[python] Allow python sklearn interface's fit() to pass init_model to train() (#2447)

* allow python sklearn interface's fit() to pass init_model to train()

* Fix whitespace issues, and change ordering of parameters to be backward
compatible

* Formatting fixes

* allow python sklearn interface's fit() to pass init_model to train()

* Fix whitespace issues, and change ordering of parameters to be backward
compatible

* Formatting fixes

* Recognize LGBModel objects for init_model

* simplified condition

* updated docstring

* added test

python-package/lightgbm/sklearn.py

@@ -376,7 +376,8 @@ class LGBMModel(_LGBMModelBase):
             eval_set=None, eval_names=None, eval_sample_weight=None,
             eval_class_weight=None, eval_init_score=None, eval_group=None,
             eval_metric=None, early_stopping_rounds=None, verbose=True,
-            feature_name='auto', categorical_feature='auto', callbacks=None):
+            feature_name='auto', categorical_feature='auto',
+            callbacks=None, init_model=None):
         """Build a gradient boosting model from the training set (X, y).
 
         Parameters
@@ -442,6 +443,8 @@ class LGBMModel(_LGBMModelBase):
         callbacks : list of callback functions or None, optional (default=None)
             List of callback functions that are applied at each iteration.
             See Callbacks in Python API for more information.
+        init_model : string, Booster, LGBMModel or None, optional (default=None)
+            Filename of LightGBM model, Booster instance or LGBMModel instance used for continue training.
 
         Returns
         -------
@@ -593,13 +596,16 @@ class LGBMModel(_LGBMModelBase):
                                                    valid_weight, valid_init_score, valid_group, params)
                 valid_sets.append(valid_set)
 
+        if isinstance(init_model, LGBMModel):
+            init_model = init_model.booster_
+
         self._Booster = train(params, train_set,
                               self.n_estimators, valid_sets=valid_sets, valid_names=eval_names,
                               early_stopping_rounds=early_stopping_rounds,
                               evals_result=evals_result, fobj=self._fobj, feval=feval,
                               verbose_eval=verbose, feature_name=feature_name,
                               categorical_feature=categorical_feature,
-                              callbacks=callbacks)
+                              callbacks=callbacks, init_model=init_model)
 
         if evals_result:
             self._evals_result = evals_result
@@ -731,7 +737,8 @@ class LGBMRegressor(LGBMModel, _LGBMRegressorBase):
             sample_weight=None, init_score=None,
             eval_set=None, eval_names=None, eval_sample_weight=None,
             eval_init_score=None, eval_metric=None, early_stopping_rounds=None,
-            verbose=True, feature_name='auto', categorical_feature='auto', callbacks=None):
+            verbose=True, feature_name='auto', categorical_feature='auto',
+            callbacks=None, init_model=None):
         """Docstring is inherited from the LGBMModel."""
         super(LGBMRegressor, self).fit(X, y, sample_weight=sample_weight,
                                        init_score=init_score, eval_set=eval_set,
@@ -742,7 +749,7 @@ class LGBMRegressor(LGBMModel, _LGBMRegressorBase):
                                        early_stopping_rounds=early_stopping_rounds,
                                        verbose=verbose, feature_name=feature_name,
                                        categorical_feature=categorical_feature,
-                                       callbacks=callbacks)
+                                       callbacks=callbacks, init_model=init_model)
         return self
 
     _base_doc = LGBMModel.fit.__doc__
@@ -758,7 +765,8 @@ class LGBMClassifier(LGBMModel, _LGBMClassifierBase):
             eval_set=None, eval_names=None, eval_sample_weight=None,
             eval_class_weight=None, eval_init_score=None, eval_metric=None,
             early_stopping_rounds=None, verbose=True,
-            feature_name='auto', categorical_feature='auto', callbacks=None):
+            feature_name='auto', categorical_feature='auto',
+            callbacks=None, init_model=None):
         """Docstring is inherited from the LGBMModel."""
         _LGBMAssertAllFinite(y)
         _LGBMCheckClassificationTargets(y)
@@ -804,7 +812,7 @@ class LGBMClassifier(LGBMModel, _LGBMClassifierBase):
                                         early_stopping_rounds=early_stopping_rounds,
                                         verbose=verbose, feature_name=feature_name,
                                         categorical_feature=categorical_feature,
-                                        callbacks=callbacks)
+                                        callbacks=callbacks, init_model=init_model)
         return self
 
     fit.__doc__ = LGBMModel.fit.__doc__
@@ -896,7 +904,8 @@ class LGBMRanker(LGBMModel):
             eval_set=None, eval_names=None, eval_sample_weight=None,
             eval_init_score=None, eval_group=None, eval_metric=None,
             eval_at=[1], early_stopping_rounds=None, verbose=True,
-            feature_name='auto', categorical_feature='auto', callbacks=None):
+            feature_name='auto', categorical_feature='auto',
+            callbacks=None, init_model=None):
         """Docstring is inherited from the LGBMModel."""
         # check group data
         if group is None:
@@ -924,7 +933,7 @@ class LGBMRanker(LGBMModel):
                                     early_stopping_rounds=early_stopping_rounds,
                                     verbose=verbose, feature_name=feature_name,
                                     categorical_feature=categorical_feature,
-                                    callbacks=callbacks)
+                                    callbacks=callbacks, init_model=init_model)
         return self
 
     _base_doc = LGBMModel.fit.__doc__

tests/python_package_test/test_sklearn.py

@@ -790,3 +790,16 @@ class TestSklearn(unittest.TestCase):
             for metric in gbm.evals_result_[eval_set]:
                 np.testing.assert_allclose(gbm.evals_result_[eval_set][metric],
                                            gbm_str.evals_result_[eval_set][metric])
+
+    def test_continue_training_with_model(self):
+        X, y = load_digits(3, True)
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
+        init_gbm = lgb.LGBMClassifier(n_estimators=5).fit(X_train, y_train, eval_set=(X_test, y_test),
+                                                          verbose=False)
+        gbm = lgb.LGBMClassifier(n_estimators=5).fit(X_train, y_train, eval_set=(X_test, y_test),
+                                                     verbose=False, init_model=init_gbm)
+        self.assertEqual(len(init_gbm.evals_result_['valid_0']['multi_logloss']),
+                         len(gbm.evals_result_['valid_0']['multi_logloss']))
+        self.assertEqual(len(init_gbm.evals_result_['valid_0']['multi_logloss']), 5)
+        self.assertLess(gbm.evals_result_['valid_0']['multi_logloss'][-1],
+                        init_gbm.evals_result_['valid_0']['multi_logloss'][-1])