Zapline: retain copy of input in case no artifact present

[Randomidous]

If the average residual power at the line frequency after polynomial fitting (mean_score) is negative on the first iteration, the function will return data that has already been modified by dss_line(), even though the stopping criterion suggests the artifact wasn't actually present or the denoising made things worse.

I suggest retaining a copy of the input, which is returned if the mean_score is negative on iteration 0.

[nbara]

Hi @Randomidous, thanks for the suggestion and the PR, will look into it!

Do you think you could come up with a minimal example to reproduce the issue? I could integrate it in the tests

[Randomidous]

Hey there!

I have added a test in the PR :)

Looks like this:

def test_dss_line_iter_no_noise():
    """
    Test that dss_line_iter returns original data unchanged when DSS
    cannot improve the signal.
    """
    sr = 200
    fline = 50
    n_samples = 9000
    n_chans = 10
    rng = np.random.RandomState(42)

    # create data without line noise at target frequency
    x = rng.randn(n_samples, n_chans)

    x_original = x.copy()

    x_out, n_iters = dss.dss_line_iter(x, fline, sr, n_iter_max=10)

    assert n_iters == 0, f"Expected 0 iterations (no improvement), got {n_iters}"
    assert np.allclose(x_out, x_original), (
        "When DSS cannot improve signal, should return original data unchanged"
    )
    assert np.allclose(x, x_original), "Input data should never be mutated"

Fails on master and passes with added changes