def test_generate_code_nonlinear_pipeline_error(
nonlinear_binary_pipeline_class):
pipeline = nonlinear_binary_pipeline_class({})
with pytest.raises(
ValueError,
match="Code generation for nonlinear pipelines is not supported yet"
):
generate_pipeline_code(pipeline)
def test_generate_code_pipeline_with_custom_components():
class CustomTransformer(Transformer):
name = "My Custom Transformer"
hyperparameter_ranges = {}
def __init__(self, random_seed=0):
parameters = {}
super().__init__(parameters=parameters,
component_obj=None,
random_seed=random_seed)
class CustomEstimator(Estimator):
name = "My Custom Estimator"
hyperparameter_ranges = {}
supported_problem_types = [
ProblemTypes.BINARY, ProblemTypes.MULTICLASS
]
model_family = ModelFamily.NONE
def __init__(self, random_arg=False, random_seed=0):
parameters = {'random_arg': random_arg}
super().__init__(parameters=parameters,
component_obj=None,
random_seed=random_seed)
mock_pipeline_with_custom_components = BinaryClassificationPipeline(
[CustomTransformer, CustomEstimator])
expected_code = "from evalml.pipelines.binary_classification_pipeline import BinaryClassificationPipeline\n" \
"pipeline = BinaryClassificationPipeline(component_graph=[CustomTransformer, CustomEstimator], " \
"parameters={'My Custom Estimator':{'random_arg': False}}, random_seed=0)"
pipeline = generate_pipeline_code(mock_pipeline_with_custom_components)
assert pipeline == expected_code
def test_generate_code_pipeline():
custom_hyperparameters = {
"Imputer": {
"numeric_impute_strategy": 'most_frequent'
}
}
binary_pipeline = BinaryClassificationPipeline(
['Imputer', 'Random Forest Classifier'],
custom_hyperparameters=custom_hyperparameters)
expected_code = "from evalml.pipelines.binary_classification_pipeline import BinaryClassificationPipeline\n" \
"pipeline = BinaryClassificationPipeline(component_graph=['Imputer', 'Random Forest Classifier'], " \
"parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, " \
"'Random Forest Classifier':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}, custom_hyperparameters={'Imputer':{'numeric_impute_strategy': 'most_frequent'}}, random_seed=0)"
pipeline = generate_pipeline_code(binary_pipeline)
assert expected_code == pipeline
regression_pipeline = RegressionPipeline(
['Imputer', 'Random Forest Regressor'],
custom_name="Mock Regression Pipeline")
expected_code = "from evalml.pipelines.regression_pipeline import RegressionPipeline\n" \
"pipeline = RegressionPipeline(component_graph=['Imputer', 'Random Forest Regressor'], parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, " \
"'Random Forest Regressor':{'n_estimators': 100, 'max_depth': 6, 'n_jobs': -1}}, custom_name='Mock Regression Pipeline', random_seed=0)"
pipeline = generate_pipeline_code(regression_pipeline)
assert pipeline == expected_code
regression_pipeline_with_params = RegressionPipeline(
['Imputer', 'Random Forest Regressor'],
custom_name="Mock Regression Pipeline",
parameters={
"Imputer": {
"numeric_impute_strategy": "most_frequent"
},
"Random Forest Regressor": {
"n_estimators": 50
}
})
expected_code_params = "from evalml.pipelines.regression_pipeline import RegressionPipeline\n" \
"pipeline = RegressionPipeline(component_graph=['Imputer', 'Random Forest Regressor'], " \
"parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'most_frequent', 'categorical_fill_value': None, 'numeric_fill_value': None}, " \
"'Random Forest Regressor':{'n_estimators': 50, 'max_depth': 6, 'n_jobs': -1}}, custom_name='Mock Regression Pipeline', random_seed=0)"
pipeline = generate_pipeline_code(regression_pipeline_with_params)
assert pipeline == expected_code_params
def test_generate_code_pipeline_json_with_objects():
class CustomEstimator(Estimator):
name = "My Custom Estimator"
hyperparameter_ranges = {}
supported_problem_types = [
ProblemTypes.BINARY, ProblemTypes.MULTICLASS
]
model_family = ModelFamily.NONE
def __init__(self, random_arg=False, numpy_arg=[], random_seed=0):
parameters = {'random_arg': random_arg, 'numpy_arg': numpy_arg}
super().__init__(parameters=parameters,
component_obj=None,
random_seed=random_seed)
component_graph = ['Imputer', CustomEstimator]
pipeline = BinaryClassificationPipeline(
component_graph,
custom_name="Mock Binary Pipeline with Transformer",
parameters={'My Custom Estimator': {
'numpy_arg': np.array([0])
}})
generated_pipeline_code = generate_pipeline_code(pipeline)
assert generated_pipeline_code == "from evalml.pipelines.binary_classification_pipeline import BinaryClassificationPipeline\n" \
"pipeline = BinaryClassificationPipeline(component_graph=['Imputer', CustomEstimator], " \
"parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, " \
"'My Custom Estimator':{'random_arg': False, 'numpy_arg': array([0])}}, custom_name='Mock Binary Pipeline with Transformer', random_seed=0)"
pipeline = BinaryClassificationPipeline(
component_graph,
custom_name="Mock Binary Pipeline with Transformer",
parameters={'My Custom Estimator': {
'random_arg': Imputer()
}})
generated_pipeline_code = generate_pipeline_code(pipeline)
assert generated_pipeline_code == "from evalml.pipelines.binary_classification_pipeline import BinaryClassificationPipeline\n" \
"pipeline = BinaryClassificationPipeline(component_graph=['Imputer', CustomEstimator], " \
"parameters={'Imputer':{'categorical_impute_strategy': 'most_frequent', 'numeric_impute_strategy': 'mean', 'categorical_fill_value': None, 'numeric_fill_value': None}, " \
"'My Custom Estimator':{'random_arg': Imputer(categorical_impute_strategy='most_frequent', numeric_impute_strategy='mean', categorical_fill_value=None, numeric_fill_value=None), 'numpy_arg': []}}, " \
"custom_name='Mock Binary Pipeline with Transformer', random_seed=0)"
def test_generate_code_pipeline_errors():
class MockBinaryPipeline(BinaryClassificationPipeline):
name = "Mock Binary Pipeline"
component_graph = ['Imputer', 'Random Forest Classifier']
class MockMulticlassPipeline(MulticlassClassificationPipeline):
name = "Mock Multiclass Pipeline"
component_graph = ['Imputer', 'Random Forest Classifier']
class MockRegressionPipeline(RegressionPipeline):
name = "Mock Regression Pipeline"
component_graph = ['Imputer', 'Random Forest Regressor']
with pytest.raises(ValueError,
match="Element must be a pipeline instance"):
generate_pipeline_code(MockBinaryPipeline)
with pytest.raises(ValueError,
match="Element must be a pipeline instance"):
generate_pipeline_code(MockMulticlassPipeline)
with pytest.raises(ValueError,
match="Element must be a pipeline instance"):
generate_pipeline_code(MockRegressionPipeline)
with pytest.raises(ValueError,
match="Element must be a pipeline instance"):
generate_pipeline_code([Imputer])
with pytest.raises(ValueError,
match="Element must be a pipeline instance"):
generate_pipeline_code([Imputer, LogisticRegressionClassifier])
with pytest.raises(ValueError,
match="Element must be a pipeline instance"):
generate_pipeline_code([Imputer(), LogisticRegressionClassifier()])