lint.

python-package/xgboost/testing/__init__.py

@@ -447,7 +447,6 @@ def make_position_biased_ltr(
     n_samples: int, n_features: int, n_query_groups: int, max_rel: int
 ) -> ClickFold:
     """Generate position-biased data for testing learning to rank."""
-    datasets = pytest.importorskip("sklearn.datasets")
     ms = pytest.importorskip("sklearn.model_selection")
     # from sklearn.model_selection import train_test_split
     X, y = datasets.make_classification(n_samples, n_features, n_classes=max_rel)

python-package/xgboost/testing/data.py

@@ -1,3 +1,4 @@
+# pylint: disable=invalid-name
 """Utilities for data generation."""
 import os
 import zipfile
@@ -349,15 +350,18 @@ def get_mq2008(
 
 
 class RelDataCV(NamedTuple):
+    """Simple data struct for train-test split."""
+
     train: RelData
     test: RelData
     max_rel: int
 
     def is_binary(self) -> bool:
+        """Whether it's binary relevance."""
         return self.max_rel == 1
 
 
-class PBM:
+class PBM:  # pylint: disable=too-few-public-methods
     """Simulate click data with position bias model. There are other models available in
     `ULTRA <https://github.com/ULTR-Community/ULTRA.git>`_ like the cascading model.
 
@@ -470,7 +474,6 @@ def simulate_one_fold(
         """Simulate clicks for one fold."""
         X_fold, y_fold, qid_fold = fold
         assert qid_fold.dtype == np.int32
-        indptr, lengths, values = rlencode(qid_fold)
 
         qids = np.unique(qid_fold)
 
@@ -554,7 +557,7 @@ def sort_ltr_samples(
     qid = qid[sorted_idx]
     pos = pos[sorted_idx]
 
-    indptr, lengths, values = rlencode(qid)
+    indptr, _, _ = rlencode(qid)
 
     for i in range(1, indptr.size):
         beg = indptr[i - 1]