Fixes for multiple and default metric (#1239)

* fix multiple evaluation metrics

* create DefaultEvalMetric only when really necessary

* py test for #1239

* make travis happy

Author: khotilov

src/c_api/c_api.cc

@@ -33,7 +33,10 @@ class Booster {
 
   inline void SetParam(const std::string& name, const std::string& val) {
     auto it = std::find_if(cfg_.begin(), cfg_.end(),
-      [&name](decltype(*cfg_.begin()) &x) {
+      [&name, &val](decltype(*cfg_.begin()) &x) {
+        if (name == "eval_metric") {
+          return x.first == name && x.second == val;
+        }
         return x.first == name;
       });
     if (it == cfg_.end()) {

src/learner.cc

@@ -256,9 +256,6 @@ class LearnerImpl : public Learner {
       attributes_ = std::map<std::string, std::string>(
           attr.begin(), attr.end());
     }
-    if (metrics_.size() == 0) {
-      metrics_.emplace_back(Metric::Create(obj_->DefaultEvalMetric()));
-    }
     this->base_score_ = mparam.base_score;
     gbm_->ResetPredBuffer(pred_buffer_size_);
     cfg_["num_class"] = common::ToString(mparam.num_class);
@@ -307,6 +304,9 @@ class LearnerImpl : public Learner {
     std::ostringstream os;
     os << '[' << iter << ']'
        << std::setiosflags(std::ios::fixed);
+    if (metrics_.size() == 0) {
+      metrics_.emplace_back(Metric::Create(obj_->DefaultEvalMetric()));
+    }
     for (size_t i = 0; i < data_sets.size(); ++i) {
       this->PredictRaw(data_sets[i], &preds_);
       obj_->EvalTransform(&preds_);
@@ -445,9 +445,6 @@ class LearnerImpl : public Learner {
 
     // reset the base score
     mparam.base_score = obj_->ProbToMargin(mparam.base_score);
-    if (metrics_.size() == 0) {
-      metrics_.emplace_back(Metric::Create(obj_->DefaultEvalMetric()));
-    }
 
     this->base_score_ = mparam.base_score;
     gbm_->ResetPredBuffer(pred_buffer_size_);

tests/python/test_basic_models.py

@@ -105,6 +105,16 @@ def neg_evalerror(preds, dtrain):
                    if int(preds2[i] > 0.5) != labels[i]) / float(len(preds2))
         assert err == err2
 
+    def test_multi_eval_metric(self):
+        watchlist = [(dtest, 'eval'), (dtrain, 'train')]
+        param = {'max_depth': 2, 'eta': 0.2, 'silent': 1, 'objective': 'binary:logistic'}
+        param['eval_metric'] = ["auc", "logloss", 'error']
+        evals_result = {}
+        bst = xgb.train(param, dtrain, 4, watchlist, evals_result=evals_result)
+        assert isinstance(bst, xgb.core.Booster)
+        assert len(evals_result['eval']) == 3
+        assert set(evals_result['eval'].keys()) == {'auc', 'error', 'logloss'}
+
     def test_fpreproc(self):
         param = {'max_depth': 2, 'eta': 1, 'silent': 1,
                  'objective': 'binary:logistic'}