wrapper(mean_squared_error)},
shuffle_in_each_epoch_at_train=True,
n_epochs=N_EPOCHS,
epochs_metrics={'mse': to_numpy_metric_wrapper(mean_squared_error)},
scoring_function=to_numpy_metric_wrapper(mean_squared_error),
)
# When
p, outputs = p.fit_transform(data_inputs, expected_outputs)
# Then
batch_mse_train = p.get_batch_metric_train('mse')
epoch_mse_train = p.get_epoch_metric_train('mse')
batch_mse_validation = p.get_batch_metric_validation('mse')
epoch_mse_validation = p.get_epoch_metric_validation('mse')
assert len(epoch_mse_train) == N_EPOCHS
assert len(epoch_mse_validation) == N_EPOCHS
expected_len_batch_mse_train = math.ceil(
(len(data_inputs) / BATCH_SIZE) * (1 - VALIDATION_SIZE)) * N_EPOCHS
expected_len_batch_mse_validation = math.ceil(
(len(data_inputs) / BATCH_SIZE) * VALIDATION_SIZE) * N_EPOCHS
assert len(batch_mse_train) == expected_len_batch_mse_train
assert len(batch_mse_validation) == expected_len_batch_mse_validation
last_batch_mse_validation = batch_mse_validation[-1]
last_batch_mse_train = batch_mse_train[-1]
last_epoch_mse_train = epoch_mse_train[-1]
last_epoch_mse_validation = epoch_mse_validation[-1]
assert last_batch_mse_train < last_batch_mse_validation
assert last_epoch_mse_train < last_epoch_mse_validation
assert last_batch_mse_train < 1
assert last_epoch_mse_train < 1