From eb1efb54b5a5010b53bf4405eac81d49e1aeba44 Mon Sep 17 00:00:00 2001
From: Jiaming Yuan <jm.yuan@outlook.com>
Date: Wed, 5 Jan 2022 01:35:34 +0800
Subject: [PATCH] Define `feature_names_in_`. (#7526)

* Define `feature_names_in_`.
* Raise attribute error if it's not defined.
---
 python-package/xgboost/sklearn.py | 12 ++++++++++++
 tests/python/test_with_sklearn.py |  5 +++++
 2 files changed, 17 insertions(+)

diff --git a/python-package/xgboost/sklearn.py b/python-package/xgboost/sklearn.py
index a25890606..b2db142fa 100644
--- a/python-package/xgboost/sklearn.py
+++ b/python-package/xgboost/sklearn.py
@@ -1105,6 +1105,18 @@ class XGBModel(XGBModelBase):
         booster = self.get_booster()
         return booster.num_features()
 
+    @property
+    def feature_names_in_(self) -> np.ndarray:
+        """Names of features seen during :py:meth:`fit`.  Defined only when `X` has feature
+        names that are all strings."""
+        feature_names = self.get_booster().feature_names
+        if feature_names is None:
+            raise AttributeError(
+                "`feature_names_in_` is defined only when `X` has feature names that "
+                "are all strings."
+            )
+        return np.array(feature_names)
+
     def _early_stopping_attr(self, attr: str) -> Union[float, int]:
         booster = self.get_booster()
         try:
diff --git a/tests/python/test_with_sklearn.py b/tests/python/test_with_sklearn.py
index 83c73932b..aef4657ea 100644
--- a/tests/python/test_with_sklearn.py
+++ b/tests/python/test_with_sklearn.py
@@ -372,6 +372,9 @@ def test_boston_housing_regression():
         assert mean_squared_error(preds3, labels) < 25
         assert mean_squared_error(preds4, labels) < 350
 
+        with pytest.raises(AttributeError, match="feature_names_in_"):
+            xgb_model.feature_names_in_
+
 
 def run_boston_housing_rf_regression(tree_method):
     from sklearn.metrics import mean_squared_error
@@ -1017,6 +1020,8 @@ def test_pandas_input():
     train = df.drop(columns=['status'])
     model = xgb.XGBClassifier()
     model.fit(train, target)
+    np.testing.assert_equal(model.feature_names_in_, np.array(feature_names))
+
     clf_isotonic = CalibratedClassifierCV(model,
                                           cv='prefit', method='isotonic')
     clf_isotonic.fit(train, target)