def fit_transform(self, x, y):
self.fit(x, y, use_oof=True)
predictions = np.zeros((x.shape[0], ))
for i, (train_index, test_index) in enumerate(self.indices):
predictions[test_index] = self.fitted_models[i].predict(
x[test_index])
self.cv_scores = [
scoring_metric(y[test_index], predictions[test_index])**0.5
for i, (train_index, test_index) in enumerate(self.indices)
]
self.avg_cv_score = np.mean(self.cv_scores)
self.overall_cv_score = scoring_metric(y, predictions)**0.5
return predictions
def fit_transform(self, x, y):
self.fit(x, y, use_oof=True)
predictions = np.zeros((x.shape[0], ))
for i, (train_index, test_index) in enumerate(self.indices):
if isinstance(self.model, LogisticRegression):
predictions[test_index] = self.fitted_models[i].predict_proba(
self.scaler_models[i].transform(x[test_index]))[:, 1]
else:
predictions[test_index] = self.fitted_models[i].predict(
x[test_index])
self.cv_scores = [
scoring_metric(y[test_index], predictions[test_index])**0.5
for i, (train_index, test_index) in enumerate(self.indices)
]
self.avg_cv_score = np.mean(self.cv_scores)
self.overall_cv_score = scoring_metric(y, predictions)
return predictions
def fit_transform(self, x, y):
self.fit(x, y, use_oof=True)
predictions = np.zeros((x.shape[0],))
for i, (train_index, test_index) in enumerate(self.indices):
if self.model.__class__.__name__ == 'LogisticRegression':
predictions[test_index] = self.fitted_models[i].predict_proba(x[test_index])[:,1]
elif self.model.__class__.__name__ == 'SVC' or self.model.__class__.__name__ == 'SVR':
predictions[test_index] = self.fitted_models[i].predict(x[test_index])
else:
predictions[test_index] = self.fitted_models[i].predict_proba(x[test_index])[:,1]
self.cv_scores = [
scoring_metric(y[test_index], predictions[test_index])
for i, (train_index, test_index) in enumerate(self.indices)
]
self.avg_cv_score = np.mean(self.cv_scores)
self.overall_cv_score = scoring_metric(y, predictions)
return predictions
def fit(self, x, y, use_oof=False, n_jobs=-1):
if not hasattr(self.model, 'fit'):
raise Exception("Model/algorithm needs to implement fit()")
fitted_models = []
if use_oof:
# kf = StratifiedKFold(n_splits=self.kfolds, random_state=self.random_state, shuffle=self.shuffle)
# self.indices = [(train_index, test_index) for (train_index, test_index) in kf.split(x, y)]
folds = CustomFolds(num_folds=self.n_splits,
random_state=self.random_state,
shuffle=self.shuffle,
validation_scheme=self.validation_scheme)
self.indices = folds.split(x, y, group=self.cv_group_col)
for i, (train_index, test_index) in enumerate(self.indices):
model = clone(self.model)
model.n_jobs = n_jobs
if (isinstance(model, LGBMRegressor)
and self.early_stopping_rounds is not None):
model.fit(X=x[train_index],
y=y[train_index],
eval_set=[(x[test_index], y[test_index]),
(x[train_index], y[train_index])],
verbose=100,
eval_metric='rmse',
early_stopping_rounds=self.early_stopping_rounds)
elif (isinstance(model, XGBRegressor)
and self.early_stopping_rounds is not None):
model.fit(X=x[train_index],
y=y[train_index],
eval_set=[(x[test_index], y[test_index])],
verbose=100,
eval_metric='rmse',
early_stopping_rounds=self.early_stopping_rounds)
else:
# model.n_jobs=-1
model.fit(x[train_index], y[train_index])
print scoring_metric(y[test_index],
model.predict(x[test_index]))**0.5
fitted_models.append(model)
else:
model = clone(self.model)
model.n_jobs = n_jobs
x_train, x_val, y_train, y_val = train_test_split(x,
y,
test_size=0.2,
shuffle=True,
random_state=100)
if isinstance(model, LGBMRegressor):
if self.early_stopping_rounds is not None:
model.fit(X=x_train,
y=y_train,
eval_set=[(x_val, y_val)],
verbose=False,
eval_metric='rmse',
early_stopping_rounds=self.early_stopping_rounds)
elif isinstance(model, XGBRegressor):
if self.early_stopping_rounds is not None:
model.fit(X=x_train,
y=y_train,
eval_set=[(x_val, y_val)],
verbose=False,
eval_metric='rmse',
early_stopping_rounds=self.early_stopping_rounds)
model.fit(x, y)
fitted_models.append(model)
self.fitted_models = fitted_models
return self