[python][scikit-learn] Fixes a bug that prevented using multiple eval_metrics...

[python][scikit-learn] Fixes a bug that prevented using multiple eval_metrics in LGBMClassifier (#3222)

* Fixes a bug that prevented using multiple eval_metrics in LGBMClassifier

* Move bug-fix test to the test_metrics unit-test

* Fix test to avoid issues with existing tests

* Fix coding-style error
Co-authored-by: German I Ramirez-Espinoza <gire@home>

python-package/lightgbm/sklearn.py

@@ -782,20 +782,28 @@ class LGBMClassifier(LGBMModel, _LGBMClassifierBase):
 
         self._classes = self._le.classes_
         self._n_classes = len(self._classes)
+
         if self._n_classes > 2:
             # Switch to using a multiclass objective in the underlying LGBM instance
             ova_aliases = {"multiclassova", "multiclass_ova", "ova", "ovr"}
             if self._objective not in ova_aliases and not callable(self._objective):
                 self._objective = "multiclass"
-            if eval_metric in {'logloss', 'binary_logloss'}:
-                eval_metric = "multi_logloss"
-            elif eval_metric in {'error', 'binary_error'}:
-                eval_metric = "multi_error"
-        else:
-            if eval_metric in {'logloss', 'multi_logloss'}:
-                eval_metric = 'binary_logloss'
-            elif eval_metric in {'error', 'multi_error'}:
-                eval_metric = 'binary_error'
+
+        if not callable(eval_metric):
+            if isinstance(eval_metric, (string_type, type(None))):
+                eval_metric = [eval_metric]
+            if self._n_classes > 2:
+                for index, metric in enumerate(eval_metric):
+                    if metric in {'logloss', 'binary_logloss'}:
+                        eval_metric[index] = "multi_logloss"
+                    elif metric in {'error', 'binary_error'}:
+                        eval_metric[index] = "multi_error"
+            else:
+                for index, metric in enumerate(eval_metric):
+                    if metric in {'logloss', 'multi_logloss'}:
+                        eval_metric[index] = 'binary_logloss'
+                    elif metric in {'error', 'multi_error'}:
+                        eval_metric[index] = 'binary_error'
 
         # do not modify args, as it causes errors in model selection tools
         valid_sets = None

tests/python_package_test/test_sklearn.py

@@ -540,6 +540,20 @@ class TestSklearn(unittest.TestCase):
         self.assertIn('l2', gbm.evals_result_['training'])
         self.assertIn('mape', gbm.evals_result_['training'])
 
+        # non-default metric with multiple metrics in eval_metric for LGBMClassifier
+        X_classification, y_classification = load_breast_cancer(True)
+        params_classification = {'n_estimators': 2, 'verbose': -1,
+                                 'objective': 'binary', 'metric': 'binary_logloss'}
+        params_fit_classification = {'X': X_classification, 'y': y_classification,
+                                     'eval_set': (X_classification, y_classification),
+                                     'verbose': False}
+        gbm = lgb.LGBMClassifier(**params_classification).fit(eval_metric=['fair', 'error'],
+                                                              **params_fit_classification)
+        self.assertEqual(len(gbm.evals_result_['training']), 3)
+        self.assertIn('fair', gbm.evals_result_['training'])
+        self.assertIn('binary_error', gbm.evals_result_['training'])
+        self.assertIn('binary_logloss', gbm.evals_result_['training'])
+
         # default metric for non-default objective
         gbm = lgb.LGBMRegressor(objective='regression_l1', **params).fit(**params_fit)
         self.assertEqual(len(gbm.evals_result_['training']), 1)