Export JSON config in get_params. (#5256)

python-package/xgboost/sklearn.py

@@ -200,7 +200,7 @@ Parameters
 @xgboost_model_doc("""Implementation of the Scikit-Learn API for XGBoost.""",
                    ['estimators', 'model', 'objective'])
 class XGBModel(XGBModelBase):
-    # pylint: disable=too-many-arguments, too-many-instance-attributes, invalid-name, missing-docstring
+    # pylint: disable=too-many-arguments, too-many-instance-attributes, missing-docstring
     def __init__(self, max_depth=None, learning_rate=None, n_estimators=100,
                  verbosity=None, objective=None, booster=None,
                  tree_method=None, n_jobs=None, gamma=None,
@@ -210,7 +210,8 @@ class XGBModel(XGBModelBase):
                  scale_pos_weight=None, base_score=None, random_state=None,
                  missing=None, num_parallel_tree=None,
                  monotone_constraints=None, interaction_constraints=None,
-                 importance_type="gain", gpu_id=None, **kwargs):
+                 importance_type="gain", gpu_id=None,
+                 validate_parameters=False, **kwargs):
         if not SKLEARN_INSTALLED:
             raise XGBoostError(
                 'sklearn needs to be installed in order to use this module')
@@ -243,6 +244,10 @@ class XGBModel(XGBModelBase):
         self.interaction_constraints = interaction_constraints
         self.importance_type = importance_type
         self.gpu_id = gpu_id
+        # Parameter validation is not working with Scikit-Learn interface, as
+        # it passes all paraemters into XGBoost core, whether they are used or
+        # not.
+        self.validate_parameters = validate_parameters
 
     def __setstate__(self, state):
         # backward compatibility code
@@ -314,11 +319,35 @@ class XGBModel(XGBModelBase):
         if isinstance(params['random_state'], np.random.RandomState):
             params['random_state'] = params['random_state'].randint(
                 np.iinfo(np.int32).max)
-        # Parameter validation is not working with Scikit-Learn interface, as
+
-        # it passes all paraemters into XGBoost core, whether they are used or
+        def parse_parameter(value):
-        # not.
+            for t in (int, float):
-        if 'validate_parameters' not in params.keys():
+                try:
-            params['validate_parameters'] = False
+                    ret = t(value)
+                    return ret
+                except ValueError:
+                    continue
+            return None
+
+        # Get internal parameter values
+        try:
+            config = json.loads(self.get_booster().save_config())
+            stack = [config]
+            internal = {}
+            while stack:
+                obj = stack.pop()
+                for k, v in obj.items():
+                    if k.endswith('_param'):
+                        for p_k, p_v in v.items():
+                            internal[p_k] = p_v
+                    elif isinstance(v, dict):
+                        stack.append(v)
+
+            for k, v in internal.items():
+                if k in params.keys() and params[k] is None:
+                    params[k] = parse_parameter(v)
+        except XGBoostError:
+            pass
         return params
 
     def get_xgb_params(self):

src/learner.cc

@@ -269,10 +269,7 @@ class LearnerImpl : public Learner {
         }
       }
     }
-    auto learner_model_param = mparam_.ToJson();
+
-    for (auto const& kv : get<Object>(learner_model_param)) {
-      keys.emplace_back(kv.first);
-    }
     keys.emplace_back(kEvalMetric);
     keys.emplace_back("verbosity");
     keys.emplace_back("num_output_group");
@@ -425,6 +422,7 @@ class LearnerImpl : public Learner {
     auto& learner_parameters = out["learner"];
 
     learner_parameters["learner_train_param"] = toJson(tparam_);
+    learner_parameters["learner_model_param"] = mparam_.ToJson();
     learner_parameters["gradient_booster"] = Object();
     auto& gradient_booster = learner_parameters["gradient_booster"];
     gbm_->SaveConfig(&gradient_booster);

tests/python/test_with_sklearn.py

@@ -490,6 +490,13 @@ def test_kwargs():
     assert clf.get_params()['n_estimators'] == 1000
 
 
+def test_kwargs_error():
+    params = {'updater': 'grow_gpu_hist', 'subsample': .5, 'n_jobs': -1}
+    with pytest.raises(TypeError):
+        clf = xgb.XGBClassifier(n_jobs=1000, **params)
+        assert isinstance(clf, xgb.XGBClassifier)
+
+
 def test_kwargs_grid_search():
     from sklearn.model_selection import GridSearchCV
     from sklearn import datasets
@@ -510,13 +517,6 @@ def test_kwargs_grid_search():
     assert len(means) == len(set(means))
 
 
-def test_kwargs_error():
-    params = {'updater': 'grow_gpu_hist', 'subsample': .5, 'n_jobs': -1}
-    with pytest.raises(TypeError):
-        clf = xgb.XGBClassifier(n_jobs=1000, **params)
-        assert isinstance(clf, xgb.XGBClassifier)
-
-
 def test_sklearn_clone():
     from sklearn.base import clone
 
@@ -525,6 +525,17 @@ def test_sklearn_clone():
     clone(clf)
 
 
+def test_sklearn_get_default_params():
+    from sklearn.datasets import load_digits
+    digits_2class = load_digits(2)
+    X = digits_2class['data']
+    y = digits_2class['target']
+    cls = xgb.XGBClassifier()
+    assert cls.get_params()['base_score'] is None
+    cls.fit(X[:4, ...], y[:4, ...])
+    assert cls.get_params()['base_score'] is not None
+
+
 def test_validation_weights_xgbmodel():
     from sklearn.datasets import make_hastie_10_2