def test_metalearning():
from mosaic_ml.autosklearn_wrapper.autosklearn import get_autosklearn_metalearning
for task in [252, 9971]:
X_train, y_train, X_test, y_test, cat = load_task(task)
autoML = AutoML(time_budget=60,
time_limit_for_evaluation=30,
memory_limit=3024,
seed=1,
scoring_func="balanced_accuracy",
verbose=0,
ensemble_size=0)
intial_configuration_metalearning_AS, _ = get_autosklearn_metalearning(
X_train, y_train, cat, "balanced_accuracy", 50)
best_config, best_score = autoML.fit(
X_train,
y_train,
X_test,
y_test,
categorical_features=cat,
initial_configurations=intial_configuration_metalearning_AS[:30]
) # init with 30 configs
assert best_config is not None
assert isinstance(best_score, float)
def test_get_metalearning_AS():
from mosaic_ml.autosklearn_wrapper.autosklearn import get_autosklearn_metalearning
for task in classification_tasks[::10]:
X_train, y_train, X_test, y_test, cat = load_task(task)
initial_configuration_metalearning_AS, list_dataset_NN = get_autosklearn_metalearning(
X_train, y_train, cat, "balanced_accuracy", 50)
assert len(initial_configuration_metalearning_AS) == len(
list_dataset_NN)
assert sum([
config is not None
for config in initial_configuration_metalearning_AS
]) == len(initial_configuration_metalearning_AS)
assert sum([dataset is not None
for dataset in list_dataset_NN]) == len(list_dataset_NN)
def test_example():
for task in [252, 9971]:
X_train, y_train, X_test, y_test, cat = load_task(task)
autoML = AutoML(time_budget=60,
time_limit_for_evaluation=30,
memory_limit=3024,
seed=1,
scoring_func="balanced_accuracy",
verbose=0
)
best_config, best_score = autoML.fit(X_train, y_train, X_test, y_test, categorical_features=cat)
assert best_config is not None
assert isinstance(best_score, float)