fix: record_temp datetime extraction with ambiguous level names

qlib/workflow/record_temp.py

@@ -471,7 +471,8 @@ def _generate(self, **kwargs):
             setattr(self, k, fill_placeholder(getattr(self, k), placeholder_value))
 
         # if the backtesting time range is not set, it will automatically extract time range from the prediction file
-        dt_values = pred.index.get_level_values("datetime")
+        dt_level = pred.index.names.index("datetime")
+        dt_values = pred.index.get_level_values(dt_level)
         if self.backtest_config["start_time"] is None:
             self.backtest_config["start_time"] = dt_values.min()
         if self.backtest_config["end_time"] is None:
@@ -617,7 +618,8 @@ def __init__(self, recorder, pass_num=10, shuffle_init_score=True, **kwargs):
     def random_init(self):
         pred_df = self.load("pred.pkl")
 
-        all_pred_dates = pred_df.index.get_level_values("datetime")
+        dt_level = pred_df.index.names.index("datetime")
+        all_pred_dates = pred_df.index.get_level_values(dt_level)
         bt_start_date = pd.to_datetime(self.backtest_config.get("start_time"))
         if bt_start_date is None:
             first_bt_pred_date = all_pred_dates.min()

tests/test_record_temp_datetime.py

@@ -0,0 +1,56 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+"""Tests for record_temp datetime level fix (Fixes #1909).
+
+The fix replaces hard-coded get_level_values("datetime") with a
+positional lookup via index.names.index("datetime"), so it works
+regardless of the position of the datetime level in the MultiIndex.
+"""
+
+import numpy as np
+import pandas as pd
+import pytest
+
+
+def _make_multiindex(names, n_dates=3, n_instruments=2):
+    """Helper to build a MultiIndex with the given level names."""
+    dates = pd.date_range("2023-01-01", periods=n_dates, freq="D")
+    instruments = [f"STOCK_{chr(65 + i)}" for i in range(n_instruments)]
+    arrays = {
+        "datetime": np.repeat(dates, n_instruments),
+        "instrument": np.tile(instruments, n_dates),
+    }
+    # Build in the order specified by `names`
+    return pd.MultiIndex.from_arrays([arrays[n] for n in names], names=names)
+
+
+def test_datetime_first_level():
+    """Standard order: (datetime, instrument)."""
+    idx = _make_multiindex(["datetime", "instrument"])
+    dt_level = idx.names.index("datetime")
+    dt_values = idx.get_level_values(dt_level)
+    assert len(dt_values) == 6
+    assert dt_values[0] == pd.Timestamp("2023-01-01")
+
+
+def test_datetime_second_level():
+    """Reversed order: (instrument, datetime)."""
+    idx = _make_multiindex(["instrument", "datetime"])
+    dt_level = idx.names.index("datetime")
+    dt_values = idx.get_level_values(dt_level)
+    assert len(dt_values) == 6
+    assert dt_values[0] == pd.Timestamp("2023-01-01")
+
+
+def test_missing_datetime_raises():
+    """If datetime level is absent, index() should raise ValueError."""
+    idx = pd.MultiIndex.from_tuples(
+        [("A", 1), ("B", 2)], names=["instrument", "id"]
+    )
+    with pytest.raises(ValueError):
+        idx.names.index("datetime")
+
+
+if __name__ == "__main__":
+    pytest.main([__file__, "-v"])