fix: reuse filtered instruments across field groups

qlib/data/dataset/loader.py

@@ -199,6 +199,11 @@ def __init__(
                     self.inst_processors
                 ), f"freq(={self.freq}), inst_processors(={self.inst_processors}) cannot be None/empty"
 
+    def load(self, instruments=None, start_time=None, end_time=None) -> pd.DataFrame:
+        if self.is_group and isinstance(instruments, str):
+            instruments = D.instruments(instruments, filter_pipe=self.filter_pipe)
+        return super().load(instruments, start_time, end_time)
+
     def load_group_df(
         self,
         instruments,
@@ -213,7 +218,7 @@ def load_group_df(
             instruments = "all"
         if isinstance(instruments, str):
             instruments = D.instruments(instruments, filter_pipe=self.filter_pipe)
-        elif self.filter_pipe is not None:
+        elif self.filter_pipe is not None and not isinstance(instruments, dict):
             warnings.warn("`filter_pipe` is not None, but it will not be used with `instruments` as list")
 
         freq = self.freq[gp_name] if isinstance(self.freq, dict) else self.freq

tests/data_mid_layer_tests/test_dataloader.py

@@ -4,7 +4,9 @@
 import sys
 import unittest
 import qlib
+import pandas as pd
 from pathlib import Path
+from unittest.mock import patch
 
 sys.path.append(str(Path(__file__).resolve().parent))
 from qlib.data.dataset.loader import NestedDataLoader, QlibDataLoader
@@ -84,6 +86,38 @@ def test_nested_data_loader(self):
         print(data)
         """
 
+    def test_grouped_loader_applies_filter_pipe_once(self):
+        loader = QlibDataLoader(
+            config={
+                "feature": (["$close"], ["close"]),
+                "label": (["Ref($close, -1)"], ["label"]),
+            },
+            filter_pipe=[{"filter_type": "SeriesDFilter"}],
+        )
+        instruments_config = {"market": "csi300", "filter_pipe": loader.filter_pipe}
+        index = pd.MultiIndex.from_tuples(
+            [("SH600000", pd.Timestamp("2020-01-01"))],
+            names=["instrument", "datetime"],
+        )
+
+        def fake_features(instruments, exprs, start_time, end_time, freq="day", inst_processors=None):
+            self.assertIs(instruments, instruments_config)
+            return pd.DataFrame([[1.0] * len(exprs)], index=index, columns=exprs)
+
+        with (
+            patch(
+                "qlib.data.dataset.loader.D.instruments",
+                return_value=instruments_config,
+                create=True,
+            ) as instruments_mock,
+            patch("qlib.data.dataset.loader.D.features", side_effect=fake_features, create=True) as features_mock,
+        ):
+            df = loader.load("csi300", start_time="2020-01-01", end_time="2020-01-02")
+
+        instruments_mock.assert_called_once_with("csi300", filter_pipe=loader.filter_pipe)
+        self.assertEqual(features_mock.call_count, 2)
+        self.assertEqual({"feature", "label"}, set(df.columns.get_level_values(0)))
+
 
 if __name__ == "__main__":
     unittest.main()