def fetch_predict_estimator(task_type,
config,
X_train,
y_train,
weight_balance=0,
data_balance=0,
combined=False):
# Build the ML estimator.
from solnml.components.utils.balancing import get_weights, smote
_fit_params = {}
config_dict = config.get_dictionary().copy()
if weight_balance == 1:
_init_params, _fit_params = get_weights(y_train, config['estimator'],
None, {}, {})
for key, val in _init_params.items():
config_dict[key] = val
if data_balance == 1:
X_train, y_train = smote(X_train, y_train)
if task_type in CLS_TASKS:
if combined:
from solnml.utils.combined_evaluator import get_estimator
else:
from solnml.components.evaluators.cls_evaluator import get_estimator
else:
from solnml.components.evaluators.reg_evaluator import get_estimator
_, estimator = get_estimator(config_dict)
estimator.fit(X_train, y_train, **_fit_params)
return estimator
def holdout_validation(estimator,
scorer,
X,
y,
test_size=0.33,
fit_params=None,
if_stratify=True,
onehot=None,
random_state=1):
with warnings.catch_warnings():
# ignore all caught warnings
warnings.filterwarnings("ignore")
if if_stratify:
ss = StratifiedShuffleSplit(n_splits=1,
test_size=test_size,
random_state=random_state)
else:
ss = ShuffleSplit(n_splits=1,
test_size=test_size,
random_state=random_state)
for train_index, test_index in ss.split(X, y):
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
_fit_params = dict()
if fit_params:
if 'sample_weight' in fit_params:
_fit_params['sample_weight'] = fit_params['sample_weight'][
train_index]
elif 'data_balance' in fit_params:
X_train, y_train = smote(X_train, y_train)
estimator.fit(X_train, y_train, **_fit_params)
if onehot is not None:
y_test = get_onehot_y(onehot, y_test)
return scorer(estimator, X_test, y_test)
def fetch_predict_estimator(task_type,
estimator_id,
config,
X_train,
y_train,
weight_balance=0,
data_balance=0):
# Build the ML estimator.
from solnml.components.utils.balancing import get_weights, smote
_fit_params = {}
config_dict = config.copy()
if weight_balance == 1:
_init_params, _fit_params = get_weights(y_train, estimator_id, None,
{}, {})
for key, val in _init_params.items():
config_dict[key] = val
if data_balance == 1:
X_train, y_train = smote(X_train, y_train)
if task_type in CLS_TASKS:
from solnml.components.evaluators.cls_evaluator import get_estimator
elif task_type in RGS_TASKS:
from solnml.components.evaluators.rgs_evaluator import get_estimator
_, estimator = get_estimator(config_dict, estimator_id)
estimator.fit(X_train, y_train, **_fit_params)
return estimator
def validation(estimator,
scorer,
X_train,
y_train,
X_val,
y_val,
fit_params=None,
onehot=None,
random_state=1):
with warnings.catch_warnings():
# ignore all caught warnings
warnings.filterwarnings("ignore")
_fit_params = dict()
if fit_params:
if 'sample_weight' in fit_params:
_fit_params['sample_weight'] = fit_params['sample_weight']
elif 'data_balance' in fit_params:
X_train, y_train = smote(X_train, y_train)
estimator.fit(X_train, y_train, **_fit_params)
if onehot is not None:
y_val = get_onehot_y(onehot, y_val)
return scorer(estimator, X_val, y_val)
def cross_validation(estimator,
scorer,
X,
y,
n_fold=5,
shuffle=True,
fit_params=None,
if_stratify=True,
onehot=None,
random_state=1):
with warnings.catch_warnings():
# ignore all caught warnings
warnings.filterwarnings("ignore")
if if_stratify:
kfold = StratifiedKFold(n_splits=n_fold,
random_state=random_state,
shuffle=shuffle)
else:
kfold = KFold(n_splits=n_fold,
random_state=random_state,
shuffle=shuffle)
scores = list()
for train_idx, valid_idx in kfold.split(X, y):
train_x, valid_x = X[train_idx], X[valid_idx]
train_y, valid_y = y[train_idx], y[valid_idx]
_fit_params = dict()
if fit_params:
if 'sample_weight' in fit_params:
_fit_params['sample_weight'] = fit_params['sample_weight'][
train_idx]
elif 'data_balance' in fit_params:
X_train, y_train = smote(X_train, y_train)
estimator.fit(train_x, train_y, **_fit_params)
if onehot is not None:
valid_y = get_onehot_y(onehot, valid_y)
scores.append(scorer(estimator, valid_x, valid_y))
return np.mean(scores)
