def _xgboost(self):
params = self.params
if self.problem_type == 'regression':
model = XGBRegressor(**params)
model.fit(self.training_data.drop(["TARGET"], axis=1),
self.training_data['TARGET'])
preds = model.predict(self.validation_data.drop(['TARGET'],
axis=1))
return model, preds, self.validation_data['TARGET']
elif self.problem_type == 'classification':
model = XGBClassifier(**params)
model.fit(self.training_data.drop(["TARGET"], axis=1),
self.training_data['TARGET'])
preds = model.predict_proba(
self.validation_data.drop(['TARGET'], axis=1))[:, 1]
return model, preds, self.validation_data['TARGET']
elif self.problem_type == 'multiclass':
model = XGBClassifier(**params)
model.fit(self.training_data.drop(["TARGET"], axis=1),
self.training_data['TARGET'])
preds = model.predict_proba(
self.validation_data.drop(['TARGET'], axis=1))
preds = [np.argmax(p) for p in preds]
return model, preds, self.validation_data['TARGET']
else:
raise Exception("Problem Type not supported")
class XGboost:
"""docstring for XGboost"""
def __init__(self, model_configs, task_type='Regression'):
self.model_configs = model_configs
self.max_depth = model_configs['max_depth']
self.learning_rate = model_configs['learning_rate']
self.n_estimators = model_configs['n_estimators']
self.objective = model_configs['objective']
self.booster = model_configs['booster']
self.subsample = model_configs['subsample']
self.colsample_bylevel = model_configs['colsample_bylevel']
self.colsample_bytree = model_configs['colsample_bytree']
self.min_child_weight = model_configs['min_child_weight']
self.reg_alpha = model_configs['reg_alpha']
self.reg_lambda = model_configs['reg_lambda']
self.scale_pos_weight = model_configs['scale_pos_weight']
self.max_delta_step = model_configs['max_delta_step']
self.random_seed = model_configs['random_seed']
self.eval_metric = model_configs['early_stopping']['eval_metric']
self.early_stopping_round = model_configs['early_stopping']['round']
self.task_type = task_type
np.random.seed(seed=self.random_seed)
self.setup_model()
def setup_model(self):
if self.task_type == 'Classification':
self.model = XGBClassifier(
max_depth=self.max_depth,
learning_rate=self.learning_rate,
n_estimators=self.n_estimators,
objective=self.objective,
booster=self.booster,
subsample=self.subsample,
colsample_bylevel=self.colsample_bylevel,
colsample_bytree=self.colsample_bytree,
min_child_weight=self.min_child_weight,
reg_alpha=self.reg_alpha,
reg_lambda=self.reg_lambda,
scale_pos_weight=self.scale_pos_weight,
max_delta_step=self.max_delta_step,
random_state=self.random_seed,
silent=False,
n_jobs=8)
elif self.task_type == 'Regression':
self.model = XGBRegressor(max_depth=self.max_depth,
learning_rate=self.learning_rate,
n_estimators=self.n_estimators,
objective=self.objective,
booster=self.booster,
subsample=self.subsample,
colsample_bylevel=self.colsample_bylevel,
colsample_bytree=self.colsample_bytree,
min_child_weight=self.min_child_weight,
reg_alpha=self.reg_alpha,
reg_lambda=self.reg_lambda,
scale_pos_weight=self.scale_pos_weight,
random_state=self.random_seed,
silent=False,
n_jobs=8)
else:
raise "Task type Error!"
def fit_model(self, train_loader):
self.model.fit(train_loader[0], train_loader[1])
def predict(self, test_loader):
if self.task_type == 'Classification':
return self.model.predict_proba(test_loader[0])
else:
return self.model.predict(test_loader[0])
class Stacking(BaseEnsembleModel):
def __init__(self,
stats,
ensemble_size: int,
task_type: int,
metric: _BaseScorer,
output_dir=None,
meta_learner='xgboost',
kfold=5):
super().__init__(stats=stats,
ensemble_method='blending',
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
output_dir=output_dir)
self.kfold = kfold
try:
from xgboost import XGBClassifier
except:
warnings.warn(
"Xgboost is not imported! Blending will use linear model instead!"
)
meta_learner = 'linear'
# We use Xgboost as default meta-learner
if self.task_type in CLS_TASKS:
if meta_learner == 'linear':
from sklearn.linear_model.logistic import LogisticRegression
self.meta_learner = LogisticRegression(max_iter=1000)
elif meta_learner == 'gb':
from sklearn.ensemble.gradient_boosting import GradientBoostingClassifier
self.meta_learner = GradientBoostingClassifier(
learning_rate=0.05,
subsample=0.7,
max_depth=4,
n_estimators=250)
elif meta_learner == 'xgboost':
from xgboost import XGBClassifier
self.meta_learner = XGBClassifier(max_depth=4,
learning_rate=0.05,
n_estimators=150)
else:
if meta_learner == 'linear':
from sklearn.linear_model import LinearRegression
self.meta_learner = LinearRegression()
elif meta_learner == 'xgboost':
from xgboost import XGBRegressor
self.meta_learner = XGBRegressor(max_depth=4,
learning_rate=0.05,
n_estimators=70)
def fit(self, data):
# Split training data for phase 1 and phase 2
if self.task_type in CLS_TASKS:
kf = StratifiedKFold(n_splits=self.kfold)
else:
kf = KFold(n_splits=self.kfold)
# Train basic models using a part of training data
model_cnt = 0
suc_cnt = 0
feature_p2 = None
for algo_id in self.stats["include_algorithms"]:
train_list = self.stats[algo_id]['train_data_list']
configs = self.stats[algo_id]['configurations']
for idx in range(len(train_list)):
X, y = train_list[idx].data
for _config in configs:
if self.base_model_mask[model_cnt] == 1:
for j, (train, test) in enumerate(kf.split(X, y)):
x_p1, x_p2, y_p1, _ = X[train], X[test], y[
train], y[test]
estimator = fetch_predict_estimator(
self.task_type, _config, x_p1, y_p1)
with open(
os.path.join(
self.output_dir, '%s-model%d_part%d' %
(self.timestamp, model_cnt, j)),
'wb') as f:
pkl.dump(estimator, f)
if self.task_type in CLS_TASKS:
pred = estimator.predict_proba(x_p2)
n_dim = np.array(pred).shape[1]
if n_dim == 2:
# Binary classificaion
n_dim = 1
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(train) + len(test)
feature_p2 = np.zeros(
(num_samples,
self.ensemble_size * n_dim))
if n_dim == 1:
feature_p2[test,
suc_cnt * n_dim:(suc_cnt + 1) *
n_dim] = pred[:, 1:2]
else:
feature_p2[test,
suc_cnt * n_dim:(suc_cnt + 1) *
n_dim] = pred
else:
pred = estimator.predict(x_p2).reshape(-1, 1)
n_dim = 1
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(train) + len(test)
feature_p2 = np.zeros(
(num_samples,
self.ensemble_size * n_dim))
feature_p2[test, suc_cnt *
n_dim:(suc_cnt + 1) * n_dim] = pred
suc_cnt += 1
model_cnt += 1
# Train model for stacking using the other part of training data
self.meta_learner.fit(feature_p2, y)
return self
def get_feature(self, data, solvers):
# Predict the labels via stacking
feature_p2 = None
model_cnt = 0
suc_cnt = 0
for algo_id in self.stats["include_algorithms"]:
train_list = self.stats[algo_id]['train_data_list']
configs = self.stats[algo_id]['configurations']
for train_node in train_list:
test_node = solvers[algo_id].optimizer['fe'].apply(
data, train_node)
for _ in configs:
if self.base_model_mask[model_cnt] == 1:
for j in range(self.kfold):
with open(
os.path.join(
self.output_dir, '%s-model%d_part%d' %
(self.timestamp, model_cnt, j)),
'rb') as f:
estimator = pkl.load(f)
if self.task_type in CLS_TASKS:
pred = estimator.predict_proba(
test_node.data[0])
n_dim = np.array(pred).shape[1]
if n_dim == 2:
n_dim = 1
if feature_p2 is None:
num_samples = len(test_node.data[0])
feature_p2 = np.zeros(
(num_samples,
self.ensemble_size * n_dim))
# Get average predictions
if n_dim == 1:
feature_p2[:, suc_cnt * n_dim:(suc_cnt + 1) * n_dim] = \
feature_p2[:, suc_cnt * n_dim:(suc_cnt + 1) * n_dim] + pred[:,
1:2] / self.kfold
else:
feature_p2[:, suc_cnt * n_dim:(suc_cnt + 1) * n_dim] = \
feature_p2[:, suc_cnt * n_dim:(suc_cnt + 1) * n_dim] + pred / self.kfold
else:
pred = estimator.predict(
test_node.data[0]).reshape(-1, 1)
n_dim = 1
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(test_node.data[0])
feature_p2 = np.zeros(
(num_samples,
self.ensemble_size * n_dim))
# Get average predictions
feature_p2[:, suc_cnt * n_dim:(suc_cnt + 1) * n_dim] = \
feature_p2[:, suc_cnt * n_dim:(suc_cnt + 1) * n_dim] + pred / self.kfold
suc_cnt += 1
model_cnt += 1
return feature_p2
def predict(self, data, solvers):
feature_p2 = self.get_feature(data, solvers)
# Get predictions from meta-learner
if self.task_type in CLS_TASKS:
final_pred = self.meta_learner.predict_proba(feature_p2)
else:
final_pred = self.meta_learner.predict(feature_p2)
return final_pred
class Stacking(BaseEnsembleModel):
def __init__(self,
model_info,
ensemble_size,
task_type,
metric,
evaluator,
model_type='ml',
meta_learner='xgboost',
kfold=3,
save_dir=None,
random_state=None):
super().__init__(model_info=model_info,
ensemble_size=ensemble_size,
task_type=task_type,
metric=metric,
evaluator=evaluator,
model_type=model_type,
save_dir=save_dir,
random_state=random_state)
self.kfold = kfold
# We use Xgboost as default meta-learner
if self.task_type == CLASSIFICATION:
if meta_learner == 'logistic':
from sklearn.linear_model.logistic import LogisticRegression
self.meta_learner = LogisticRegression(max_iter=1000)
elif meta_learner == 'gb':
from sklearn.ensemble.gradient_boosting import GradientBoostingClassifier
self.meta_learner = GradientBoostingClassifier(
learning_rate=0.05,
subsample=0.7,
max_depth=4,
n_estimators=250)
elif meta_learner == 'xgboost':
from xgboost import XGBClassifier
self.meta_learner = XGBClassifier(max_depth=4,
learning_rate=0.05,
n_estimators=150)
elif self.task_type == REGRESSION:
if meta_learner == 'linear':
from sklearn.linear_model import LinearRegression
self.meta_learner = LinearRegression()
elif meta_learner == 'xgboost':
from xgboost import XGBRegressor
self.meta_learner = XGBRegressor(max_depth=4,
learning_rate=0.05,
n_estimators=70)
def fit(self, dm: DataManager):
# Split training data for phase 1 and phase 2
if self.task_type == CLASSIFICATION:
kf = StratifiedKFold(n_splits=self.kfold)
elif self.task_type == REGRESSION:
kf = KFold(n_splits=self.kfold)
feature_p2 = None
if self.model_type == 'ml':
# Train basic models using a part of training data
for i, config in enumerate(self.config_list):
for j, (train,
test) in enumerate(kf.split(dm.train_X, dm.train_y)):
x_p1, x_p2, y_p1, _ = dm.train_X[train], dm.train_X[
test], dm.train_y[train], dm.train_y[test]
estimator = self.get_estimator(config, x_p1, y_p1)
# The final list will contain self.kfold * self.ensemble_size models
self.ensemble_models.append(estimator)
pred = self.get_proba_predictions(estimator, x_p2)
if self.task_type == CLASSIFICATION:
n_dim = np.array(pred).shape[1]
if n_dim == 2:
# Binary classificaion
n_dim = 1
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(train) + len(test)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
if n_dim == 1:
feature_p2[test,
i * n_dim:(i + 1) * n_dim] = pred[:,
1:2]
else:
feature_p2[test, i * n_dim:(i + 1) * n_dim] = pred
elif self.task_type == REGRESSION:
shape = np.array(pred).shape
n_dim = shape[1]
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(train) + len(test)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
feature_p2[test, i * n_dim:(i + 1) * n_dim] = pred
# Train model for stacking using the other part of training data
self.meta_learner.fit(feature_p2, dm.train_y)
elif self.model_type == 'dl':
pass
return self
def get_feature(self, X):
# Predict the labels via stacking
feature_p2 = None
for i, model in enumerate(self.ensemble_models):
pred = self.get_proba_predictions(model, X)
if self.task_type == CLASSIFICATION:
n_dim = np.array(pred).shape[1]
if n_dim == 2:
n_dim = 1
if feature_p2 is None:
num_samples = len(X)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
index = i % self.kfold
# Get average predictions
if n_dim == 1:
feature_p2[:, index * n_dim:(index + 1) * n_dim] = feature_p2[:,
index * n_dim:(index + 1) * n_dim] + \
pred[:, 1:2] / self.kfold
else:
feature_p2[:, index * n_dim:(index + 1) * n_dim] = feature_p2[:,
index * n_dim:(index + 1) * n_dim] + \
pred / self.kfold
elif self.task_type == REGRESSION:
shape = np.array(pred).shape
n_dim = shape[1]
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(X)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
index = i % self.kfold
# Get average predictions
feature_p2[:, index * n_dim:(index + 1) * n_dim] = feature_p2[:,
index * n_dim:(index + 1) * n_dim] + \
pred / self.kfold
return feature_p2
def predict(self, X):
feature_p2 = self.get_feature(X)
# Get predictions from meta-learner
final_pred = self.meta_learner.predict(feature_p2)
return final_pred
def predict_proba(self, X):
feature_p2 = self.get_feature(X)
# Get predictions from meta-learner
final_pred = self.meta_learner.predict_proba(feature_p2)
return final_pred
class Blending(BaseEnsembleModel):
def __init__(self,
model_info,
ensemble_size,
task_type,
metric,
evaluator,
model_type='ml',
meta_learner='xgboost'):
super().__init__(model_info, ensemble_size, task_type, metric,
evaluator, model_type)
# We use Xgboost as default meta-learner
if self.task_type == CLASSIFICATION:
if meta_learner == 'logistic':
from sklearn.linear_model.logistic import LogisticRegression
self.meta_learner = LogisticRegression(max_iter=1000)
elif meta_learner == 'gb':
self.meta_learner = GradientBoostingClassifier(
learning_rate=0.05,
subsample=0.7,
max_depth=4,
n_estimators=250)
elif meta_learner == 'xgboost':
from xgboost import XGBClassifier
self.meta_learner = XGBClassifier(max_depth=4,
learning_rate=0.05,
n_estimators=150)
elif self.task_type == REGRESSION:
if meta_learner == 'linear':
from sklearn.linear_model import LinearRegression
self.meta_learner = LinearRegression()
elif meta_learner == 'xgboost':
from xgboost import XGBRegressor
self.meta_learner = XGBRegressor(max_depth=4,
learning_rate=0.05,
n_estimators=70)
def fit(self, dm: DataManager):
# Split training data for phase 1 and phase 2
if self.task_type == CLASSIFICATION:
x_p1, x_p2, y_p1, y_p2 = train_test_split(dm.train_X,
dm.train_y,
test_size=0.2,
stratify=dm.train_y)
elif self.task_type == REGRESSION:
x_p1, x_p2, y_p1, y_p2 = train_test_split(dm.train_X,
dm.train_y,
test_size=0.2)
feature_p2 = None
if self.model_type == 'ml':
# Train basic models using a part of training data
for i, config in enumerate(self.config_list):
estimator = self.get_estimator(config, x_p1, y_p1)
self.ensemble_models.append(estimator)
pred = self.get_proba_predictions(estimator, x_p2)
if self.task_type == CLASSIFICATION:
n_dim = np.array(pred).shape[1]
if n_dim == 2:
# Binary classificaion
n_dim = 1
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(x_p2)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
if n_dim == 1:
feature_p2[:, i * n_dim:(i + 1) * n_dim] = pred[:, 1:2]
else:
feature_p2[:, i * n_dim:(i + 1) * n_dim] = pred
elif self.task_type == REGRESSION:
shape = np.array(pred).shape
n_dim = shape[1]
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(x_p2)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
feature_p2[:, i * n_dim:(i + 1) * n_dim] = pred
# Train model for blending using the other part of training data
self.meta_learner.fit(feature_p2, y_p2)
elif self.model_type == 'dl':
pass
return self
def get_feature(self, X):
# Predict the labels via blending
feature_p2 = None
for i, model in enumerate(self.ensemble_models):
pred = self.get_proba_predictions(model, X)
if self.task_type == CLASSIFICATION:
n_dim = np.array(pred).shape[1]
if n_dim == 2:
# Binary classificaion
n_dim = 1
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(X)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
if n_dim == 1:
feature_p2[:, i * n_dim:(i + 1) * n_dim] = pred[:, 1:2]
else:
feature_p2[:, i * n_dim:(i + 1) * n_dim] = pred
elif self.task_type == REGRESSION:
n_dim = np.array(pred).shape[1]
# Initialize training matrix for phase 2
if feature_p2 is None:
num_samples = len(X)
feature_p2 = np.zeros(
(num_samples, self.ensemble_size * n_dim))
feature_p2[:, i * n_dim:(i + 1) * n_dim] = pred
return feature_p2
def predict(self, X):
feature_p2 = self.get_feature(X)
# Get predictions from meta-learner
final_pred = self.meta_learner.predict(feature_p2)
return final_pred
def predict_proba(self, X):
feature_p2 = self.get_feature(X)
# Get predictions from meta-learner
final_pred = self.meta_learner.predict_proba(feature_p2)
return final_pred
## -------------------------------------
## SHAP values
## -------------------------------------
print("Getting SHAP values")
## fit the full regression function
cc_all = (np.sum(np.isnan(data.x_train), axis=1) == 0)
cc_all_test = (np.sum(np.isnan(data.x_test), axis=1) == 0)
start = time.time()
ensemble.fit(data.x_train[cc_all, :], np.ravel(data.y_train[cc_all]))
## print test-set error
if args.measure == "auc":
if 'nn' in args.estimator_type:
test_preds = np.mean(ensemble.transform(data.x_test[cc_all_test, :]),
axis=1)
else:
test_preds = ensemble.predict_proba(data.x_test[cc_all_test, :])
else:
test_preds = ensemble.predict(data.x_test[cc_all_test, :])
log_lik = (-1) * sl_scorer(y_true=np.ravel(data.y_test[cc_all_test]),
y_pred=test_preds,
normalize=False)
print('Estimated negative log likelihood: ' + str(log_lik))
if "tree" in args.estimator_type:
explainer = shap.TreeExplainer(ensemble)
shap_values = explainer.shap_values(data.x_test[cc_all_test, :])
else:
if args.measure == "auc":
explainer = shap.KernelExplainer(
ensemble.transform, shap.kmeans(data.x_train[cc_all, :], 100))
tmp = explainer.shap_values(data.x_test[cc_all_test, :], nsample=500)
