[python-package] respect 'verbose' setting when using custom objective function (fixes #6014) (#6428)

src/io/config.cpp

@@ -40,9 +40,24 @@ void GetFirstValueAsInt(const std::unordered_map<std::string, std::vector<std::s
 }
 
 void Config::SetVerbosity(const std::unordered_map<std::string, std::vector<std::string>>& params) {
-  int verbosity = Config().verbosity;
-  GetFirstValueAsInt(params, "verbose", &verbosity);
-  GetFirstValueAsInt(params, "verbosity", &verbosity);
+  int verbosity = 1;
+
+  // if "verbosity" was found in params, prefer that to any other aliases
+  const auto verbosity_iter = params.find("verbosity");
+  if (verbosity_iter != params.end()) {
+    GetFirstValueAsInt(params, "verbosity", &verbosity);
+  } else {
+    // if "verbose" was found in params and "verbosity" was not, use that value
+    const auto verbose_iter = params.find("verbose");
+    if (verbose_iter != params.end()) {
+      GetFirstValueAsInt(params, "verbose", &verbosity);
+    } else {
+      // if "verbosity" and "verbose" were both missing from params, don't modify LightGBM's log level
+      return;
+    }
+  }
+
+  // otherwise, update LightGBM's log level based on the passed-in value
   if (verbosity < 0) {
     LightGBM::Log::ResetLogLevel(LightGBM::LogLevel::Fatal);
   } else if (verbosity == 0) {

tests/python_package_test/test_basic.py

@@ -383,7 +383,7 @@ def test_add_features_does_not_fail_if_initial_dataset_has_zero_informative_feat
     arr_a = np.zeros((100, 1), dtype=np.float32)
     arr_b = rng.uniform(size=(100, 5))
 
-    dataset_a = lgb.Dataset(arr_a).construct()
+    dataset_a = lgb.Dataset(arr_a, params={"verbose": 0}).construct()
     expected_msg = (
         "[LightGBM] [Warning] There are no meaningful features which satisfy "
         "the provided configuration. Decreasing Dataset parameters min_data_in_bin "

tests/python_package_test/test_engine.py

@@ -1469,6 +1469,7 @@ def test_parameters_are_loaded_from_model_file(tmp_path, capsys, rng):
         "metric": ["l2", "rmse"],
         "num_leaves": 5,
         "num_threads": 1,
+        "verbosity": 0,
     }
     model_file = tmp_path / "model.txt"
     orig_bst = lgb.train(params, ds, num_boost_round=1, categorical_feature=[1, 2])
@@ -4274,11 +4275,25 @@ def test_verbosity_and_verbose(capsys):
         "verbosity": 0,
     }
     lgb.train(params, ds, num_boost_round=1)
-    expected_msg = "[LightGBM] [Warning] verbosity is set=0, verbose=1 will be ignored. " "Current value: verbosity=0"
+    expected_msg = "[LightGBM] [Warning] verbosity is set=0, verbose=1 will be ignored. Current value: verbosity=0"
     stdout = capsys.readouterr().out
     assert expected_msg in stdout
 
 
+def test_verbosity_is_respected_when_using_custom_objective(capsys):
+    X, y = make_synthetic_regression()
+    ds = lgb.Dataset(X, y)
+    params = {
+        "objective": mse_obj,
+        "nonsense": 123,
+        "num_leaves": 3,
+    }
+    lgb.train({**params, "verbosity": -1}, ds, num_boost_round=1)
+    assert capsys.readouterr().out == ""
+    lgb.train({**params, "verbosity": 0}, ds, num_boost_round=1)
+    assert "[LightGBM] [Warning] Unknown parameter: nonsense" in capsys.readouterr().out
+
+
 @pytest.mark.parametrize("verbosity_param", lgb.basic._ConfigAliases.get("verbosity"))
 @pytest.mark.parametrize("verbosity", [-1, 0])
 def test_verbosity_can_suppress_alias_warnings(capsys, verbosity_param, verbosity):

tests/python_package_test/test_sklearn.py

@@ -1290,6 +1290,19 @@ def test_max_depth_warning_is_never_raised(capsys, estimator_class, max_depth):
     assert "Provided parameters constrain tree depth" not in capsys.readouterr().out
 
 
+def test_verbosity_is_respected_when_using_custom_objective(capsys):
+    X, y = make_synthetic_regression()
+    params = {
+        "objective": objective_ls,
+        "nonsense": 123,
+        "num_leaves": 3,
+    }
+    lgb.LGBMRegressor(**params, verbosity=-1, n_estimators=1).fit(X, y)
+    assert capsys.readouterr().out == ""
+    lgb.LGBMRegressor(**params, verbosity=0, n_estimators=1).fit(X, y)
+    assert "[LightGBM] [Warning] Unknown parameter: nonsense" in capsys.readouterr().out
+
+
 @pytest.mark.parametrize("estimator_class", [lgb.LGBMModel, lgb.LGBMClassifier, lgb.LGBMRegressor, lgb.LGBMRanker])
 def test_getting_feature_names_in_np_input(estimator_class):
     # input is a numpy array, which doesn't have feature names. LightGBM adds

tests/python_package_test/test_utilities.py

@@ -31,7 +31,7 @@ def test_register_logger(tmp_path):
     eval_records = {}
     callbacks = [lgb.record_evaluation(eval_records), lgb.log_evaluation(2), lgb.early_stopping(10)]
     lgb.train(
-        {"objective": "binary", "metric": ["auc", "binary_error"]},
+        {"objective": "binary", "metric": ["auc", "binary_error"], "verbose": 1},
         lgb_train,
         num_boost_round=10,
         feval=dummy_metric,