def __init__(self, stats, ensemble_method: str,
ensemble_size: int,
task_type: int,
metric: _BaseScorer,
output_dir=None):
self.model = None
if ensemble_method == 'bagging':
self.model = Bagging(stats=stats,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
elif ensemble_method == 'blending':
self.model = Blending(stats=stats,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
elif ensemble_method == 'stacking':
self.model = Stacking(stats=stats,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
elif ensemble_method == 'ensemble_selection':
self.model = EnsembleSelection(stats=stats,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
else:
raise ValueError("%s is not supported for ensemble!" % ensemble_method)
class EnsembleBuilder:
def __init__(self,
stats,
data_node,
ensemble_method: str,
ensemble_size: int,
task_type: int,
metric: _BaseScorer,
output_dir=None):
self.model = None
if ensemble_method == 'bagging':
self.model = Bagging(stats=stats,
data_node=data_node,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
elif ensemble_method == 'blending':
self.model = Blending(stats=stats,
data_node=data_node,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
elif ensemble_method == 'stacking':
self.model = Stacking(stats=stats,
data_node=data_node,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
elif ensemble_method == 'ensemble_selection':
self.model = EnsembleSelection(stats=stats,
data_node=data_node,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
else:
raise ValueError("%s is not supported for ensemble!" %
ensemble_method)
def fit(self, data):
return self.model.fit(data)
def predict(self, data):
return self.model.predict(data)
def refit(self):
return self.model.refit()
def get_ens_model_info(self):
return self.model.get_ens_model_info()
def ensemble_implementation_examples(bandit: FirstLayerBandit, test_data: DataNode):
from sklearn.model_selection import StratifiedShuffleSplit
from sklearn.metrics import accuracy_score
from autosklearn.metrics import accuracy
n_best = 20
stats = bandit.fetch_ensemble_members(test_data)
seed = stats['split_seed']
test_size = 0.2
train_predictions = []
test_predictions = []
for algo_id in bandit.nbest_algo_ids:
X, y = stats[algo_id]['train_dataset'].data
sss = StratifiedShuffleSplit(n_splits=1, test_size=test_size, random_state=1)
for train_index, test_index in sss.split(X, y):
X_train, X_valid = X[train_index], X[test_index]
y_train, y_valid = y[train_index], y[test_index]
X_test, y_test = stats[algo_id]['test_dataset'].data
configs = stats[algo_id]['configurations']
performance = stats[algo_id]['performance']
best_index = np.argsort(-np.array(performance))
best_configs = [configs[i] for i in best_index[:n_best]]
for config in best_configs:
try:
# Build the ML estimator.
_, estimator = get_estimator(config)
# print(X_train.shape, X_test.shape)
estimator.fit(X_train, y_train)
y_valid_pred = estimator.predict_proba(X_valid)
y_test_pred = estimator.predict_proba(X_test)
train_predictions.append(y_valid_pred)
test_predictions.append(y_test_pred)
except Exception as e:
print(str(e))
es = EnsembleSelection(ensemble_size=50, task_type=1,
metric=accuracy, random_state=np.random.RandomState(seed))
assert len(train_predictions) == len(test_predictions)
es.fit(train_predictions, y_valid, identifiers=None)
y_pred = es.predict(test_predictions)
y_pred = np.argmax(y_pred, axis=-1)
test_score = accuracy_score(y_test, y_pred)
return test_score