From bc0ea4aa10b9c8789cae81fc54f1d18b23bb8432 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Tue, 18 Dec 2018 01:42:10 +0900
Subject: [PATCH 01/19] Divide metric into train_metric and valid_metric

Close #1911
---
 include/LightGBM/c_api.h         |  4 +--
 include/LightGBM/config.h        |  3 +-
 python-package/lightgbm/basic.py | 53 +++++++++++++++-----------------
 src/application/application.cpp  |  6 ++--
 src/c_api.cpp                    | 50 +++++++++++++++++++++---------
 src/io/config.cpp                | 50 +++++++++++++++++++++++++++---
 src/io/config_auto.cpp           |  4 +--
 src/lightgbm_R.cpp               |  9 +++---
 8 files changed, 119 insertions(+), 60 deletions(-)

diff --git a/include/LightGBM/c_api.h b/include/LightGBM/c_api.h
index 4d50788efea0..3d4825078897 100644
--- a/include/LightGBM/c_api.h
+++ b/include/LightGBM/c_api.h
@@ -484,7 +484,7 @@ LIGHTGBM_C_EXPORT int LGBM_BoosterNumberOfTotalModel(BoosterHandle handle, int*
 * \param out_len total number of eval results
 * \return 0 when succeed, -1 when failure happens
 */
-LIGHTGBM_C_EXPORT int LGBM_BoosterGetEvalCounts(BoosterHandle handle, int* out_len);
+LIGHTGBM_C_EXPORT int LGBM_BoosterGetEvalCounts(BoosterHandle handle, int data_idx, int* out_len);
 
 /*!
 * \brief Get name of eval
@@ -492,7 +492,7 @@ LIGHTGBM_C_EXPORT int LGBM_BoosterGetEvalCounts(BoosterHandle handle, int* out_l
 * \param out_strs names of eval result, need to pre-allocate memory before call this
 * \return 0 when succeed, -1 when failure happens
 */
-LIGHTGBM_C_EXPORT int LGBM_BoosterGetEvalNames(BoosterHandle handle, int* out_len, char** out_strs);
+LIGHTGBM_C_EXPORT int LGBM_BoosterGetEvalNames(BoosterHandle handle, int data_idx, int* out_len, char** out_strs);
 
 /*!
 * \brief Get name of features
diff --git a/include/LightGBM/config.h b/include/LightGBM/config.h
index 7f29b0595395..b73a691827fb 100644
--- a/include/LightGBM/config.h
+++ b/include/LightGBM/config.h
@@ -697,7 +697,8 @@ struct Config {
   // descl2 = ``xentlambda``, "intensity-weighted" cross-entropy, aliases: ``cross_entropy_lambda``
   // descl2 = ``kldiv``, `Kullback-Leibler divergence <https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence>`__, aliases: ``kullback_leibler``
   // desc = support multiple metrics, separated by ``,``
-  std::vector<std::string> metric;
+  std::vector<std::string> train_metric;
+  std::vector<std::string> valid_metric;
 
   // check = >0
   // alias = output_freq
diff --git a/python-package/lightgbm/basic.py b/python-package/lightgbm/basic.py
index b347f73ce41a..d6fd7edb172c 100644
--- a/python-package/lightgbm/basic.py
+++ b/python-package/lightgbm/basic.py
@@ -1472,7 +1472,6 @@ def __init__(self, params=None, train_set=None, model_file=None, silent=False):
         """
         self.handle = None
         self.network = False
-        self.__need_reload_eval_info = True
         self.__train_data_name = "training"
         self.__attr = {}
         self.__set_objective_to_none = False
@@ -1528,7 +1527,7 @@ def __init__(self, params=None, train_set=None, model_file=None, silent=False):
             # buffer for inner predict
             self.__inner_predict_buffer = [None]
             self.__is_predicted_cur_iter = [False]
-            self.__get_eval_info()
+            self.__get_eval_info(0)  # init by train data (data_idx=0)
             self.pandas_categorical = train_set.pandas_categorical
         elif model_file is not None:
             # Prediction task
@@ -1708,8 +1707,6 @@ def reset_parameter(self, params):
         self : Booster
             Booster with new parameters.
         """
-        if any(metric_alias in params for metric_alias in ('metric', 'metrics', 'metric_types')):
-            self.__need_reload_eval_info = True
         params_str = param_dict_to_str(params)
         if params_str:
             _safe_call(_LIB.LGBM_BoosterResetParameter(
@@ -2310,7 +2307,7 @@ def __inner_eval(self, data_name, data_idx, feval=None):
         """Evaluate training or validation data."""
         if data_idx >= self.__num_dataset:
             raise ValueError("Data_idx should be smaller than number of dataset")
-        self.__get_eval_info()
+        self.__get_eval_info(data_idx)
         ret = []
         if self.__num_inner_eval > 0:
             result = np.zeros(self.__num_inner_eval, dtype=np.float64)
@@ -2363,31 +2360,31 @@ def __inner_predict(self, data_idx):
             self.__is_predicted_cur_iter[data_idx] = True
         return self.__inner_predict_buffer[data_idx]
 
-    def __get_eval_info(self):
+    def __get_eval_info(self, data_idx):
         """Get inner evaluation count and names."""
-        if self.__need_reload_eval_info:
-            self.__need_reload_eval_info = False
-            out_num_eval = ctypes.c_int(0)
-            # Get num of inner evals
-            _safe_call(_LIB.LGBM_BoosterGetEvalCounts(
+        out_num_eval = ctypes.c_int(0)
+        # Get num of inner evals
+        _safe_call(_LIB.LGBM_BoosterGetEvalCounts(
+            self.handle,
+            ctypes.c_int(data_idx),
+            ctypes.byref(out_num_eval)))
+        self.__num_inner_eval = out_num_eval.value
+        if self.__num_inner_eval > 0:
+            # Get name of evals
+            tmp_out_len = ctypes.c_int(0)
+            string_buffers = [ctypes.create_string_buffer(255) for i in range_(self.__num_inner_eval)]
+            ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))
+            _safe_call(_LIB.LGBM_BoosterGetEvalNames(
                 self.handle,
-                ctypes.byref(out_num_eval)))
-            self.__num_inner_eval = out_num_eval.value
-            if self.__num_inner_eval > 0:
-                # Get name of evals
-                tmp_out_len = ctypes.c_int(0)
-                string_buffers = [ctypes.create_string_buffer(255) for i in range_(self.__num_inner_eval)]
-                ptr_string_buffers = (ctypes.c_char_p * self.__num_inner_eval)(*map(ctypes.addressof, string_buffers))
-                _safe_call(_LIB.LGBM_BoosterGetEvalNames(
-                    self.handle,
-                    ctypes.byref(tmp_out_len),
-                    ptr_string_buffers))
-                if self.__num_inner_eval != tmp_out_len.value:
-                    raise ValueError("Length of eval names doesn't equal with num_evals")
-                self.__name_inner_eval = \
-                    [string_buffers[i].value.decode() for i in range_(self.__num_inner_eval)]
-                self.__higher_better_inner_eval = \
-                    [name.startswith(('auc', 'ndcg@', 'map@')) for name in self.__name_inner_eval]
+                ctypes.c_int(data_idx),
+                ctypes.byref(tmp_out_len),
+                ptr_string_buffers))
+            if self.__num_inner_eval != tmp_out_len.value:
+                raise ValueError("Length of eval names doesn't equal with num_evals")
+            self.__name_inner_eval = \
+                [string_buffers[i].value.decode() for i in range_(self.__num_inner_eval)]
+            self.__higher_better_inner_eval = \
+                [name.startswith(('auc', 'ndcg@', 'map@')) for name in self.__name_inner_eval]
 
     def attr(self, key):
         """Get attribute string from the Booster.
diff --git a/src/application/application.cpp b/src/application/application.cpp
index 3fbc7d3a0d91..15363ee70813 100644
--- a/src/application/application.cpp
+++ b/src/application/application.cpp
@@ -116,7 +116,7 @@ void Application::LoadData() {
   }
   // create training metric
   if (config_.is_provide_training_metric) {
-    for (auto metric_type : config_.metric) {
+    for (auto metric_type : config_.train_metric) {
       auto metric = std::unique_ptr<Metric>(Metric::CreateMetric(metric_type, config_));
       if (metric == nullptr) { continue; }
       metric->Init(train_data_->metadata(), train_data_->num_data());
@@ -126,7 +126,7 @@ void Application::LoadData() {
   train_metric_.shrink_to_fit();
 
 
-  if (!config_.metric.empty()) {
+  if (!config_.train_metric.empty()) {
     // only when have metrics then need to construct validation data
 
     // Add validation data, if it exists
@@ -146,7 +146,7 @@ void Application::LoadData() {
 
       // add metric for validation data
       valid_metrics_.emplace_back();
-      for (auto metric_type : config_.metric) {
+      for (auto metric_type : config_.valid_metric) {
         auto metric = std::unique_ptr<Metric>(Metric::CreateMetric(metric_type, config_));
         if (metric == nullptr) { continue; }
         metric->Init(valid_datas_.back()->metadata(),
diff --git a/src/c_api.cpp b/src/c_api.cpp
index 3a8fbe86079e..4ddc510aed98 100644
--- a/src/c_api.cpp
+++ b/src/c_api.cpp
@@ -103,7 +103,7 @@ class Booster {
 
     // create training metric
     train_metric_.clear();
-    for (auto metric_type : config_.metric) {
+    for (auto metric_type : config_.train_metric) {
       auto metric = std::unique_ptr<Metric>(
         Metric::CreateMetric(metric_type, config_));
       if (metric == nullptr) { continue; }
@@ -164,7 +164,7 @@ class Booster {
   void AddValidData(const Dataset* valid_data) {
     std::lock_guard<std::mutex> lock(mutex_);
     valid_metrics_.emplace_back();
-    for (auto metric_type : config_.metric) {
+    for (auto metric_type : config_.valid_metric) {
       auto metric = std::unique_ptr<Metric>(Metric::CreateMetric(metric_type, config_));
       if (metric == nullptr) { continue; }
       metric->Init(valid_data->metadata(), valid_data->num_data());
@@ -297,21 +297,41 @@ class Booster {
     boosting_->ShuffleModels(start_iter, end_iter);
   }
 
-  int GetEvalCounts() const {
+  int GetEvalCounts(int data_idx) const {
     int ret = 0;
-    for (const auto& metric : train_metric_) {
-      ret += static_cast<int>(metric->GetName().size());
+    if (data_idx == 0) {
+        for (const auto& metric : train_metric_) {
+          ret += static_cast<int>(metric->GetName().size());
+        }
+    } else {
+        for (size_t i = 0; i < valid_metrics_.size(); ++i) {
+          for (size_t j = 0; j < valid_metrics_[i].size(); ++j) {
+            ret += static_cast<int>(valid_metrics_[i][j]->GetName().size());
+          }
+        }
     }
     return ret;
   }
 
-  int GetEvalNames(char** out_strs) const {
+  int GetEvalNames(int data_idx, char** out_strs) const {
     int idx = 0;
-    for (const auto& metric : train_metric_) {
-      for (const auto& name : metric->GetName()) {
-        std::memcpy(out_strs[idx], name.c_str(), name.size() + 1);
-        ++idx;
-      }
+    if (data_idx == 0) {
+        for (const auto& metric : train_metric_) {
+          for (const auto& name : metric->GetName()) {
+            std::memcpy(out_strs[idx], name.c_str(), name.size() + 1);
+            ++idx;
+          }
+        }
+    } else {
+        for (size_t i = 0; i < valid_metrics_.size(); ++i) {
+          for (size_t j = 0; j < valid_metrics_[i].size(); ++j) {
+              auto& metric = valid_metrics_[i][j];
+              for (const auto& name : metric->GetName()) {
+                std::memcpy(out_strs[idx], name.c_str(), name.size() + 1);
+                ++idx;
+              }
+          }
+        }
     }
     return idx;
   }
@@ -1029,17 +1049,17 @@ int LGBM_BoosterNumberOfTotalModel(BoosterHandle handle, int* out_models) {
   API_END();
 }
 
-int LGBM_BoosterGetEvalCounts(BoosterHandle handle, int* out_len) {
+int LGBM_BoosterGetEvalCounts(BoosterHandle handle, int data_idx, int* out_len) {
   API_BEGIN();
   Booster* ref_booster = reinterpret_cast<Booster*>(handle);
-  *out_len = ref_booster->GetEvalCounts();
+  *out_len = ref_booster->GetEvalCounts(data_idx);
   API_END();
 }
 
-int LGBM_BoosterGetEvalNames(BoosterHandle handle, int* out_len, char** out_strs) {
+int LGBM_BoosterGetEvalNames(BoosterHandle handle, int data_idx, int* out_len, char** out_strs) {
   API_BEGIN();
   Booster* ref_booster = reinterpret_cast<Booster*>(handle);
-  *out_len = ref_booster->GetEvalNames(out_strs);
+  *out_len = ref_booster->GetEvalNames(data_idx, out_strs);
   API_END();
 }
 
diff --git a/src/io/config.cpp b/src/io/config.cpp
index 414ecafcfd79..f67a4d49a854 100644
--- a/src/io/config.cpp
+++ b/src/io/config.cpp
@@ -68,9 +68,9 @@ void GetObjectiveType(const std::unordered_map<std::string, std::string>& params
   }
 }
 
-void GetMetricType(const std::unordered_map<std::string, std::string>& params, std::vector<std::string>* metric) {
+void GetTrainMetricType(const std::unordered_map<std::string, std::string>& params, std::vector<std::string>* metric) {
   std::string value;
-  if (Config::GetString(params, "metric", &value)) {
+  if (Config::GetString(params, "train_metric", &value)) {
     // clear old metrics
     metric->clear();
     // to lower
@@ -99,6 +99,31 @@ void GetMetricType(const std::unordered_map<std::string, std::string>& params, s
   }
 }
 
+void GetValidMetricType(const std::unordered_map<std::string, std::string>& params, std::vector<std::string>* metric) {
+  std::string value;
+  if (Config::GetString(params, "valid_metric", &value)) {
+    // clear old metrics
+    metric->clear();
+    // to lower
+    std::transform(value.begin(), value.end(), value.begin(), Common::tolower);
+    // split
+    std::vector<std::string> metrics = Common::Split(value.c_str(), ',');
+    // remove duplicate
+    std::unordered_set<std::string> metric_sets;
+    for (auto& met : metrics) {
+      std::transform(met.begin(), met.end(), met.begin(), Common::tolower);
+      if (metric_sets.count(met) <= 0) {
+        metric_sets.insert(met);
+      }
+    }
+    for (auto& met : metric_sets) {
+      metric->push_back(met);
+    }
+    metric->shrink_to_fit();
+  }
+}
+
+
 void GetTaskType(const std::unordered_map<std::string, std::string>& params, TaskType* task) {
   std::string value;
   if (Config::GetString(params, "task", &value)) {
@@ -164,7 +189,8 @@ void Config::Set(const std::unordered_map<std::string, std::string>& params) {
 
   GetTaskType(params, &task);
   GetBoostingType(params, &boosting);
-  GetMetricType(params, &metric);
+  GetTrainMetricType(params, &train_metric);
+  GetValidMetricType(params, &valid_metric);
   GetObjectiveType(params, &objective);
   GetDeviceType(params, &device_type);
   GetTreeLearnerType(params, &tree_learner);
@@ -215,7 +241,7 @@ void Config::CheckParamConflict() {
       Log::Fatal("Number of classes must be 1 for non-multiclass training");
     }
   }
-  for (std::string metric_type : metric) {
+  for (std::string metric_type : train_metric) {
     bool metric_custom_or_none = metric_type == std::string("none") || metric_type == std::string("null") 
                                  || metric_type == std::string("custom") || metric_type == std::string("na");
     bool metric_type_multiclass = (CheckMultiClassObjective(metric_type)
@@ -227,6 +253,19 @@ void Config::CheckParamConflict() {
       Log::Fatal("Multiclass objective and metrics don't match");
     }
   }
+  for (std::string metric_type : valid_metric) {
+    bool metric_custom_or_none = metric_type == std::string("none") || metric_type == std::string("null") 
+                                 || metric_type == std::string("custom") || metric_type == std::string("na");
+    bool metric_type_multiclass = (CheckMultiClassObjective(metric_type)
+                                   || metric_type == std::string("multi_logloss")
+                                   || metric_type == std::string("multi_error")
+                                   || (metric_custom_or_none && num_class_check > 1));
+    if ((objective_type_multiclass && !metric_type_multiclass)
+        || (!objective_type_multiclass && metric_type_multiclass)) {
+      Log::Fatal("Multiclass objective and metrics don't match");
+    }
+  }
+
 
   if (num_machines > 1) {
     is_parallel = true;
@@ -270,7 +309,8 @@ std::string Config::ToString() const {
   std::stringstream str_buf;
   str_buf << "[boosting: " << boosting << "]\n";
   str_buf << "[objective: " << objective << "]\n";
-  str_buf << "[metric: " << Common::Join(metric, ",") << "]\n";
+  str_buf << "[train_metric: " << Common::Join(train_metric, ",") << "]\n";
+  str_buf << "[valid_metric: " << Common::Join(valid_metric, ",") << "]\n";
   str_buf << "[tree_learner: " << tree_learner << "]\n";
   str_buf << "[device_type: " << device_type << "]\n";
   str_buf << SaveMembersToString();
diff --git a/src/io/config_auto.cpp b/src/io/config_auto.cpp
index 3803d986da5e..809eed4e1b39 100644
--- a/src/io/config_auto.cpp
+++ b/src/io/config_auto.cpp
@@ -140,7 +140,6 @@ std::unordered_map<std::string, std::string> Config::alias_table({
   {"output_freq", "metric_freq"},
   {"training_metric", "is_provide_training_metric"},
   {"is_training_metric", "is_provide_training_metric"},
-  {"train_metric", "is_provide_training_metric"},
   {"ndcg_eval_at", "eval_at"},
   {"ndcg_at", "eval_at"},
   {"map_eval_at", "eval_at"},
@@ -249,7 +248,8 @@ std::unordered_set<std::string> Config::parameter_set({
   "tweedie_variance_power",
   "max_position",
   "label_gain",
-  "metric",
+  "train_metric",
+  "valid_metric",
   "metric_freq",
   "is_provide_training_metric",
   "eval_at",
diff --git a/src/lightgbm_R.cpp b/src/lightgbm_R.cpp
index b140cae0ff04..6a715bcfe0d4 100644
--- a/src/lightgbm_R.cpp
+++ b/src/lightgbm_R.cpp
@@ -426,11 +426,12 @@ LGBM_SE LGBM_BoosterGetEvalNames_R(LGBM_SE handle,
   LGBM_SE buf_len,
   LGBM_SE actual_len,
   LGBM_SE eval_names,
-  LGBM_SE call_state) {
+  LGBM_SE call_state,
+  LGBM_SE data_idx) {
 
   R_API_BEGIN();
   int len;
-  CHECK_CALL(LGBM_BoosterGetEvalCounts(R_GET_PTR(handle), &len));
+  CHECK_CALL(LGBM_BoosterGetEvalCounts(R_GET_PTR(handle), R_AS_INT(data_idx), &len));
   std::vector<std::vector<char>> names(len);
   std::vector<char*> ptr_names(len);
   for (int i = 0; i < len; ++i) {
@@ -438,7 +439,7 @@ LGBM_SE LGBM_BoosterGetEvalNames_R(LGBM_SE handle,
     ptr_names[i] = names[i].data();
   }
   int out_len;
-  CHECK_CALL(LGBM_BoosterGetEvalNames(R_GET_PTR(handle), &out_len, ptr_names.data()));
+  CHECK_CALL(LGBM_BoosterGetEvalNames(R_GET_PTR(handle), R_AS_INT(data_idx), &out_len, ptr_names.data()));
   CHECK(out_len == len);
   auto merge_names = Common::Join<char*>(ptr_names, "\t");
   EncodeChar(eval_names, merge_names.c_str(), buf_len, actual_len, merge_names.size() + 1);
@@ -451,7 +452,7 @@ LGBM_SE LGBM_BoosterGetEval_R(LGBM_SE handle,
   LGBM_SE call_state) {
   R_API_BEGIN();
   int len;
-  CHECK_CALL(LGBM_BoosterGetEvalCounts(R_GET_PTR(handle), &len));
+  CHECK_CALL(LGBM_BoosterGetEvalCounts(R_GET_PTR(handle), R_AS_INT(data_idx), &len));
   double* ptr_ret = R_REAL_PTR(out_result);
   int out_len;
   CHECK_CALL(LGBM_BoosterGetEval(R_GET_PTR(handle), R_AS_INT(data_idx), &out_len, ptr_ret));

From 7ce4d594def4474c6b3a4f1de62152aa5aa73589 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Tue, 18 Dec 2018 17:24:24 +0900
Subject: [PATCH 02/19] fix sklearn API

---
 python-package/lightgbm/sklearn.py         | 3 ++-
 tests/python_package_test/test_plotting.py | 5 ++++-
 2 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/python-package/lightgbm/sklearn.py b/python-package/lightgbm/sklearn.py
index 821feebcbc28..5db0bdb82b54 100644
--- a/python-package/lightgbm/sklearn.py
+++ b/python-package/lightgbm/sklearn.py
@@ -477,7 +477,7 @@ def fit(self, X, y,
             # concatenate metric from params (or default if not provided in params) and eval_metric
             original_metric = [original_metric] if isinstance(original_metric, (string_type, type(None))) else original_metric
             eval_metric = [eval_metric] if isinstance(eval_metric, (string_type, type(None))) else eval_metric
-            params['metric'] = set(original_metric + eval_metric)
+            params['valid_metric'] = set(original_metric + eval_metric)  # TODO: train_metric
 
         if not isinstance(X, DataFrame):
             X, y = _LGBMCheckXY(X, y, accept_sparse=True, force_all_finite=False, ensure_min_samples=2)
@@ -518,6 +518,7 @@ def _get_meta_data(collection, i):
                 # reduce cost for prediction training data
                 if valid_data[0] is X and valid_data[1] is y:
                     valid_set = train_set
+                    # params['train_metric'] = set(original_metric + eval_metric)
                 else:
                     valid_weight = _get_meta_data(eval_sample_weight, i)
                     if _get_meta_data(eval_class_weight, i) is not None:
diff --git a/tests/python_package_test/test_plotting.py b/tests/python_package_test/test_plotting.py
index 05a25e64d4a2..6ee332355ae6 100644
--- a/tests/python_package_test/test_plotting.py
+++ b/tests/python_package_test/test_plotting.py
@@ -104,7 +104,10 @@ def test_create_tree_digraph(self):
     @unittest.skipIf(not MATPLOTLIB_INSTALLED, 'matplotlib is not installed')
     def test_plot_metrics(self):
         test_data = lgb.Dataset(self.X_test, self.y_test, reference=self.train_data)
-        self.params.update({"metric": {"binary_logloss", "binary_error"}})
+        self.params.update({
+            "train_metric": ["binary_logloss", "binary_error"],
+            "valid_metric": ["binary_logloss", "binary_error"]
+        })
 
         evals_result0 = {}
         gbm0 = lgb.train(self.params, self.train_data,

From e1f2bba50bb1737964e3235737d8a0595e49287f Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Tue, 18 Dec 2018 18:09:43 +0900
Subject: [PATCH 03/19] fix tests

---
 python-package/lightgbm/engine.py        |  2 +-
 src/io/config.cpp                        |  7 ++++++
 tests/python_package_test/test_engine.py | 32 ++++++++++++------------
 3 files changed, 24 insertions(+), 17 deletions(-)

diff --git a/python-package/lightgbm/engine.py b/python-package/lightgbm/engine.py
index da7459fced71..da4e42069a29 100644
--- a/python-package/lightgbm/engine.py
+++ b/python-package/lightgbm/engine.py
@@ -446,7 +446,7 @@ def cv(params, train_set, num_boost_round=100,
              .set_categorical_feature(categorical_feature)
 
     if metrics is not None:
-        params['metric'] = metrics
+        params['valid_metric'] = metrics
 
     results = collections.defaultdict(list)
     cvfolds = _make_n_folds(train_set, folds=folds, nfold=nfold,
diff --git a/src/io/config.cpp b/src/io/config.cpp
index f67a4d49a854..4ee8f1b0a54f 100644
--- a/src/io/config.cpp
+++ b/src/io/config.cpp
@@ -121,6 +121,13 @@ void GetValidMetricType(const std::unordered_map<std::string, std::string>& para
     }
     metric->shrink_to_fit();
   }
+  // add names of objective function if not providing metric
+  if (metric->empty() && value.size() == 0) {
+    if (Config::GetString(params, "objective", &value)) {
+      std::transform(value.begin(), value.end(), value.begin(), Common::tolower);
+      metric->push_back(value);
+    }
+  }
 }
 
 
diff --git a/tests/python_package_test/test_engine.py b/tests/python_package_test/test_engine.py
index e89b59661f9f..fc73169c5b66 100644
--- a/tests/python_package_test/test_engine.py
+++ b/tests/python_package_test/test_engine.py
@@ -30,7 +30,7 @@ def test_binary(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'binary',
-            'metric': 'binary_logloss',
+            'valid_metric': 'binary_logloss',
             'verbose': -1,
             'num_iteration': 50  # test num_iteration in dict here
         }
@@ -57,7 +57,7 @@ def test_rf(self):
             'bagging_fraction': 0.5,
             'feature_fraction': 0.5,
             'num_leaves': 50,
-            'metric': 'binary_logloss',
+            'valid_metric': 'binary_logloss',
             'verbose': -1
         }
         lgb_train = lgb.Dataset(X_train, y_train)
@@ -76,7 +76,7 @@ def test_regression(self):
         X, y = load_boston(True)
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
-            'metric': 'l2',
+            'valid_metric': 'l2',
             'verbose': -1
         }
         lgb_train = lgb.Dataset(X_train, y_train)
@@ -102,7 +102,7 @@ def test_missing_value_handle(self):
         lgb_eval = lgb.Dataset(X_train, y_train)
 
         params = {
-            'metric': 'l2',
+            'valid_metric': 'l2',
             'verbose': -1,
             'boost_from_average': False
         }
@@ -127,7 +127,7 @@ def test_missing_value_handle_na(self):
 
         params = {
             'objective': 'regression',
-            'metric': 'auc',
+            'valid_metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -159,7 +159,7 @@ def test_missing_value_handle_zero(self):
 
         params = {
             'objective': 'regression',
-            'metric': 'auc',
+            'valid_metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -191,7 +191,7 @@ def test_missing_value_handle_none(self):
 
         params = {
             'objective': 'regression',
-            'metric': 'auc',
+            'valid_metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -224,7 +224,7 @@ def test_categorical_handle(self):
 
         params = {
             'objective': 'regression',
-            'metric': 'auc',
+            'valid_metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -261,7 +261,7 @@ def test_categorical_handle_na(self):
 
         params = {
             'objective': 'regression',
-            'metric': 'auc',
+            'valid_metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -292,7 +292,7 @@ def test_multiclass(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'multiclass',
-            'metric': 'multi_logloss',
+            'valid_metric': 'multi_logloss',
             'num_class': 10,
             'verbose': -1
         }
@@ -314,7 +314,7 @@ def test_multiclass_rf(self):
         params = {
             'boosting_type': 'rf',
             'objective': 'multiclass',
-            'metric': 'multi_logloss',
+            'valid_metric': 'multi_logloss',
             'bagging_freq': 1,
             'bagging_fraction': 0.6,
             'feature_fraction': 0.6,
@@ -365,7 +365,7 @@ def test_early_stopping(self):
         X, y = load_breast_cancer(True)
         params = {
             'objective': 'binary',
-            'metric': 'binary_logloss',
+            'valid_metric': 'binary_logloss',
             'verbose': -1
         }
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
@@ -397,7 +397,7 @@ def test_continue_train(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'regression',
-            'metric': 'l1',
+            'valid_metric': 'l1',
             'verbose': -1
         }
         lgb_train = lgb.Dataset(X_train, y_train, free_raw_data=False)
@@ -426,7 +426,7 @@ def test_continue_train_multiclass(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'multiclass',
-            'metric': 'multi_logloss',
+            'valid_metric': 'multi_logloss',
             'num_class': 3,
             'verbose': -1
         }
@@ -450,7 +450,7 @@ def test_cv(self):
         params = {'verbose': -1}
         lgb_train = lgb.Dataset(X_train, y_train)
         # shuffle = False, override metric in params
-        params_with_metric = {'metric': 'l2', 'verbose': -1}
+        params_with_metric = {'valid_metric': 'l2', 'verbose': -1}
         cv_res = lgb.cv(params_with_metric, lgb_train, num_boost_round=10,
                         nfold=3, stratified=False, shuffle=False,
                         metrics='l1', verbose_eval=False)
@@ -595,7 +595,7 @@ def test_reference_chain(self):
         # take subsets and train
         tmp_dat_train = tmp_dat.subset(np.arange(80))
         tmp_dat_val = tmp_dat.subset(np.arange(80, 100)).subset(np.arange(18))
-        params = {'objective': 'regression_l2', 'metric': 'rmse'}
+        params = {'objective': 'regression_l2', 'train_metric': 'rmse', 'valid_metric': 'rmse'}
         evals_result = {}
         gbm = lgb.train(params, tmp_dat_train, num_boost_round=20,
                         valid_sets=[tmp_dat_train, tmp_dat_val], evals_result=evals_result)

From ab23d73b22e4eaa8dc64ae5ec7e08b7b18860c2a Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Tue, 18 Dec 2018 18:37:26 +0900
Subject: [PATCH 04/19] fix examples

---
 examples/python-guide/advanced_example.py | 2 +-
 examples/python-guide/plot_example.py     | 3 ++-
 examples/python-guide/simple_example.py   | 2 +-
 3 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/examples/python-guide/advanced_example.py b/examples/python-guide/advanced_example.py
index cef8ad6c944a..865b1307249d 100644
--- a/examples/python-guide/advanced_example.py
+++ b/examples/python-guide/advanced_example.py
@@ -36,7 +36,7 @@
 params = {
     'boosting_type': 'gbdt',
     'objective': 'binary',
-    'metric': 'binary_logloss',
+    'train_metric': 'binary_logloss',
     'num_leaves': 31,
     'learning_rate': 0.05,
     'feature_fraction': 0.9,
diff --git a/examples/python-guide/plot_example.py b/examples/python-guide/plot_example.py
index eca896982fa9..e935930aa95b 100644
--- a/examples/python-guide/plot_example.py
+++ b/examples/python-guide/plot_example.py
@@ -25,7 +25,8 @@
 # specify your configurations as a dict
 params = {
     'num_leaves': 5,
-    'metric': ('l1', 'l2'),
+    'train_metric': ('l1', 'l2'),
+    'valid_metric': ('l1', 'l2'),
     'verbose': 0
 }
 
diff --git a/examples/python-guide/simple_example.py b/examples/python-guide/simple_example.py
index 18f07cd601cd..403956d8642a 100644
--- a/examples/python-guide/simple_example.py
+++ b/examples/python-guide/simple_example.py
@@ -22,7 +22,7 @@
 params = {
     'boosting_type': 'gbdt',
     'objective': 'regression',
-    'metric': {'l2', 'l1'},
+    'valid_metric': {'l2', 'l1'},
     'num_leaves': 31,
     'learning_rate': 0.05,
     'feature_fraction': 0.9,

From 9327ffb1a62dcddc343d771608c294f63aa37ca5 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Tue, 18 Dec 2018 18:54:10 +0900
Subject: [PATCH 05/19] fix notebook

---
 .../notebooks/interactive_plot_example.ipynb             | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/examples/python-guide/notebooks/interactive_plot_example.ipynb b/examples/python-guide/notebooks/interactive_plot_example.ipynb
index e472404d9aa4..f86377a52950 100644
--- a/examples/python-guide/notebooks/interactive_plot_example.ipynb
+++ b/examples/python-guide/notebooks/interactive_plot_example.ipynb
@@ -103,7 +103,8 @@
    "source": [
     "params = {\n",
     "    'num_leaves': 5,\n",
-    "    'metric': ['l1', 'l2'],\n",
+    "    'train_metric': ['l1', 'l2'],\n",
+    "    'valid_metric': ['l1', 'l2'],\n",
     "    'verbose': -1\n",
     "}"
    ]
@@ -197,9 +198,9 @@
    "source": [
     "if INTERACTIVE:\n",
     "    # create widget to switch between metrics\n",
-    "    interact(render_metric, metric_name=params['metric'])\n",
+    "    interact(render_metric, metric_name=params['valid_metric'])\n",
     "else:\n",
-    "    render_metric(params['metric'][0])"
+    "    render_metric(params['valid_metric'][0])"
    ]
   },
   {
@@ -336,7 +337,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.3"
+   "version": "3.6.6"
   },
   "varInspector": {
    "cols": {

From b0e7a5ca6fb1097a8248e0de0794b9cabdc006c2 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Tue, 18 Dec 2018 19:01:36 +0900
Subject: [PATCH 06/19] fix docs

---
 include/LightGBM/config.h | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/include/LightGBM/config.h b/include/LightGBM/config.h
index b73a691827fb..a4357853cf1d 100644
--- a/include/LightGBM/config.h
+++ b/include/LightGBM/config.h
@@ -698,6 +698,10 @@ struct Config {
   // descl2 = ``kldiv``, `Kullback-Leibler divergence <https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence>`__, aliases: ``kullback_leibler``
   // desc = support multiple metrics, separated by ``,``
   std::vector<std::string> train_metric;
+
+  // type = multi-enum
+  // default = ""
+  // desc = metric(s) to be evaluated on the evaluation set(s)
   std::vector<std::string> valid_metric;
 
   // check = >0
@@ -705,7 +709,7 @@ struct Config {
   // desc = frequency for metric output
   int metric_freq = 1;
 
-  // alias = training_metric, is_training_metric, train_metric
+  // alias = training_metric, is_training_metric
   // desc = set this to ``true`` to output metric result over training dataset
   // desc = **Note**: can be used only in CLI version
   bool is_provide_training_metric = false;

From 4a98ec0f7788fef7243e34239f7fe304607b9ed0 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Wed, 19 Dec 2018 00:04:58 +0900
Subject: [PATCH 07/19] update Parameters.rst and config_auto.cpp

`python helpers/parameter_generator.py`
---
 docs/Parameters.rst    | 8 ++++++--
 src/io/config_auto.cpp | 9 +++++++--
 2 files changed, 13 insertions(+), 4 deletions(-)

diff --git a/docs/Parameters.rst b/docs/Parameters.rst
index 6cac09e0588a..acda6a9291a2 100644
--- a/docs/Parameters.rst
+++ b/docs/Parameters.rst
@@ -739,7 +739,7 @@ Objective Parameters
 Metric Parameters
 -----------------
 
--  ``metric`` :raw-html:`<a id="metric" title="Permalink to this parameter" href="#metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum, aliases: ``metrics``, ``metric_types``
+-  ``train_metric`` :raw-html:`<a id="train_metric" title="Permalink to this parameter" href="#train_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum, aliases: ``metrics``, ``metric_types``
 
    -  metric(s) to be evaluated on the evaluation set(s)
 
@@ -791,11 +791,15 @@ Metric Parameters
 
    -  support multiple metrics, separated by ``,``
 
+-  ``valid_metric`` :raw-html:`<a id="valid_metric" title="Permalink to this parameter" href="#valid_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
+
+   -  metric(s) to be evaluated on the evaluation set(s)
+
 -  ``metric_freq`` :raw-html:`<a id="metric_freq" title="Permalink to this parameter" href="#metric_freq">&#x1F517;&#xFE0E;</a>`, default = ``1``, type = int, aliases: ``output_freq``, constraints: ``metric_freq > 0``
 
    -  frequency for metric output
 
--  ``is_provide_training_metric`` :raw-html:`<a id="is_provide_training_metric" title="Permalink to this parameter" href="#is_provide_training_metric">&#x1F517;&#xFE0E;</a>`, default = ``false``, type = bool, aliases: ``training_metric``, ``is_training_metric``, ``train_metric``
+-  ``is_provide_training_metric`` :raw-html:`<a id="is_provide_training_metric" title="Permalink to this parameter" href="#is_provide_training_metric">&#x1F517;&#xFE0E;</a>`, default = ``false``, type = bool, aliases: ``training_metric``, ``is_training_metric``
 
    -  set this to ``true`` to output metric result over training dataset
 
diff --git a/src/io/config_auto.cpp b/src/io/config_auto.cpp
index 809eed4e1b39..3964c3e7af9c 100644
--- a/src/io/config_auto.cpp
+++ b/src/io/config_auto.cpp
@@ -135,8 +135,8 @@ std::unordered_map<std::string, std::string> Config::alias_table({
   {"num_classes", "num_class"},
   {"unbalance", "is_unbalance"},
   {"unbalanced_sets", "is_unbalance"},
-  {"metrics", "metric"},
-  {"metric_types", "metric"},
+  {"metrics", "train_metric"},
+  {"metric_types", "train_metric"},
   {"output_freq", "metric_freq"},
   {"training_metric", "is_provide_training_metric"},
   {"is_training_metric", "is_provide_training_metric"},
@@ -491,6 +491,10 @@ void Config::GetMembersFromString(const std::unordered_map<std::string, std::str
     label_gain = Common::StringToArray<double>(tmp_str, ',');
   }
 
+  if (GetString(params, "valid_metric", &tmp_str)) {
+    valid_metric = Common::Split(tmp_str.c_str(), ',');
+  }
+
   GetInt(params, "metric_freq", &metric_freq);
   CHECK(metric_freq >0);
 
@@ -609,6 +613,7 @@ std::string Config::SaveMembersToString() const {
   str_buf << "[tweedie_variance_power: " << tweedie_variance_power << "]\n";
   str_buf << "[max_position: " << max_position << "]\n";
   str_buf << "[label_gain: " << Common::Join(label_gain,",") << "]\n";
+  str_buf << "[valid_metric: " << Common::Join(valid_metric,",") << "]\n";
   str_buf << "[metric_freq: " << metric_freq << "]\n";
   str_buf << "[is_provide_training_metric: " << is_provide_training_metric << "]\n";
   str_buf << "[eval_at: " << Common::Join(eval_at,",") << "]\n";

From a276abbfc683ed33e118cb3cd7b7a5374d522d9a Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Wed, 19 Dec 2018 00:29:00 +0900
Subject: [PATCH 08/19] fix

---
 docs/Parameters.rst       | 2 +-
 include/LightGBM/config.h | 2 +-
 src/io/config_auto.cpp    | 7 -------
 3 files changed, 2 insertions(+), 9 deletions(-)

diff --git a/docs/Parameters.rst b/docs/Parameters.rst
index acda6a9291a2..baac3ce806ea 100644
--- a/docs/Parameters.rst
+++ b/docs/Parameters.rst
@@ -739,7 +739,7 @@ Objective Parameters
 Metric Parameters
 -----------------
 
--  ``train_metric`` :raw-html:`<a id="train_metric" title="Permalink to this parameter" href="#train_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum, aliases: ``metrics``, ``metric_types``
+-  ``train_metric`` :raw-html:`<a id="train_metric" title="Permalink to this parameter" href="#train_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
 
    -  metric(s) to be evaluated on the evaluation set(s)
 
diff --git a/include/LightGBM/config.h b/include/LightGBM/config.h
index a4357853cf1d..0702046529d0 100644
--- a/include/LightGBM/config.h
+++ b/include/LightGBM/config.h
@@ -669,7 +669,6 @@ struct Config {
   #pragma region Metric Parameters
 
   // [doc-only]
-  // alias = metrics, metric_types
   // default = ""
   // type = multi-enum
   // desc = metric(s) to be evaluated on the evaluation set(s)
@@ -699,6 +698,7 @@ struct Config {
   // desc = support multiple metrics, separated by ``,``
   std::vector<std::string> train_metric;
 
+  // [doc-only]
   // type = multi-enum
   // default = ""
   // desc = metric(s) to be evaluated on the evaluation set(s)
diff --git a/src/io/config_auto.cpp b/src/io/config_auto.cpp
index 3964c3e7af9c..d341f844d2f4 100644
--- a/src/io/config_auto.cpp
+++ b/src/io/config_auto.cpp
@@ -135,8 +135,6 @@ std::unordered_map<std::string, std::string> Config::alias_table({
   {"num_classes", "num_class"},
   {"unbalance", "is_unbalance"},
   {"unbalanced_sets", "is_unbalance"},
-  {"metrics", "train_metric"},
-  {"metric_types", "train_metric"},
   {"output_freq", "metric_freq"},
   {"training_metric", "is_provide_training_metric"},
   {"is_training_metric", "is_provide_training_metric"},
@@ -491,10 +489,6 @@ void Config::GetMembersFromString(const std::unordered_map<std::string, std::str
     label_gain = Common::StringToArray<double>(tmp_str, ',');
   }
 
-  if (GetString(params, "valid_metric", &tmp_str)) {
-    valid_metric = Common::Split(tmp_str.c_str(), ',');
-  }
-
   GetInt(params, "metric_freq", &metric_freq);
   CHECK(metric_freq >0);
 
@@ -613,7 +607,6 @@ std::string Config::SaveMembersToString() const {
   str_buf << "[tweedie_variance_power: " << tweedie_variance_power << "]\n";
   str_buf << "[max_position: " << max_position << "]\n";
   str_buf << "[label_gain: " << Common::Join(label_gain,",") << "]\n";
-  str_buf << "[valid_metric: " << Common::Join(valid_metric,",") << "]\n";
   str_buf << "[metric_freq: " << metric_freq << "]\n";
   str_buf << "[is_provide_training_metric: " << is_provide_training_metric << "]\n";
   str_buf << "[eval_at: " << Common::Join(eval_at,",") << "]\n";

From bea28512f2fbd7212d3e700c814104b39d1b6e4d Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Wed, 19 Dec 2018 20:18:51 +0900
Subject: [PATCH 09/19] backward compatibility

---
 docs/Parameters.rst       |  4 +++
 include/LightGBM/config.h |  6 ++++
 src/io/config.cpp         | 69 +++++++++++++++++++--------------------
 src/io/config_auto.cpp    |  1 +
 4 files changed, 45 insertions(+), 35 deletions(-)

diff --git a/docs/Parameters.rst b/docs/Parameters.rst
index baac3ce806ea..e92fb2a2a5eb 100644
--- a/docs/Parameters.rst
+++ b/docs/Parameters.rst
@@ -795,6 +795,10 @@ Metric Parameters
 
    -  metric(s) to be evaluated on the evaluation set(s)
 
+-  ``metric`` :raw-html:`<a id="metric" title="Permalink to this parameter" href="#metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
+
+   -  dummpy parameter
+
 -  ``metric_freq`` :raw-html:`<a id="metric_freq" title="Permalink to this parameter" href="#metric_freq">&#x1F517;&#xFE0E;</a>`, default = ``1``, type = int, aliases: ``output_freq``, constraints: ``metric_freq > 0``
 
    -  frequency for metric output
diff --git a/include/LightGBM/config.h b/include/LightGBM/config.h
index 0702046529d0..97777faca389 100644
--- a/include/LightGBM/config.h
+++ b/include/LightGBM/config.h
@@ -704,6 +704,12 @@ struct Config {
   // desc = metric(s) to be evaluated on the evaluation set(s)
   std::vector<std::string> valid_metric;
 
+  // [doc-only]
+  // type = multi-enum
+  // default = ""
+  // desc = dummpy parameter
+  std::vector<std::string> metric;
+
   // check = >0
   // alias = output_freq
   // desc = frequency for metric output
diff --git a/src/io/config.cpp b/src/io/config.cpp
index 4ee8f1b0a54f..5fec4653464a 100644
--- a/src/io/config.cpp
+++ b/src/io/config.cpp
@@ -68,9 +68,12 @@ void GetObjectiveType(const std::unordered_map<std::string, std::string>& params
   }
 }
 
-void GetTrainMetricType(const std::unordered_map<std::string, std::string>& params, std::vector<std::string>* metric) {
+
+void SetMetircType(const std::unordered_map<std::string, std::string>& params,
+                   const std::string& name,
+                   std::vector<std::string>* metric) {
   std::string value;
-  if (Config::GetString(params, "train_metric", &value)) {
+  if (Config::GetString(params, name, &value)) {
     // clear old metrics
     metric->clear();
     // to lower
@@ -90,47 +93,44 @@ void GetTrainMetricType(const std::unordered_map<std::string, std::string>& para
     }
     metric->shrink_to_fit();
   }
-  // add names of objective function if not providing metric
-  if (metric->empty() && value.size() == 0) {
-    if (Config::GetString(params, "objective", &value)) {
-      std::transform(value.begin(), value.end(), value.begin(), Common::tolower);
-      metric->push_back(value);
-    }
-  }
 }
 
-void GetValidMetricType(const std::unordered_map<std::string, std::string>& params, std::vector<std::string>* metric) {
+void SetDefaultMetircType(const std::unordered_map<std::string, std::string>& params,
+                          std::vector<std::string>* metric) {
   std::string value;
-  if (Config::GetString(params, "valid_metric", &value)) {
-    // clear old metrics
-    metric->clear();
-    // to lower
+  // add names of objective function if not providing metric
+  if (Config::GetString(params, "objective", &value)) {
     std::transform(value.begin(), value.end(), value.begin(), Common::tolower);
-    // split
-    std::vector<std::string> metrics = Common::Split(value.c_str(), ',');
-    // remove duplicate
-    std::unordered_set<std::string> metric_sets;
-    for (auto& met : metrics) {
-      std::transform(met.begin(), met.end(), met.begin(), Common::tolower);
-      if (metric_sets.count(met) <= 0) {
-        metric_sets.insert(met);
-      }
+    metric->push_back(value);
+  }
+}
+
+void GetMetricType(const std::unordered_map<std::string, std::string>& params,
+                   std::vector<std::string>* train_metric, std::vector<std::string>* valid_metric) {
+  if (params.count("metric") > 0) {
+    //printf("HOGE\n");
+    //printf("metric: %s\n", params.at("metric").c_str());
+    SetMetircType(params, "metric", train_metric);
+    SetMetircType(params, "metric", valid_metric);
+  } else {
+    //printf("FOOO\n");
+    if (params.count("train_metric") > 0) {
+      //printf("train_metric: %s\n", params.at("train_metric").c_str());
+      SetMetircType(params, "train_metric", train_metric);
     }
-    for (auto& met : metric_sets) {
-      metric->push_back(met);
+    if (params.count("valid_metric") > 0) {
+      //printf("valid_metric: %s\n", params.at("valid_metric").c_str());
+      SetMetircType(params, "valid_metric", valid_metric);
     }
-    metric->shrink_to_fit();
   }
-  // add names of objective function if not providing metric
-  if (metric->empty() && value.size() == 0) {
-    if (Config::GetString(params, "objective", &value)) {
-      std::transform(value.begin(), value.end(), value.begin(), Common::tolower);
-      metric->push_back(value);
-    }
+  if (train_metric->empty()) {
+    SetDefaultMetircType(params, train_metric);
+  }
+  if (valid_metric->empty()) {
+    SetDefaultMetircType(params, valid_metric);
   }
 }
 
-
 void GetTaskType(const std::unordered_map<std::string, std::string>& params, TaskType* task) {
   std::string value;
   if (Config::GetString(params, "task", &value)) {
@@ -196,8 +196,7 @@ void Config::Set(const std::unordered_map<std::string, std::string>& params) {
 
   GetTaskType(params, &task);
   GetBoostingType(params, &boosting);
-  GetTrainMetricType(params, &train_metric);
-  GetValidMetricType(params, &valid_metric);
+  GetMetricType(params, &train_metric, &valid_metric);
   GetObjectiveType(params, &objective);
   GetDeviceType(params, &device_type);
   GetTreeLearnerType(params, &tree_learner);
diff --git a/src/io/config_auto.cpp b/src/io/config_auto.cpp
index d341f844d2f4..e8feaa405abe 100644
--- a/src/io/config_auto.cpp
+++ b/src/io/config_auto.cpp
@@ -248,6 +248,7 @@ std::unordered_set<std::string> Config::parameter_set({
   "label_gain",
   "train_metric",
   "valid_metric",
+  "metric",
   "metric_freq",
   "is_provide_training_metric",
   "eval_at",

From e0ca9e0cdcc0e1098297d971bfb3cb8b93eda998 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Wed, 19 Dec 2018 22:05:01 +0900
Subject: [PATCH 10/19] reset

---
 examples/python-guide/advanced_example.py     |  2 +-
 .../notebooks/interactive_plot_example.ipynb  |  9 +++---
 examples/python-guide/plot_example.py         |  3 +-
 examples/python-guide/simple_example.py       |  2 +-
 python-package/lightgbm/engine.py             |  2 +-
 python-package/lightgbm/sklearn.py            |  2 +-
 tests/python_package_test/test_engine.py      | 32 +++++++++----------
 tests/python_package_test/test_plotting.py    |  5 +--
 8 files changed, 26 insertions(+), 31 deletions(-)

diff --git a/examples/python-guide/advanced_example.py b/examples/python-guide/advanced_example.py
index 865b1307249d..cef8ad6c944a 100644
--- a/examples/python-guide/advanced_example.py
+++ b/examples/python-guide/advanced_example.py
@@ -36,7 +36,7 @@
 params = {
     'boosting_type': 'gbdt',
     'objective': 'binary',
-    'train_metric': 'binary_logloss',
+    'metric': 'binary_logloss',
     'num_leaves': 31,
     'learning_rate': 0.05,
     'feature_fraction': 0.9,
diff --git a/examples/python-guide/notebooks/interactive_plot_example.ipynb b/examples/python-guide/notebooks/interactive_plot_example.ipynb
index f86377a52950..e472404d9aa4 100644
--- a/examples/python-guide/notebooks/interactive_plot_example.ipynb
+++ b/examples/python-guide/notebooks/interactive_plot_example.ipynb
@@ -103,8 +103,7 @@
    "source": [
     "params = {\n",
     "    'num_leaves': 5,\n",
-    "    'train_metric': ['l1', 'l2'],\n",
-    "    'valid_metric': ['l1', 'l2'],\n",
+    "    'metric': ['l1', 'l2'],\n",
     "    'verbose': -1\n",
     "}"
    ]
@@ -198,9 +197,9 @@
    "source": [
     "if INTERACTIVE:\n",
     "    # create widget to switch between metrics\n",
-    "    interact(render_metric, metric_name=params['valid_metric'])\n",
+    "    interact(render_metric, metric_name=params['metric'])\n",
     "else:\n",
-    "    render_metric(params['valid_metric'][0])"
+    "    render_metric(params['metric'][0])"
    ]
   },
   {
@@ -337,7 +336,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.6"
+   "version": "3.6.3"
   },
   "varInspector": {
    "cols": {
diff --git a/examples/python-guide/plot_example.py b/examples/python-guide/plot_example.py
index e935930aa95b..eca896982fa9 100644
--- a/examples/python-guide/plot_example.py
+++ b/examples/python-guide/plot_example.py
@@ -25,8 +25,7 @@
 # specify your configurations as a dict
 params = {
     'num_leaves': 5,
-    'train_metric': ('l1', 'l2'),
-    'valid_metric': ('l1', 'l2'),
+    'metric': ('l1', 'l2'),
     'verbose': 0
 }
 
diff --git a/examples/python-guide/simple_example.py b/examples/python-guide/simple_example.py
index 403956d8642a..18f07cd601cd 100644
--- a/examples/python-guide/simple_example.py
+++ b/examples/python-guide/simple_example.py
@@ -22,7 +22,7 @@
 params = {
     'boosting_type': 'gbdt',
     'objective': 'regression',
-    'valid_metric': {'l2', 'l1'},
+    'metric': {'l2', 'l1'},
     'num_leaves': 31,
     'learning_rate': 0.05,
     'feature_fraction': 0.9,
diff --git a/python-package/lightgbm/engine.py b/python-package/lightgbm/engine.py
index da4e42069a29..da7459fced71 100644
--- a/python-package/lightgbm/engine.py
+++ b/python-package/lightgbm/engine.py
@@ -446,7 +446,7 @@ def cv(params, train_set, num_boost_round=100,
              .set_categorical_feature(categorical_feature)
 
     if metrics is not None:
-        params['valid_metric'] = metrics
+        params['metric'] = metrics
 
     results = collections.defaultdict(list)
     cvfolds = _make_n_folds(train_set, folds=folds, nfold=nfold,
diff --git a/python-package/lightgbm/sklearn.py b/python-package/lightgbm/sklearn.py
index 5db0bdb82b54..f69d28f683ed 100644
--- a/python-package/lightgbm/sklearn.py
+++ b/python-package/lightgbm/sklearn.py
@@ -477,7 +477,7 @@ def fit(self, X, y,
             # concatenate metric from params (or default if not provided in params) and eval_metric
             original_metric = [original_metric] if isinstance(original_metric, (string_type, type(None))) else original_metric
             eval_metric = [eval_metric] if isinstance(eval_metric, (string_type, type(None))) else eval_metric
-            params['valid_metric'] = set(original_metric + eval_metric)  # TODO: train_metric
+            params['metric'] = set(original_metric + eval_metric)  # TODO: train_metric
 
         if not isinstance(X, DataFrame):
             X, y = _LGBMCheckXY(X, y, accept_sparse=True, force_all_finite=False, ensure_min_samples=2)
diff --git a/tests/python_package_test/test_engine.py b/tests/python_package_test/test_engine.py
index fc73169c5b66..e89b59661f9f 100644
--- a/tests/python_package_test/test_engine.py
+++ b/tests/python_package_test/test_engine.py
@@ -30,7 +30,7 @@ def test_binary(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'binary',
-            'valid_metric': 'binary_logloss',
+            'metric': 'binary_logloss',
             'verbose': -1,
             'num_iteration': 50  # test num_iteration in dict here
         }
@@ -57,7 +57,7 @@ def test_rf(self):
             'bagging_fraction': 0.5,
             'feature_fraction': 0.5,
             'num_leaves': 50,
-            'valid_metric': 'binary_logloss',
+            'metric': 'binary_logloss',
             'verbose': -1
         }
         lgb_train = lgb.Dataset(X_train, y_train)
@@ -76,7 +76,7 @@ def test_regression(self):
         X, y = load_boston(True)
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
-            'valid_metric': 'l2',
+            'metric': 'l2',
             'verbose': -1
         }
         lgb_train = lgb.Dataset(X_train, y_train)
@@ -102,7 +102,7 @@ def test_missing_value_handle(self):
         lgb_eval = lgb.Dataset(X_train, y_train)
 
         params = {
-            'valid_metric': 'l2',
+            'metric': 'l2',
             'verbose': -1,
             'boost_from_average': False
         }
@@ -127,7 +127,7 @@ def test_missing_value_handle_na(self):
 
         params = {
             'objective': 'regression',
-            'valid_metric': 'auc',
+            'metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -159,7 +159,7 @@ def test_missing_value_handle_zero(self):
 
         params = {
             'objective': 'regression',
-            'valid_metric': 'auc',
+            'metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -191,7 +191,7 @@ def test_missing_value_handle_none(self):
 
         params = {
             'objective': 'regression',
-            'valid_metric': 'auc',
+            'metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -224,7 +224,7 @@ def test_categorical_handle(self):
 
         params = {
             'objective': 'regression',
-            'valid_metric': 'auc',
+            'metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -261,7 +261,7 @@ def test_categorical_handle_na(self):
 
         params = {
             'objective': 'regression',
-            'valid_metric': 'auc',
+            'metric': 'auc',
             'verbose': -1,
             'boost_from_average': False,
             'min_data': 1,
@@ -292,7 +292,7 @@ def test_multiclass(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'multiclass',
-            'valid_metric': 'multi_logloss',
+            'metric': 'multi_logloss',
             'num_class': 10,
             'verbose': -1
         }
@@ -314,7 +314,7 @@ def test_multiclass_rf(self):
         params = {
             'boosting_type': 'rf',
             'objective': 'multiclass',
-            'valid_metric': 'multi_logloss',
+            'metric': 'multi_logloss',
             'bagging_freq': 1,
             'bagging_fraction': 0.6,
             'feature_fraction': 0.6,
@@ -365,7 +365,7 @@ def test_early_stopping(self):
         X, y = load_breast_cancer(True)
         params = {
             'objective': 'binary',
-            'valid_metric': 'binary_logloss',
+            'metric': 'binary_logloss',
             'verbose': -1
         }
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
@@ -397,7 +397,7 @@ def test_continue_train(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'regression',
-            'valid_metric': 'l1',
+            'metric': 'l1',
             'verbose': -1
         }
         lgb_train = lgb.Dataset(X_train, y_train, free_raw_data=False)
@@ -426,7 +426,7 @@ def test_continue_train_multiclass(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         params = {
             'objective': 'multiclass',
-            'valid_metric': 'multi_logloss',
+            'metric': 'multi_logloss',
             'num_class': 3,
             'verbose': -1
         }
@@ -450,7 +450,7 @@ def test_cv(self):
         params = {'verbose': -1}
         lgb_train = lgb.Dataset(X_train, y_train)
         # shuffle = False, override metric in params
-        params_with_metric = {'valid_metric': 'l2', 'verbose': -1}
+        params_with_metric = {'metric': 'l2', 'verbose': -1}
         cv_res = lgb.cv(params_with_metric, lgb_train, num_boost_round=10,
                         nfold=3, stratified=False, shuffle=False,
                         metrics='l1', verbose_eval=False)
@@ -595,7 +595,7 @@ def test_reference_chain(self):
         # take subsets and train
         tmp_dat_train = tmp_dat.subset(np.arange(80))
         tmp_dat_val = tmp_dat.subset(np.arange(80, 100)).subset(np.arange(18))
-        params = {'objective': 'regression_l2', 'train_metric': 'rmse', 'valid_metric': 'rmse'}
+        params = {'objective': 'regression_l2', 'metric': 'rmse'}
         evals_result = {}
         gbm = lgb.train(params, tmp_dat_train, num_boost_round=20,
                         valid_sets=[tmp_dat_train, tmp_dat_val], evals_result=evals_result)
diff --git a/tests/python_package_test/test_plotting.py b/tests/python_package_test/test_plotting.py
index 6ee332355ae6..05a25e64d4a2 100644
--- a/tests/python_package_test/test_plotting.py
+++ b/tests/python_package_test/test_plotting.py
@@ -104,10 +104,7 @@ def test_create_tree_digraph(self):
     @unittest.skipIf(not MATPLOTLIB_INSTALLED, 'matplotlib is not installed')
     def test_plot_metrics(self):
         test_data = lgb.Dataset(self.X_test, self.y_test, reference=self.train_data)
-        self.params.update({
-            "train_metric": ["binary_logloss", "binary_error"],
-            "valid_metric": ["binary_logloss", "binary_error"]
-        })
+        self.params.update({"metric": {"binary_logloss", "binary_error"}})
 
         evals_result0 = {}
         gbm0 = lgb.train(self.params, self.train_data,

From 95729b2103cedc2cdc2fa8782e00d0f396a168f1 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Wed, 19 Dec 2018 22:19:53 +0900
Subject: [PATCH 11/19] add a test

---
 tests/python_package_test/test_engine.py | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/tests/python_package_test/test_engine.py b/tests/python_package_test/test_engine.py
index e89b59661f9f..3cfeb7054ff0 100644
--- a/tests/python_package_test/test_engine.py
+++ b/tests/python_package_test/test_engine.py
@@ -808,3 +808,23 @@ def test_constant_features_multiclassova(self):
         }
         self.test_constant_features([0.0, 1.0, 2.0, 0.0], [0.5, 0.25, 0.25], params)
         self.test_constant_features([0.0, 1.0, 2.0, 1.0], [0.25, 0.5, 0.25], params)
+
+    def test_train_and_valid_metric(self):
+        X, y = load_boston(True)
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
+        params = {
+            'train_metric': ['l2'],
+            'valid_metric': ['l1', 'l2'],
+            'verbose': -1
+        }
+        lgb_train = lgb.Dataset(X_train, y_train)
+        lgb_eval = lgb.Dataset(X_test, y_test, reference=lgb_train)
+        evals_result = {}
+        gbm = lgb.train(params, lgb_train,
+                        num_boost_round=50,
+                        valid_sets=[lgb_train, lgb_eval],
+                        verbose_eval=False,
+                        evals_result=evals_result)
+        self.assertTrue('l2' in evals_result['training'])
+        self.assertTrue('l1' in evals_result['valid_1'])
+        self.assertTrue('l2' in evals_result['valid_1'])

From 452d21f1d83cc6f2dbd0022fef4afb27c8e91ce5 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 00:16:58 +0900
Subject: [PATCH 12/19] fix c_api

---
 src/c_api.cpp | 38 +++++++++++++++++---------------------
 1 file changed, 17 insertions(+), 21 deletions(-)

diff --git a/src/c_api.cpp b/src/c_api.cpp
index 4ddc510aed98..298ed854dc5b 100644
--- a/src/c_api.cpp
+++ b/src/c_api.cpp
@@ -300,15 +300,13 @@ class Booster {
   int GetEvalCounts(int data_idx) const {
     int ret = 0;
     if (data_idx == 0) {
-        for (const auto& metric : train_metric_) {
-          ret += static_cast<int>(metric->GetName().size());
-        }
+      for (const auto& metric : train_metric_) {
+        ret += static_cast<int>(metric->GetName().size());
+      }
     } else {
-        for (size_t i = 0; i < valid_metrics_.size(); ++i) {
-          for (size_t j = 0; j < valid_metrics_[i].size(); ++j) {
-            ret += static_cast<int>(valid_metrics_[i][j]->GetName().size());
-          }
-        }
+      for (size_t j = 0; j < valid_metrics_[data_idx - 1].size(); ++j) {
+        ret += static_cast<int>(valid_metrics_[data_idx - 1][j]->GetName().size());
+      }
     }
     return ret;
   }
@@ -316,22 +314,20 @@ class Booster {
   int GetEvalNames(int data_idx, char** out_strs) const {
     int idx = 0;
     if (data_idx == 0) {
-        for (const auto& metric : train_metric_) {
-          for (const auto& name : metric->GetName()) {
-            std::memcpy(out_strs[idx], name.c_str(), name.size() + 1);
-            ++idx;
-          }
+      for (const auto& metric : train_metric_) {
+        for (const auto& name : metric->GetName()) {
+          std::memcpy(out_strs[idx], name.c_str(), name.size() + 1);
+          ++idx;
         }
+      }
     } else {
-        for (size_t i = 0; i < valid_metrics_.size(); ++i) {
-          for (size_t j = 0; j < valid_metrics_[i].size(); ++j) {
-              auto& metric = valid_metrics_[i][j];
-              for (const auto& name : metric->GetName()) {
-                std::memcpy(out_strs[idx], name.c_str(), name.size() + 1);
-                ++idx;
-              }
-          }
+      for (size_t j = 0; j < valid_metrics_[data_idx - 1].size(); ++j) {
+        auto& metric = valid_metrics_[data_idx - 1][j];
+        for (const auto& name : metric->GetName()) {
+          std::memcpy(out_strs[idx], name.c_str(), name.size() + 1);
+          ++idx;
         }
+      }
     }
     return idx;
   }

From 6892408a4027736c271a3ff4a27b22f899004c78 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 00:38:13 +0900
Subject: [PATCH 13/19] update sklearn API

[ci skip]
---
 python-package/lightgbm/sklearn.py        | 76 ++++++++++++++---------
 tests/python_package_test/test_sklearn.py | 16 ++++-
 2 files changed, 60 insertions(+), 32 deletions(-)

diff --git a/python-package/lightgbm/sklearn.py b/python-package/lightgbm/sklearn.py
index f69d28f683ed..5c053cc128f6 100644
--- a/python-package/lightgbm/sklearn.py
+++ b/python-package/lightgbm/sklearn.py
@@ -330,7 +330,8 @@ def fit(self, X, y,
             sample_weight=None, init_score=None, group=None,
             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,
+            eval_metric=None, eval_train_metric=None, eval_valid_metric=None,
+            early_stopping_rounds=None, verbose=True,
             feature_name='auto', categorical_feature='auto', callbacks=None):
         """Build a gradient boosting model from the training set (X, y).
 
@@ -460,24 +461,29 @@ def fit(self, X, y,
             feval = _eval_function_wrapper(eval_metric)
         else:
             feval = None
-            # register default metric for consistency with callable eval_metric case
-            original_metric = self._objective if isinstance(self._objective, string_type) else None
-            if original_metric is None:
-                # try to deduce from class instance
-                if isinstance(self, LGBMRegressor):
-                    original_metric = "l2"
-                elif isinstance(self, LGBMClassifier):
-                    original_metric = "multi_logloss" if self._n_classes > 2 else "binary_logloss"
-                elif isinstance(self, LGBMRanker):
-                    original_metric = "ndcg"
-            # overwrite default metric by explicitly set metric
-            for metric_alias in ['metric', 'metrics', 'metric_types']:
-                if metric_alias in params:
-                    original_metric = params.pop(metric_alias)
-            # concatenate metric from params (or default if not provided in params) and eval_metric
-            original_metric = [original_metric] if isinstance(original_metric, (string_type, type(None))) else original_metric
-            eval_metric = [eval_metric] if isinstance(eval_metric, (string_type, type(None))) else eval_metric
-            params['metric'] = set(original_metric + eval_metric)  # TODO: train_metric
+            if eval_metric:
+                params['metric'] = eval_metric
+            else:
+                params['train_metric'] = eval_train_metric
+                params['valid_metric'] = eval_valid_metric
+            # # register default metric for consistency with callable eval_metric case
+            # original_metric = self._objective if isinstance(self._objective, string_type) else None
+            # if original_metric is None:
+            #     # try to deduce from class instance
+            #     if isinstance(self, LGBMRegressor):
+            #         original_metric = "l2"
+            #     elif isinstance(self, LGBMClassifier):
+            #         original_metric = "multi_logloss" if self._n_classes > 2 else "binary_logloss"
+            #     elif isinstance(self, LGBMRanker):
+            #         original_metric = "ndcg"
+            # # overwrite default metric by explicitly set metric
+            # for metric_alias in ['metric', 'metrics', 'metric_types']:
+            #     if metric_alias in params:
+            #         original_metric = params.pop(metric_alias)
+            # # concatenate metric from params (or default if not provided in params) and eval_metric
+            # original_metric = [original_metric] if isinstance(original_metric, (string_type, type(None))) else original_metric
+            # eval_metric = [eval_metric] if isinstance(eval_metric, (string_type, type(None))) else eval_metric
+            # params['metric'] = set(original_metric + eval_metric)  # TODO: train_metric
 
         if not isinstance(X, DataFrame):
             X, y = _LGBMCheckXY(X, y, accept_sparse=True, force_all_finite=False, ensure_min_samples=2)
@@ -518,7 +524,6 @@ def _get_meta_data(collection, i):
                 # reduce cost for prediction training data
                 if valid_data[0] is X and valid_data[1] is y:
                     valid_set = train_set
-                    # params['train_metric'] = set(original_metric + eval_metric)
                 else:
                     valid_weight = _get_meta_data(eval_sample_weight, i)
                     if _get_meta_data(eval_class_weight, i) is not None:
@@ -669,7 +674,8 @@ class LGBMRegressor(LGBMModel, _LGBMRegressorBase):
     def fit(self, X, y,
             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,
+            eval_init_score=None, eval_metric=None, eval_train_metric=None,
+            eval_valid_metric=None, early_stopping_rounds=None,
             verbose=True, feature_name='auto', categorical_feature='auto', callbacks=None):
         """Docstring is inherited from the LGBMModel."""
         super(LGBMRegressor, self).fit(X, y, sample_weight=sample_weight,
@@ -678,6 +684,8 @@ def fit(self, X, y,
                                        eval_sample_weight=eval_sample_weight,
                                        eval_init_score=eval_init_score,
                                        eval_metric=eval_metric,
+                                       eval_train_metric=eval_train_metric,
+                                       eval_valid_metric=eval_valid_metric,
                                        early_stopping_rounds=early_stopping_rounds,
                                        verbose=verbose, feature_name=feature_name,
                                        categorical_feature=categorical_feature,
@@ -696,6 +704,7 @@ def fit(self, X, y,
             sample_weight=None, init_score=None,
             eval_set=None, eval_names=None, eval_sample_weight=None,
             eval_class_weight=None, eval_init_score=None, eval_metric=None,
+            eval_train_metric=None, eval_valid_metric=None,
             early_stopping_rounds=None, verbose=True,
             feature_name='auto', categorical_feature='auto', callbacks=None):
         """Docstring is inherited from the LGBMModel."""
@@ -711,15 +720,17 @@ def fit(self, X, y,
             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"
+            for _metric in (eval_metric, eval_train_metric, eval_valid_metric):
+                if _metric in ('logloss', 'binary_logloss'):
+                    _metric = "multi_logloss"
+                elif _metric in ('error', 'binary_error'):
+                    _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'
+            for _metric in (eval_metric, eval_train_metric, eval_valid_metric):
+                if _metric in ('logloss', 'multi_logloss'):
+                    _metric = 'binary_logloss'
+                elif _metric in ('error', 'multi_error'):
+                    _metric = 'binary_error'
 
         if eval_set is not None:
             if isinstance(eval_set, tuple):
@@ -737,6 +748,8 @@ def fit(self, X, y,
                                         eval_class_weight=eval_class_weight,
                                         eval_init_score=eval_init_score,
                                         eval_metric=eval_metric,
+                                        eval_train_metric=eval_train_metric,
+                                        eval_valid_metric=eval_valid_metric,
                                         early_stopping_rounds=early_stopping_rounds,
                                         verbose=verbose, feature_name=feature_name,
                                         categorical_feature=categorical_feature,
@@ -824,7 +837,8 @@ def fit(self, X, y,
             sample_weight=None, init_score=None, group=None,
             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,
+            eval_train_metric=None, eval_valid_metric=None, eval_at=[1],
+            early_stopping_rounds=None, verbose=True,
             feature_name='auto', categorical_feature='auto', callbacks=None):
         """Docstring is inherited from the LGBMModel."""
         # check group data
@@ -850,6 +864,8 @@ def fit(self, X, y,
                                     eval_sample_weight=eval_sample_weight,
                                     eval_init_score=eval_init_score, eval_group=eval_group,
                                     eval_metric=eval_metric,
+                                    eval_train_metric=eval_train_metric,
+                                    eval_valid_metric=eval_valid_metric,
                                     early_stopping_rounds=early_stopping_rounds,
                                     verbose=verbose, feature_name=feature_name,
                                     categorical_feature=categorical_feature,
diff --git a/tests/python_package_test/test_sklearn.py b/tests/python_package_test/test_sklearn.py
index 7ee85649bfa0..e2c4495ff32f 100644
--- a/tests/python_package_test/test_sklearn.py
+++ b/tests/python_package_test/test_sklearn.py
@@ -82,7 +82,8 @@ def objective_ls(y_true, y_pred):
         X, y = load_boston(True)
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         gbm = lgb.LGBMRegressor(n_estimators=50, silent=True, objective=objective_ls)
-        gbm.fit(X_train, y_train, eval_set=[(X_test, y_test)], early_stopping_rounds=5, verbose=False)
+        gbm.fit(X_train, y_train, eval_set=[(X_test, y_test)], eval_metric='regression',
+                early_stopping_rounds=5, verbose=False)
         ret = mean_squared_error(y_test, gbm.predict(X_test))
         self.assertLess(ret, 100)
         self.assertAlmostEqual(ret, gbm.evals_result_['valid_0']['l2'][gbm.best_iteration_ - 1], places=5)
@@ -100,7 +101,8 @@ def binary_error(y_test, y_pred):
         X, y = load_digits(2, True)
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         gbm = lgb.LGBMClassifier(n_estimators=50, silent=True, objective=logregobj)
-        gbm.fit(X_train, y_train, eval_set=[(X_test, y_test)], early_stopping_rounds=5, verbose=False)
+        gbm.fit(X_train, y_train, eval_set=[(X_test, y_test)], eval_metric='binary_error',
+                early_stopping_rounds=5, verbose=False)
         ret = binary_error(y_test, gbm.predict(X_test))
         self.assertLess(ret, 0.1)
 
@@ -279,3 +281,13 @@ def test_predict(self):
         self.assertRaises(AssertionError,
                           np.testing.assert_allclose,
                           res_engine, res_sklearn_params)
+
+    def test_regressor_train_and_valid_metric(self):
+        X, y = load_boston(True)
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
+        gbm = lgb.LGBMRegressor(n_estimators=10, silent=True)
+        gbm.fit(X_train, y_train, eval_set=[(X_train, y_train), (X_test, y_test)],
+                eval_train_metric='l2', eval_valid_metric=['l1', 'l2'], verbose=False)
+        self.assertTrue('l2' in gbm.evals_result_['training'])
+        self.assertTrue('l1' in gbm.evals_result_['valid_1'])
+        self.assertTrue('l2' in gbm.evals_result_['valid_1'])

From c6062de560dc5ef618e85e8b84152e01282c01c6 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 12:36:34 +0900
Subject: [PATCH 14/19] remove printf

---
 src/io/config.cpp | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/src/io/config.cpp b/src/io/config.cpp
index 5fec4653464a..0da8ff313e5c 100644
--- a/src/io/config.cpp
+++ b/src/io/config.cpp
@@ -108,18 +108,13 @@ void SetDefaultMetircType(const std::unordered_map<std::string, std::string>& pa
 void GetMetricType(const std::unordered_map<std::string, std::string>& params,
                    std::vector<std::string>* train_metric, std::vector<std::string>* valid_metric) {
   if (params.count("metric") > 0) {
-    //printf("HOGE\n");
-    //printf("metric: %s\n", params.at("metric").c_str());
     SetMetircType(params, "metric", train_metric);
     SetMetircType(params, "metric", valid_metric);
   } else {
-    //printf("FOOO\n");
     if (params.count("train_metric") > 0) {
-      //printf("train_metric: %s\n", params.at("train_metric").c_str());
       SetMetircType(params, "train_metric", train_metric);
     }
     if (params.count("valid_metric") > 0) {
-      //printf("valid_metric: %s\n", params.at("valid_metric").c_str());
       SetMetircType(params, "valid_metric", valid_metric);
     }
   }

From a3fb1d632c3490e680152546d3695bac6b93106a Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 12:51:10 +0900
Subject: [PATCH 15/19] add tests

---
 tests/python_package_test/test_sklearn.py | 40 ++++++++++++++++++++---
 1 file changed, 36 insertions(+), 4 deletions(-)

diff --git a/tests/python_package_test/test_sklearn.py b/tests/python_package_test/test_sklearn.py
index fa60cf3e4a7d..e41fa2509a5b 100644
--- a/tests/python_package_test/test_sklearn.py
+++ b/tests/python_package_test/test_sklearn.py
@@ -297,7 +297,39 @@ def test_regressor_train_and_valid_metric(self):
         X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
         gbm = lgb.LGBMRegressor(n_estimators=10, silent=True)
         gbm.fit(X_train, y_train, eval_set=[(X_train, y_train), (X_test, y_test)],
-                eval_train_metric='l2', eval_valid_metric=['l1', 'l2'], verbose=False)
-        self.assertTrue('l2' in gbm.evals_result_['training'])
-        self.assertTrue('l1' in gbm.evals_result_['valid_1'])
-        self.assertTrue('l2' in gbm.evals_result_['valid_1'])
+                eval_train_metric='l2', eval_valid_metric=['l1', 'l2'],
+                early_stopping_rounds=5, verbose=False)
+        self.assertIn('l2', gbm.evals_result_['training'])
+        self.assertIn('l1', gbm.evals_result_['valid_1'])
+        self.assertIn('l2', gbm.evals_result_['valid_1'])
+
+    def test_classifier_train_and_valid_metric(self):
+        X, y = load_breast_cancer(True)
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
+        gbm = lgb.LGBMClassifier(n_estimators=10, silent=True)
+        gbm.fit(X_train, y_train, eval_set=[(X_train, y_train), (X_test, y_test)],
+                eval_train_metric='binary_error', eval_valid_metric=['binary_logloss', 'binary_error'],
+                early_stopping_rounds=5, verbose=False)
+        self.assertIn('binary_error', gbm.evals_result_['training'])
+        self.assertIn('binary_error', gbm.evals_result_['valid_1'])
+        self.assertIn('binary_logloss', gbm.evals_result_['valid_1'])
+
+    def test_ranker_train_and_valid_metric(self):
+        X_train, y_train = load_svmlight_file(os.path.join(os.path.dirname(os.path.realpath(__file__)),
+                                                           '../../examples/lambdarank/rank.train'))
+        X_test, y_test = load_svmlight_file(os.path.join(os.path.dirname(os.path.realpath(__file__)),
+                                                         '../../examples/lambdarank/rank.test'))
+        q_train = np.loadtxt(os.path.join(os.path.dirname(os.path.realpath(__file__)),
+                                          '../../examples/lambdarank/rank.train.query'))
+        q_test = np.loadtxt(os.path.join(os.path.dirname(os.path.realpath(__file__)),
+                                         '../../examples/lambdarank/rank.test.query'))
+        gbm = lgb.LGBMRanker(n_estimators=10, silent=True)
+        gbm.fit(X_train, y_train, group=q_train, eval_set=[(X_train, y_train), (X_test, y_test)],
+                eval_group=[q_train, q_test], eval_at=[1, 3],
+                eval_train_metric='ndcg', eval_valid_metric=['l2', 'ndcg'],
+                early_stopping_rounds=5, verbose=False)
+        self.assertIn('ndcg@1', gbm.evals_result_['training'])
+        self.assertIn('ndcg@3', gbm.evals_result_['training'])
+        self.assertIn('ndcg@1', gbm.evals_result_['valid_1'])
+        self.assertIn('ndcg@3', gbm.evals_result_['valid_1'])
+        self.assertIn('l2', gbm.evals_result_['valid_1'])

From 9d5882b6ed5213f987b06a16fce429e35a437863 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 12:53:07 +0900
Subject: [PATCH 16/19] cleanup

---
 python-package/lightgbm/sklearn.py | 18 ------------------
 1 file changed, 18 deletions(-)

diff --git a/python-package/lightgbm/sklearn.py b/python-package/lightgbm/sklearn.py
index f85a555efb36..6b4cca23736e 100644
--- a/python-package/lightgbm/sklearn.py
+++ b/python-package/lightgbm/sklearn.py
@@ -466,24 +466,6 @@ def fit(self, X, y,
             else:
                 params['train_metric'] = eval_train_metric
                 params['valid_metric'] = eval_valid_metric
-            # # register default metric for consistency with callable eval_metric case
-            # original_metric = self._objective if isinstance(self._objective, string_type) else None
-            # if original_metric is None:
-            #     # try to deduce from class instance
-            #     if isinstance(self, LGBMRegressor):
-            #         original_metric = "l2"
-            #     elif isinstance(self, LGBMClassifier):
-            #         original_metric = "multi_logloss" if self._n_classes > 2 else "binary_logloss"
-            #     elif isinstance(self, LGBMRanker):
-            #         original_metric = "ndcg"
-            # # overwrite default metric by explicitly set metric
-            # for metric_alias in ['metric', 'metrics', 'metric_types']:
-            #     if metric_alias in params:
-            #         original_metric = params.pop(metric_alias)
-            # # concatenate metric from params (or default if not provided in params) and eval_metric
-            # original_metric = [original_metric] if isinstance(original_metric, (string_type, type(None))) else original_metric
-            # eval_metric = [eval_metric] if isinstance(eval_metric, (string_type, type(None))) else eval_metric
-            # params['metric'] = set(original_metric + eval_metric)  # TODO: train_metric
 
         if not isinstance(X, DataFrame):
             _X, _y = _LGBMCheckXY(X, y, accept_sparse=True, force_all_finite=False, ensure_min_samples=2)

From b137ad5f630ff9f4530d8949fba299925d07cd63 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 13:01:02 +0900
Subject: [PATCH 17/19] update

---
 python-package/lightgbm/basic.py         | 1 -
 tests/python_package_test/test_engine.py | 6 ++++--
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/python-package/lightgbm/basic.py b/python-package/lightgbm/basic.py
index d6fd7edb172c..ccd31fb59996 100644
--- a/python-package/lightgbm/basic.py
+++ b/python-package/lightgbm/basic.py
@@ -1527,7 +1527,6 @@ def __init__(self, params=None, train_set=None, model_file=None, silent=False):
             # buffer for inner predict
             self.__inner_predict_buffer = [None]
             self.__is_predicted_cur_iter = [False]
-            self.__get_eval_info(0)  # init by train data (data_idx=0)
             self.pandas_categorical = train_set.pandas_categorical
         elif model_file is not None:
             # Prediction task
diff --git a/tests/python_package_test/test_engine.py b/tests/python_package_test/test_engine.py
index 3cfeb7054ff0..1725b1d72ef7 100644
--- a/tests/python_package_test/test_engine.py
+++ b/tests/python_package_test/test_engine.py
@@ -821,10 +821,12 @@ def test_train_and_valid_metric(self):
         lgb_eval = lgb.Dataset(X_test, y_test, reference=lgb_train)
         evals_result = {}
         gbm = lgb.train(params, lgb_train,
-                        num_boost_round=50,
-                        valid_sets=[lgb_train, lgb_eval],
+                        num_boost_round=10,
+                        valid_sets=[lgb_train, lgb_eval, lgb_eval],
                         verbose_eval=False,
                         evals_result=evals_result)
         self.assertTrue('l2' in evals_result['training'])
         self.assertTrue('l1' in evals_result['valid_1'])
         self.assertTrue('l2' in evals_result['valid_1'])
+        self.assertTrue('l1' in evals_result['valid_2'])
+        self.assertTrue('l2' in evals_result['valid_2'])

From 2fb6066cb3409f190dfa8ab3f20b9d554ee78267 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 13:10:48 +0900
Subject: [PATCH 18/19] backward compatibility of config

---
 docs/Parameters.rst       |  8 ++++----
 include/LightGBM/config.h | 13 +++++++------
 src/io/config_auto.cpp    |  4 +++-
 3 files changed, 14 insertions(+), 11 deletions(-)

diff --git a/docs/Parameters.rst b/docs/Parameters.rst
index e92fb2a2a5eb..e34b360dc4b2 100644
--- a/docs/Parameters.rst
+++ b/docs/Parameters.rst
@@ -739,7 +739,7 @@ Objective Parameters
 Metric Parameters
 -----------------
 
--  ``train_metric`` :raw-html:`<a id="train_metric" title="Permalink to this parameter" href="#train_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
+-  ``metric`` :raw-html:`<a id="metric" title="Permalink to this parameter" href="#metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum, aliases: ``metrics``, ``metric_types``
 
    -  metric(s) to be evaluated on the evaluation set(s)
 
@@ -791,13 +791,13 @@ Metric Parameters
 
    -  support multiple metrics, separated by ``,``
 
--  ``valid_metric`` :raw-html:`<a id="valid_metric" title="Permalink to this parameter" href="#valid_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
+-  ``train_metric`` :raw-html:`<a id="train_metric" title="Permalink to this parameter" href="#train_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
 
    -  metric(s) to be evaluated on the evaluation set(s)
 
--  ``metric`` :raw-html:`<a id="metric" title="Permalink to this parameter" href="#metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
+-  ``valid_metric`` :raw-html:`<a id="valid_metric" title="Permalink to this parameter" href="#valid_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
 
-   -  dummpy parameter
+   -  metric(s) to be evaluated on the evaluation set(s)
 
 -  ``metric_freq`` :raw-html:`<a id="metric_freq" title="Permalink to this parameter" href="#metric_freq">&#x1F517;&#xFE0E;</a>`, default = ``1``, type = int, aliases: ``output_freq``, constraints: ``metric_freq > 0``
 
diff --git a/include/LightGBM/config.h b/include/LightGBM/config.h
index 97777faca389..5f8303c8a7f4 100644
--- a/include/LightGBM/config.h
+++ b/include/LightGBM/config.h
@@ -669,6 +669,7 @@ struct Config {
   #pragma region Metric Parameters
 
   // [doc-only]
+  // alias = metrics, metric_types
   // default = ""
   // type = multi-enum
   // desc = metric(s) to be evaluated on the evaluation set(s)
@@ -696,19 +697,19 @@ struct Config {
   // descl2 = ``xentlambda``, "intensity-weighted" cross-entropy, aliases: ``cross_entropy_lambda``
   // descl2 = ``kldiv``, `Kullback-Leibler divergence <https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence>`__, aliases: ``kullback_leibler``
   // desc = support multiple metrics, separated by ``,``
-  std::vector<std::string> train_metric;
+  std::vector<std::string> metric;
 
   // [doc-only]
-  // type = multi-enum
   // default = ""
+  // type = multi-enum
   // desc = metric(s) to be evaluated on the evaluation set(s)
-  std::vector<std::string> valid_metric;
+  std::vector<std::string> train_metric;
 
   // [doc-only]
-  // type = multi-enum
   // default = ""
-  // desc = dummpy parameter
-  std::vector<std::string> metric;
+  // type = multi-enum
+  // desc = metric(s) to be evaluated on the evaluation set(s)
+  std::vector<std::string> valid_metric;
 
   // check = >0
   // alias = output_freq
diff --git a/src/io/config_auto.cpp b/src/io/config_auto.cpp
index e8feaa405abe..d2105c340f0d 100644
--- a/src/io/config_auto.cpp
+++ b/src/io/config_auto.cpp
@@ -135,6 +135,8 @@ std::unordered_map<std::string, std::string> Config::alias_table({
   {"num_classes", "num_class"},
   {"unbalance", "is_unbalance"},
   {"unbalanced_sets", "is_unbalance"},
+  {"metrics", "metric"},
+  {"metric_types", "metric"},
   {"output_freq", "metric_freq"},
   {"training_metric", "is_provide_training_metric"},
   {"is_training_metric", "is_provide_training_metric"},
@@ -246,9 +248,9 @@ std::unordered_set<std::string> Config::parameter_set({
   "tweedie_variance_power",
   "max_position",
   "label_gain",
+  "metric",
   "train_metric",
   "valid_metric",
-  "metric",
   "metric_freq",
   "is_provide_training_metric",
   "eval_at",

From b89c1ecb60f7a09fabcfa2ce5972251b2311cd02 Mon Sep 17 00:00:00 2001
From: Tsukasa OMOTO <henry0312@gmail.com>
Date: Thu, 20 Dec 2018 23:24:54 +0900
Subject: [PATCH 19/19] update documents

---
 docs/Parameters.rst                | 8 ++++++--
 include/LightGBM/config.h          | 6 ++++--
 python-package/lightgbm/sklearn.py | 7 +++++++
 src/io/config.cpp                  | 1 +
 4 files changed, 18 insertions(+), 4 deletions(-)

diff --git a/docs/Parameters.rst b/docs/Parameters.rst
index e34b360dc4b2..686bc16a960d 100644
--- a/docs/Parameters.rst
+++ b/docs/Parameters.rst
@@ -793,11 +793,15 @@ Metric Parameters
 
 -  ``train_metric`` :raw-html:`<a id="train_metric" title="Permalink to this parameter" href="#train_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
 
-   -  metric(s) to be evaluated on the evaluation set(s)
+   -  metric(s) to be evaluated on the train set passed into evaluation set
+
+   -  if ``metric`` is set, ``train_metric`` will be overwritten
 
 -  ``valid_metric`` :raw-html:`<a id="valid_metric" title="Permalink to this parameter" href="#valid_metric">&#x1F517;&#xFE0E;</a>`, default = ``""``, type = multi-enum
 
-   -  metric(s) to be evaluated on the evaluation set(s)
+   -  metric(s) to be evaluated on the valid set passed into evaluation set
+
+   -  if ``metric`` is set, ``valid_metric`` will be overwritten
 
 -  ``metric_freq`` :raw-html:`<a id="metric_freq" title="Permalink to this parameter" href="#metric_freq">&#x1F517;&#xFE0E;</a>`, default = ``1``, type = int, aliases: ``output_freq``, constraints: ``metric_freq > 0``
 
diff --git a/include/LightGBM/config.h b/include/LightGBM/config.h
index 5f8303c8a7f4..e6363352edf3 100644
--- a/include/LightGBM/config.h
+++ b/include/LightGBM/config.h
@@ -702,13 +702,15 @@ struct Config {
   // [doc-only]
   // default = ""
   // type = multi-enum
-  // desc = metric(s) to be evaluated on the evaluation set(s)
+  // desc = metric(s) to be evaluated on the train set passed into evaluation set
+  // desc = if ``metric`` is set, ``train_metric`` will be overwritten
   std::vector<std::string> train_metric;
 
   // [doc-only]
   // default = ""
   // type = multi-enum
-  // desc = metric(s) to be evaluated on the evaluation set(s)
+  // desc = metric(s) to be evaluated on the valid set passed into evaluation set
+  // desc = if ``metric`` is set, ``valid_metric`` will be overwritten
   std::vector<std::string> valid_metric;
 
   // check = >0
diff --git a/python-package/lightgbm/sklearn.py b/python-package/lightgbm/sklearn.py
index 6b4cca23736e..5f87ae8b7679 100644
--- a/python-package/lightgbm/sklearn.py
+++ b/python-package/lightgbm/sklearn.py
@@ -364,6 +364,12 @@ def fit(self, X, y,
             If callable, it should be a custom evaluation metric, see note below for more details.
             In either case, the ``metric`` from the model parameters will be evaluated and used as well.
             Default: 'l2' for LGBMRegressor, 'logloss' for LGBMClassifier, 'ndcg' for LGBMRanker.
+        eval_train_metric : string, list of strings, callable or None, optional (default=None)
+            metric(s) to be evaluated on the train set passed into ``eval_set``
+            if ``metric`` is set, ``eval_train_metric`` will be overwritten
+        eval_valid_metric : string, list of strings, callable or None, optional (default=None)
+            metric(s) to be evaluated on the valid set passed into ``eval_set``
+            if ``metric`` is set, ``eval_valid_metric`` will be overwritten
         early_stopping_rounds : int or None, optional (default=None)
             Activates early stopping. The model will train until the validation score stops improving.
             Validation score needs to improve at least every ``early_stopping_rounds`` round(s)
@@ -462,6 +468,7 @@ def fit(self, X, y,
         else:
             feval = None
             if eval_metric:
+                # for backward compatibility, `metric` is prefered to `train_metric` and `valid_metric`
                 params['metric'] = eval_metric
             else:
                 params['train_metric'] = eval_train_metric
diff --git a/src/io/config.cpp b/src/io/config.cpp
index 0da8ff313e5c..d18c559f7fca 100644
--- a/src/io/config.cpp
+++ b/src/io/config.cpp
@@ -108,6 +108,7 @@ void SetDefaultMetircType(const std::unordered_map<std::string, std::string>& pa
 void GetMetricType(const std::unordered_map<std::string, std::string>& params,
                    std::vector<std::string>* train_metric, std::vector<std::string>* valid_metric) {
   if (params.count("metric") > 0) {
+    // for backward compatibility, `metric` is prefered to `train_metric` and `valid_metric`
     SetMetircType(params, "metric", train_metric);
     SetMetircType(params, "metric", valid_metric);
   } else {