def with_params(pipeline, params):
for param in params:
pipeline.set_params(**{
pipeline_utils.find_xgbmodel_param_prefix(pipeline)[0] + param: params[param]
})
print(pipeline_utils.get_final_estimator(pipeline).get_params())
return pipeline
def fit(self, X, y):
n_samples = X.shape[0]
# if isinstance(self.hold_out_size, int):
# self.hold_out_size = self.hold_out_size / n_samples
# X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=self.hold_out_size,
# random_state=self.rng)
# print(type(X))
# self.hold_out_size = self.hold_out_size * n_samples
# print(self.hold_out_size)
X_train_idx, X_test_idx = train_test_split(
X.index, test_size=self.hold_out_size, random_state=self.rng)
self._best_its = []
for union, estimator in deepcopy(self.union_estimator_tuples):
# if not isinstance(estimator, xgb.XGBModel):
# raise ValueError('best_iteration_ can only be determined for XGBModel. Given {} instead.'
# .format(estimator.__class__.__name__))
xgb_estimator_path, _ = pipeline_utils.find_xgbmodel_param_prefix(
estimator)
# print(xgb_estimator_path)
# raise SystemExit(1)
# print(X_train.shape, X_test.shape)
X_train, X_test = X.ix[X_train_idx], X.ix[X_test_idx]
y_train, y_test = y.ix[X_train_idx], y.ix[X_test_idx]
X_train = union.fit_transform(X_train, y_train)
X_test = union.transform(X_test)
eval_set = [(X_test, y_test)]
fit_params = {
xgb_estimator_path + 'eval_set': eval_set,
xgb_estimator_path + 'early_stopping_rounds':
self.early_stopping_rounds,
xgb_estimator_path + 'eval_metric': self.eval_metric,
xgb_estimator_path + 'maximize_score': self.maximize_score,
xgb_estimator_path + 'verbose': self.verbose_eval
}
estimator.fit(X_train, y_train, **fit_params)
# eval_set=eval_set,
# early_stopping_rounds=self.early_stopping_rounds)
if isinstance(estimator, Pipeline):
final_ = pipeline_utils.get_final_estimator(estimator)
else:
final_ = estimator
self._best_its.append(final_.best_iteration)
return self
from src.xgb_early import XGBEarlyStop
from kaggle_tools.metrics import xgb_normalized_gini
early_stop = XGBEarlyStop([(union, estimator)],
hold_out_size=0.25,
maximize_score=True,
eval_metric=xgb_normalized_gini,
random_state=2,
early_stopping_rounds=80)
early_stop.fit(dataset, target)
best_iteration = early_stop.best_iteration_[0]
from kaggle_tools.utils import pipeline_utils
pipeline = overall_pipeline()
pipeline.set_params(
**{
pipeline_utils.find_xgbmodel_param_prefix(pipeline)[0] + 'n_estimators':
best_iteration
})
print(pipeline_utils.get_final_estimator(pipeline).get_params())
cv = KFold(len(target), n_folds=4, random_state=2, shuffle=False)
submission = SqrtHazardSubmission(pipeline, 'XGB_Direct_Mean', cv=cv)
submission.fit(dataset,
target,
perform_cv=True,
scoring=scorer_normalized_gini,
n_jobs=2,
verbose=3)
print('fitted. time:', time.time() - before)