refine early stopping and add a test case (#369)

python-package/lightgbm/callback.py

@@ -201,7 +201,6 @@ def callback(env):
                             env.iteration + 1, '\t'.join([_format_eval_result(x) for x in env.evaluation_result_list]))
                     best_msg[i] = best_msg_buffer
             elif env.iteration - best_iter[i] >= stopping_rounds:
-                env.model.set_attr(best_iteration=str(best_iter[i]))
                 if verbose:
                     print('Early stopping, best iteration is:\n' + best_msg[i])
                 raise EarlyStopException(best_iter[i])

python-package/lightgbm/engine.py

@@ -165,6 +165,7 @@ def train(params, train_set, num_boost_round=100,
         booster.set_train_data_name(train_data_name)
     for valid_set, name_valid_set in zip(reduced_valid_sets, name_valid_sets):
         booster.add_valid(valid_set, name_valid_set)
+    booster.best_iteration = -1
 
     """start training"""
     for i in range_(init_iteration, init_iteration + num_boost_round):
@@ -192,19 +193,17 @@ def train(params, train_set, num_boost_round=100,
                                         begin_iteration=init_iteration,
                                         end_iteration=init_iteration + num_boost_round,
                                         evaluation_result_list=evaluation_result_list))
-        except callback.EarlyStopException:
+        except callback.EarlyStopException as earlyStopException:
+            booster.best_iteration = earlyStopException.best_iteration + 1
             break
-    if booster.attr('best_iteration') is not None:
-        booster.best_iteration = int(booster.attr('best_iteration')) + 1
-    else:
-        booster.best_iteration = -1
     return booster
 
 
 class CVBooster(object):
     """"Auxiliary data struct to hold all boosters of CV."""
     def __init__(self):
         self.boosters = []
+        self.best_iteration = -1
 
     def append(self, booster):
         """add a booster to CVBooster"""
@@ -408,8 +407,9 @@ def cv(params, train_set, num_boost_round=10,
                                         begin_iteration=0,
                                         end_iteration=num_boost_round,
                                         evaluation_result_list=res))
-        except callback.EarlyStopException as e:
+        except callback.EarlyStopException as earlyStopException:
+            cvfolds.best_iteration = earlyStopException.best_iteration + 1
             for k in results:
-                results[k] = results[k][:e.best_iteration + 1]
+                results[k] = results[k][:cvfolds.best_iteration]
             break
     return dict(results)

tests/python_package_test/test_engine.py

@@ -86,6 +86,32 @@ def test_multiclass(self):
         self.assertLess(ret, 0.2)
         self.assertAlmostEqual(min(evals_result['eval']['multi_logloss']), ret, places=5)
 
+    def test_early_stopping(self):
+        X_y = load_breast_cancer(True)
+        params = {
+            'objective': 'binary',
+            'metric': 'binary_logloss',
+            'verbose': -1,
+            'seed': 42
+        }
+        X_train, X_test, y_train, y_test = train_test_split(*X_y, test_size=0.1, random_state=42)
+        lgb_train = lgb.Dataset(X_train, y_train)
+        lgb_eval = lgb.Dataset(X_test, y_test, reference=lgb_train)
+        # no early stopping
+        gbm = lgb.train(params, lgb_train,
+                        num_boost_round=10,
+                        valid_sets=lgb_eval,
+                        verbose_eval=False,
+                        early_stopping_rounds=5)
+        self.assertEqual(gbm.best_iteration, -1)
+        # early stopping occurs
+        gbm = lgb.train(params, lgb_train,
+                        num_boost_round=100,
+                        valid_sets=lgb_eval,
+                        verbose_eval=False,
+                        early_stopping_rounds=5)
+        self.assertLessEqual(gbm.best_iteration, 100)
+
     def test_continue_train_and_other(self):
         params = {
             'objective': 'regression',