Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix divide by 0 in feature importance when no split is found. (#6676)

Browse Source
This commit is contained in:
Jiaming Yuan 2021-02-05 03:39:30 +08:00 committed by GitHub
parent 72892cc80d
commit a4101de678
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 24 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
python-package/xgboost/sklearn.py
View File

@ -920,7 +920,10 @@ class XGBModel(XGBModelBase):
score = b.get_score(importance_type=self.importance_type) score = b.get_score(importance_type=self.importance_type)
all_features = [score.get(f, 0.) for f in b.feature_names] all_features = [score.get(f, 0.) for f in b.feature_names]
all_features = np.array(all_features, dtype=np.float32) all_features = np.array(all_features, dtype=np.float32)
return all_features / all_features.sum() total = all_features.sum()
if total == 0:
return all_features
return all_features / total
@property @property
def coef_(self): def coef_(self):

25
tests/python/test_with_sklearn.py
View File

@ -252,7 +252,9 @@ def test_feature_importances_gain():
xgb_model = xgb.XGBClassifier( xgb_model = xgb.XGBClassifier(
random_state=0, tree_method="exact", random_state=0, tree_method="exact",
learning_rate=0.1, learning_rate=0.1,
importance_type="gain").fit(X, y) importance_type="gain",
use_label_encoder=False,
).fit(X, y)
exp = np.array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., exp = np.array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0.00326159, 0., 0., 0., 0., 0., 0., 0., 0., 0.00326159, 0., 0., 0., 0., 0., 0., 0., 0.,
@ -270,17 +272,30 @@ def test_feature_importances_gain():
y = pd.Series(digits['target']) y = pd.Series(digits['target'])
X = pd.DataFrame(digits['data']) X = pd.DataFrame(digits['data'])
xgb_model = xgb.XGBClassifier( xgb_model = xgb.XGBClassifier(
random_state=0, tree_method="exact", random_state=0,
tree_method="exact",
learning_rate=0.1, learning_rate=0.1,
importance_type="gain").fit(X, y) importance_type="gain",
use_label_encoder=False,
).fit(X, y)
np.testing.assert_almost_equal(xgb_model.feature_importances_, exp) np.testing.assert_almost_equal(xgb_model.feature_importances_, exp)
xgb_model = xgb.XGBClassifier( xgb_model = xgb.XGBClassifier(
random_state=0, tree_method="exact", random_state=0,
tree_method="exact",
learning_rate=0.1, learning_rate=0.1,
importance_type="gain").fit(X, y) importance_type="gain",
use_label_encoder=False,
).fit(X, y)
np.testing.assert_almost_equal(xgb_model.feature_importances_, exp) np.testing.assert_almost_equal(xgb_model.feature_importances_, exp)
# no split can be found
cls = xgb.XGBClassifier(
min_child_weight=1000, tree_method="hist", n_estimators=1, use_label_encoder=False
)
cls.fit(X, y)
assert np.all(cls.feature_importances_ == 0)
def test_select_feature(): def test_select_feature():
from sklearn.datasets import load_digits from sklearn.datasets import load_digits