BUG: Raise early on non-finite values in PSD (Welch) and ICA.fit (Fix…

[zhijingz]

…es #13364)

Reference issue (if any)

Fixes #13364

What does this implement/fix?

psd_array_welch

• Added an explicit early check for Inf values inside the analyzed span.
• Replaced an internal assert with a user-facing ValueError when NaN masks are not aligned across channels.
• Retained existing intended behavior that allows “sentinel” NaNs — aligned across channels — to mark annotation boundaries without error.

ICA.fit

• Added an early ValueError guard when input data contain any NaN or Inf, producing a clear message instead of downstream runtime warnings.

Additional information

[welcome]

Hello! 👋 Thanks for opening your first pull request here! ❤️ We will try to get back to you soon. 🚴

[zhijingz]

Hi! First-time contributor here. I’ve rebased on current main, pre-commit passes locally, and targeted tests are green:

pytest -q mne/time_frequency/tests -k psd → 21 passed, 6 skipped
pytest -q mne/preprocessing/tests -k ica → 1 passed

Could someone please approve CI to run on this PR? Thanks!

[zhijingz]

Hi, I revised the contritbution doc - anything else I need to do for the workflow to be approved?

[drammock]

I made a few comments/suggestions about specific bits of the code/tests. Additional high-level comments:

#13364 (comment) suggested that we error for ICA.fit when there are NaNs present, but for PSD we should warn if NaNs are on specific channels and compute the PSD for the other channels that have good data. I see two new ValueErrors and no new warnings in the time_frequency code, so it seems like you're doing something different from what @larsoner recommended.

[zhijingz]

Hi thanks for the suggestions - I've included the suggestions into this updated version + psd part now warns on channels with nan instead of raising but still computes psd for valid channels. Let me know your feedback on these new changes

[larsoner]

Pushed a tiny commit to use mne.utils.warn instead of warnings.warn because our version does some stacklevel and logger.level-based magic. Then pushed a commit to get CircleCI to run all examples to make sure none break because of this PR 🤞

I'll watch it and push another commit to fix any broken examples so this can land in 1.11 which we want to release in the next day or so. Thanks in advance @zhijingz !

[welcome]

🎉 Congrats on merging your first pull request! 🥳 Looking forward to seeing more from you in the future! 💪