Disable per-group weight.

python-package/xgboost/dask.py

@@ -208,6 +208,8 @@ class DaskDMatrix:
         Weight for each instance.
     base_margin :
         Global bias for each instance.
+    qid :
+        Query ID for ranking.
     label_lower_bound :
         Upper bound for survival training.
     label_upper_bound :
@@ -245,6 +247,9 @@ def __init__(
         self.feature_types = feature_types
         self.missing = missing
 
+        if qid is not None and weight is not None:
+            raise NotImplementedError('per-group weight is not implemented.')
+
         if len(data.shape) != 2:
             raise ValueError(
                 'Expecting 2 dimensional input, got: {shape}'.format(
@@ -1657,8 +1662,10 @@ def fit(  # pylint: disable=arguments-differ
     ) -> "DaskXGBRanker":
         _assert_dask_support()
         msg = "Use `qid` instead of `group` on dask interface."
-        assert group is None and eval_group is None, msg
-        assert qid is not None, "`qid` is required for ranking."
+        if not (group is None and eval_group is None):
+            raise ValueError(msg)
+        if not (qid is not None):
+            raise ValueError("`qid` is required for ranking.")
         return self.client.sync(
             self._fit_async,
             X=X,

python-package/xgboost/sklearn.py

@@ -1352,10 +1352,10 @@ def fit(
 
             .. note:: Weights are per-group for ranking tasks
 
-                In ranking task, one weight is assigned to each query group
-                (not each data point). This is because we only care about the
-                relative ordering of data points within each group, so it
-                doesn't make sense to assign weights to individual data points.
+                In ranking task, one weight is assigned to each query group/id (not each
+                data point). This is because we only care about the relative ordering of
+                data points within each group, so it doesn't make sense to assign weights
+                to individual data points.
 
         base_margin : array_like
             Global bias for each instance.