[dask] Add DaskXGBRanker (#6576)

* Initial support for distributed LTR using dask.

* Support `qid` in libxgboost.
* Refactor `predict` and `n_features_in_`, `best_[score/iteration/ntree_limit]`
  to avoid duplicated code.
* Define `DaskXGBRanker`.

The dask ranker doesn't support group structure, instead it uses query id and
convert to group ptr internally.

include/xgboost/learner.h

@@ -165,7 +165,7 @@ class Learner : public Model, public Configurable, public dmlc::Serializable {
    * \brief Get the number of features of the booster.
    * \return number of features
    */
-  virtual uint32_t GetNumFeature() = 0;
+  virtual uint32_t GetNumFeature() const = 0;
 
   /*!
    * \brief Set additional attribute to the Booster.

python-package/xgboost/core.py

@@ -321,6 +321,7 @@ def next_wrapper(self, this):  # pylint: disable=unused-argument
 
         def data_handle(data, label=None, weight=None, base_margin=None,
                         group=None,
+                        qid=None,
                         label_lower_bound=None, label_upper_bound=None,
                         feature_names=None, feature_types=None,
                         feature_weights=None):
@@ -333,6 +334,7 @@ def data_handle(data, label=None, weight=None, base_margin=None,
             self.proxy.set_info(label=label, weight=weight,
                                 base_margin=base_margin,
                                 group=group,
+                                qid=qid,
                                 label_lower_bound=label_lower_bound,
                                 label_upper_bound=label_upper_bound,
                                 feature_names=feature_names,
@@ -523,12 +525,14 @@ def __del__(self):
     def set_info(self, *,
                  label=None, weight=None, base_margin=None,
                  group=None,
+                 qid=None,
                  label_lower_bound=None,
                  label_upper_bound=None,
                  feature_names=None,
                  feature_types=None,
                  feature_weights=None):
         '''Set meta info for DMatrix.'''
+        from .data import dispatch_meta_backend
         if label is not None:
             self.set_label(label)
         if weight is not None:
@@ -537,6 +541,8 @@ def set_info(self, *,
             self.set_base_margin(base_margin)
         if group is not None:
             self.set_group(group)
+        if qid is not None:
+            dispatch_meta_backend(matrix=self, data=qid, name='qid')
         if label_lower_bound is not None:
             self.set_float_info('label_lower_bound', label_lower_bound)
         if label_upper_bound is not None:
@@ -546,7 +552,6 @@ def set_info(self, *,
         if feature_types is not None:
             self.feature_types = feature_types
         if feature_weights is not None:
-            from .data import dispatch_meta_backend
             dispatch_meta_backend(matrix=self, data=feature_weights,
                                   name='feature_weights')
 
@@ -993,7 +998,7 @@ def _set_data_from_cuda_columnar(self, data):
 
 
 Objective = Callable[[np.ndarray, DMatrix], Tuple[np.ndarray, np.ndarray]]
-Metric = Callable[[np.ndarray, DMatrix], Tuple[np.ndarray, np.ndarray]]
+Metric = Callable[[np.ndarray, DMatrix], Tuple[str, float]]
 
 
 class Booster(object):
@@ -1743,10 +1748,16 @@ def num_boosted_rounds(self) -> int:
         '''
         rounds = ctypes.c_int()
         assert self.handle is not None
-        _check_call(_LIB.XGBoosterBoostedRounds(
-            self.handle, ctypes.byref(rounds)))
+        _check_call(_LIB.XGBoosterBoostedRounds(self.handle, ctypes.byref(rounds)))
         return rounds.value
 
+    def num_features(self) -> int:
+        '''Number of features in booster.'''
+        features = ctypes.c_int()
+        assert self.handle is not None
+        _check_call(_LIB.XGBoosterGetNumFeature(self.handle, ctypes.byref(features)))
+        return features.value
+
     def dump_model(self, fout, fmap='', with_stats=False, dump_format="text"):
         """Dump model into a text or JSON file.  Unlike `save_model`, the
         output format is primarily used for visualization or interpretation,