Bug: FactorDatetimeDailyEvaluator classifies hourly and other sub-daily factor data as daily

Summary

FactorDatetimeDailyEvaluator.evaluate reports sub-daily factor output as daily unless the adjacent datetime difference is exactly one minute.

For example, factor data indexed hourly, every 30 minutes, or every second is currently accepted:

("The generated dataframe is daily.", True)

These are not daily observations. The evaluator should reject sub-daily datetime frequencies instead of only rejecting the special one-minute case.

To Reproduce

1. Check out RD-Agent main at commit 6762f84f9bc0f5c6486c50a00e128a57ac6c3683.

2. Install RD-Agent from source.

3. Create test/qlib/test_factor_datetime_daily_evaluator.py:

import pandas as pd
import pytest

from rdagent.components.coder.factor_coder.eva_utils import (
    FactorDatetimeDailyEvaluator,
)


@pytest.mark.parametrize("freq", ["h", "30min", "s"])
def test_datetime_daily_evaluator_rejects_sub_daily_frequency(freq):
    datetimes = pd.date_range("2024-01-01 00:00:00", periods=3, freq=freq)
    index = pd.MultiIndex.from_product(
        [datetimes, ["instrument_1"]],
        names=["datetime", "instrument"],
    )
    gen_df = pd.DataFrame({"factor": [1.0, 2.0, 3.0]}, index=index)

    evaluator = object.__new__(FactorDatetimeDailyEvaluator)
    evaluator._get_df = lambda _gt_implementation, _implementation: (None, gen_df)

    message, is_daily = evaluator.evaluate(None, None)

    assert is_daily is False, message

1. Run:

python -m pytest test/qlib/test_factor_datetime_daily_evaluator.py -q

1. Observe that all three parameterized cases fail.

Expected Behavior

Hourly, 30-minute, and one-second datetime intervals should be classified as not daily:

is_daily is False

Actual Behavior

Each sub-daily input is classified as daily:

h     ('The generated dataframe is daily.', True)
30min ('The generated dataframe is daily.', True)
s     ('The generated dataframe is daily.', True)

A representative assertion failure is:

AssertionError: The generated dataframe is daily.
assert True is False

Screenshot

Not applicable; this is a deterministic unit-level reproduction.

Environment

• Name of current operating system: macOS
• Processor architecture: arm64
• Python version: 3.11.15
• RD-Agent version: 0.8.0, main@6762f84f9bc0f5c6486c50a00e128a57ac6c3683
• Package version: pandas 2.3.3, pytest 9.1.1

Additional Notes

The root cause appears to be that the evaluator only checks for one exact sub-daily interval:

time_diff = pd.to_datetime(
    gen_df.index.get_level_values("datetime")
).to_series().diff().dropna().unique()

if pd.Timedelta(minutes=1) in time_diff:
    return (
        "The generated dataframe is not daily. The implementation is definitely wrong. Please check the implementation.",
        False,
    )

return "The generated dataframe is daily.", True

Any other sub-daily interval, such as hourly or 30-minute data, reaches the final positive result.

A possible fix is to reject datetime differences that are shorter than one day, while still allowing valid daily data with calendar gaps such as weekends or holidays.