Support sample weight in sklearn custom objective. (#10050)

tests/ci_build/lint_python.py

@@ -100,6 +100,7 @@ class LintersPaths:
         # demo
         "demo/json-model/json_parser.py",
         "demo/guide-python/external_memory.py",
+        "demo/guide-python/sklearn_examples.py",
         "demo/guide-python/continuation.py",
         "demo/guide-python/callbacks.py",
         "demo/guide-python/cat_in_the_dat.py",

tests/python/test_with_sklearn.py

@@ -517,6 +517,12 @@ def test_regression_with_custom_objective():
         labels = y[test_index]
     assert mean_squared_error(preds, labels) < 25
 
+    w = rng.uniform(low=0.0, high=1.0, size=X.shape[0])
+    reg = xgb.XGBRegressor(objective=tm.ls_obj, n_estimators=25)
+    reg.fit(X, y, sample_weight=w)
+    y_pred = reg.predict(X)
+    assert mean_squared_error(y_true=y, y_pred=y_pred, sample_weight=w) < 25
+
     # Test that the custom objective function is actually used
     class XGBCustomObjectiveException(Exception):
         pass

tests/test_distributed/test_with_dask/test_with_dask.py

@@ -1750,9 +1750,20 @@ class TestWithDask:
             )
             tm.non_increasing(results_native["validation_0"]["rmse"])
 
+            reg = xgb.dask.DaskXGBRegressor(
+                n_estimators=rounds, objective=tm.ls_obj, tree_method="hist"
+            )
+            rng = da.random.RandomState(1994)
+            w = rng.uniform(low=0.0, high=1.0, size=y.shape[0])
+            reg.fit(
+                X, y, sample_weight=w, eval_set=[(X, y)], sample_weight_eval_set=[w]
+            )
+            results_custom = reg.evals_result()
+            tm.non_increasing(results_custom["validation_0"]["rmse"])
+
     def test_no_duplicated_partition(self) -> None:
-        """Assert each worker has the correct amount of data, and DMatrix initialization doesn't
-        generate unnecessary copies of data.
+        """Assert each worker has the correct amount of data, and DMatrix initialization
+        doesn't generate unnecessary copies of data.
 
         """
         with LocalCluster(n_workers=2, dashboard_address=":0") as cluster: