[python][scikit-learn] change MRO (#3192)

* chanche MRO

* fix MRO resolution

python-package/lightgbm/compat.py

@@ -94,9 +94,22 @@ def _check_sample_weight(sample_weight, X, dtype=None):
     _LGBMComputeSampleWeight = compute_sample_weight
 except ImportError:
     SKLEARN_INSTALLED = False
-    _LGBMModelBase = object
-    _LGBMClassifierBase = object
-    _LGBMRegressorBase = object
+
+    class _LGBMModelBase:  # type: ignore
+        """Dummy class for sklearn.base.BaseEstimator."""
+
+        pass
+
+    class _LGBMClassifierBase:  # type: ignore
+        """Dummy class for sklearn.base.ClassifierMixin."""
+
+        pass
+
+    class _LGBMRegressorBase:  # type: ignore
+        """Dummy class for sklearn.base.RegressorMixin."""
+
+        pass
+
     _LGBMLabelEncoder = None
     LGBMNotFittedError = ValueError
     _LGBMStratifiedKFold = None
@@ -118,11 +131,16 @@ def _check_sample_weight(sample_weight, X, dtype=None):
     DASK_INSTALLED = True
 except ImportError:
     DASK_INSTALLED = False
+
     delayed = None
-    Client = object
     default_client = None
     wait = None
 
+    class Client:  # type: ignore
+        """Dummy class for dask.distributed.Client."""
+
+        pass
+
     class dask_Array:  # type: ignore
         """Dummy class for dask.array.Array."""
 

python-package/lightgbm/sklearn.py

@@ -804,7 +804,7 @@ def feature_name_(self):
         return self._Booster.feature_name()
 
 
-class LGBMRegressor(LGBMModel, _LGBMRegressorBase):
+class LGBMRegressor(_LGBMRegressorBase, LGBMModel):
     """LightGBM regressor."""
 
     def fit(self, X, y,
@@ -830,7 +830,7 @@ def fit(self, X, y,
                    + _base_doc[_base_doc.find('eval_metric :'):])
 
 
-class LGBMClassifier(LGBMModel, _LGBMClassifierBase):
+class LGBMClassifier(_LGBMClassifierBase, LGBMModel):
     """LightGBM classifier."""
 
     def fit(self, X, y,