Fix coef_ and intercept_ signature to be compatible with sklearn.RFECV (#3873)

* Fix coef_ and intercept_ signature to be compatible with sklearn.RFECV

* Fix lint

* Fix lint

python-package/xgboost/sklearn.py

@@ -535,13 +535,21 @@ class XGBModel(XGBModelBase):
 
         Returns
         -------
-        coef_ : array of shape ``[n_features]``
+        coef_ : array of shape ``[n_features]`` or ``[n_classes, n_features]``
         """
         if self.booster != 'gblinear':
             raise AttributeError('Coefficients are not defined for Booster type {}'
                                  .format(self.booster))
         b = self.get_booster()
-        return json.loads(b.get_dump(dump_format='json')[0])['weight']
+        coef = np.array(json.loads(b.get_dump(dump_format='json')[0])['weight'])
+        # Logic for multiclass classification
+        n_classes = getattr(self, 'n_classes_', None)
+        if n_classes is not None:
+            if n_classes > 2:
+                assert len(coef.shape) == 1
+                assert coef.shape[0] % n_classes == 0
+                coef = coef.reshape((n_classes, -1))
+        return coef
 
     @property
     def intercept_(self):
@@ -556,13 +564,13 @@ class XGBModel(XGBModelBase):
 
         Returns
         -------
-        intercept_ : array of shape ``[n_features]``
+        intercept_ : array of shape ``(1,)`` or ``[n_classes]``
         """
         if self.booster != 'gblinear':
             raise AttributeError('Intercept (bias) is not defined for Booster type {}'
                                  .format(self.booster))
         b = self.get_booster()
-        return json.loads(b.get_dump(dump_format='json')[0])['bias']
+        return np.array(json.loads(b.get_dump(dump_format='json')[0])['bias'])
 
 
 class XGBClassifier(XGBModel, XGBClassifierBase):

tests/python/test_with_sklearn.py

@@ -544,3 +544,36 @@ def test_save_load_model():
             err = sum(1 for i in range(len(preds))
                       if int(preds[i] > 0.5) != labels[i]) / float(len(preds))
             assert err < 0.1
+
+
+def test_RFECV():
+    tm._skip_if_no_sklearn()
+    from sklearn.datasets import load_boston
+    from sklearn.datasets import load_breast_cancer
+    from sklearn.datasets import load_iris
+    from sklearn.feature_selection import RFECV
+
+    # Regression
+    X, y = load_boston(return_X_y=True)
+    bst = xgb.XGBClassifier(booster='gblinear', learning_rate=0.1,
+                            n_estimators=10, n_jobs=1, objective='reg:linear',
+                            random_state=0, silent=True)
+    rfecv = RFECV(estimator=bst, step=1, cv=3, scoring='neg_mean_squared_error')
+    rfecv.fit(X, y)
+
+    # Binary classification
+    X, y = load_breast_cancer(return_X_y=True)
+    bst = xgb.XGBClassifier(booster='gblinear', learning_rate=0.1,
+                            n_estimators=10, n_jobs=1, objective='binary:logistic',
+                            random_state=0, silent=True)
+    rfecv = RFECV(estimator=bst, step=1, cv=3, scoring='roc_auc')
+    rfecv.fit(X, y)
+
+    # Multi-class classification
+    X, y = load_iris(return_X_y=True)
+    bst = xgb.XGBClassifier(base_score=0.4, booster='gblinear', learning_rate=0.1,
+                            n_estimators=10, n_jobs=1, objective='multi:softprob',
+                            random_state=0, reg_alpha=0.001, reg_lambda=0.01,
+                            scale_pos_weight=0.5, silent=True)
+    rfecv = RFECV(estimator=bst, step=1, cv=3, scoring='neg_log_loss')
+    rfecv.fit(X, y)