fix(ops): make IdxMax and IdxMin skip NaN inside the window

[Ludwig J. Marx (LudwigJMarx)]

Description

IdxMax and IdxMin used ndarray.argmax / ndarray.argmin, which return
the position of a NaN rather than skipping it. Replaced with np.nanargmax /
np.nanargmin, four lines.

Motivation and Context

Refs #2352.

Max and Min skip NaN through pandas, IdxMax and IdxMin did not, so the
two disagreed on the same series. Mad and Rank in the same file already
filter NaN explicitly.

Alpha158 builds IMAX{d}, IMIN{d} and IMXD{d} on these operators over
$high and $low, which are NaN while an instrument is suspended.

No guard against ValueError: All-NaN slice encountered is needed, and that is
measured rather than assumed: with min_periods=1 pandas does not call apply
at all for a window that holds only NaN. Checked for both the rolling and the
expanding path. The min_periods=1 sits on the same line as the apply
call, so the assumption is local.

Rejected: filtering NaN out first, the way Mad and Rank do. The returned
position would then be the one inside the filtered array rather than inside the
window, which is exactly the quantity the field means (days since the high).

How Has This Been Tested?

• Pass the test by running: pytest qlib/tests/test_all_pipeline.py under upper directory of qlib.
• If you are adding a new feature, test on your own test scripts.

tests/test_all_pipeline.py is the one that matters here: it runs
DATASET_ALPHA158_CLASS over csi300 from 2008 to 2020, so it exercises the
fields this change touches, and test_1_backtest checks a real threshold.
It carries pytest.mark.slow, so CI skips it. Ran both sides:

	result
before the fix, on 7bd57c18	3 passed in 357 s
after the fix, on 4f5941e0	3 passed in 258 s

The runtimes differ because GBDT_MODEL sets no seed and uses
num_threads: 20. The test checks a threshold, not equality.

New file tests/ops/test_rolling_operator.py. It asserts no invented number
but the invariant against the real sibling operators: IdxMax has to point at
the value Max reports. The two commits carry the evidence: the test is red on
7bd57c18 and green on 4f5941e0.

Before the fix:

FAILED ops/test_rolling_operator.py::TestIdxRollingOperator::test_idxmax_points_at_the_maximum
FAILED ops/test_rolling_operator.py::TestIdxRollingOperator::test_idxmin_points_at_the_minimum
E   AssertionError: np.float64(nan) != np.float64(5.0) : IdxMax points at nan but Max reports 5.0
E   AssertionError: np.float64(nan) != np.float64(1.0) : IdxMin points at nan but Min reports 1.0
2 failed in 2.07s

After the fix:

2 passed in 2.21s

Rest of the suite, same command as CI (cd tests && pytest . -m "not slow"):

	result
before, untouched tree	71 passed, 1 skipped, 10 deselected, 79 subtests, 818 s
after	73 passed, 1 skipped, 10 deselected, 79 subtests, 718 s

The two extra are the new tests.

Counter-check that nothing else moves: 200000 random windows without NaN, ties
forced, argmax against nanargmax and argmin against nanargmin, zero
differences.

black -l 120 --check clean, flake8 clean, pylint 10.00/10 on
qlib/data/ops.py.

Environment: Python 3.12.13, numpy 2.5.3, pandas 2.3.3, MacOS 26.6.2 arm64,
commit be72549.

Types of changes

• Fix bugs
• Add new feature
• Update documentation

[Ludwig J. Marx (LudwigJMarx)]

@microsoft-github-policy-service agree