read_raw_bids() messes up the channel order and therefore loading raw fails

[moritz-gerster]

Description of the problem

I can load my data using mne.read_raw() but not using mne_bids.read_raw_bids(). I get the following error:

ValueError: New channel names are not unique, renaming failed

My electrodes.tsv file seems correct. When inspecting the code, it turns out that the order of ch_names_tsv and raw.ch_names is messed up. Not sure why it only happens for one specific file.

If I change the lists using sorted(), it works fine. Should the code possibly be changed?

        for bids_ch_name, raw_ch_name in zip(sorted(ch_names_tsv),
                                             sorted(raw.ch_names.copy())):
            if bids_ch_name != raw_ch_name:
                raw.rename_channels({raw_ch_name: bids_ch_name})

Maybe it's the same issue as #1117 (I didn't look into the details)?

Steps to reproduce

I can't share a replication since the patient data (even the metadata) are highly confidential.

Expected results

I would expect mne_bids.read_raw_bids() to load the data.

Actual results

Traceback (most recent call last):
  File "/Users/moritzgerster/Library/CloudStorage/Dropbox/Code/BIDS_LocalGlobal/scripts/_00_bidsify_sourcedata.py", line 600, in <module>
    bidsify_sourcedata()
  File "/Users/moritzgerster/Library/CloudStorage/Dropbox/Code/BIDS_LocalGlobal/scripts/_00_bidsify_sourcedata.py", line 28, in bidsify_sourcedata
    bidsify_sourcedata_neumann()
  File "/Users/moritzgerster/Library/CloudStorage/Dropbox/Code/BIDS_LocalGlobal/scripts/_00_bidsify_sourcedata.py", line 130, in bidsify_sourcedata_neumann
    raw = read_raw_bids(bids_path)
          ^^^^^^^^^^^^^^^^^^^^^^^^
  File "<decorator-gen-634>", line 12, in read_raw_bids
  File "/Users/moritzgerster/anaconda3/envs/localglobal/lib/python3.11/site-packages/mne_bids/read.py", line 780, in read_raw_bids
    raw = _handle_channels_reading(channels_fname, raw)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/moritzgerster/anaconda3/envs/localglobal/lib/python3.11/site-packages/mne_bids/read.py", line 606, in _handle_channels_reading
    raw.rename_channels({raw_ch_name: bids_ch_name})
  File "<decorator-gen-48>", line 12, in rename_channels
  File "/Users/moritzgerster/anaconda3/envs/localglobal/lib/python3.11/site-packages/mne/channels/channels.py", line 502, in rename_channels
    rename_channels(self.info, mapping, allow_duplicates)
  File "<decorator-gen-58>", line 12, in rename_channels
  File "/Users/moritzgerster/anaconda3/envs/localglobal/lib/python3.11/site-packages/mne/channels/channels.py", line 1290, in rename_channels
    raise ValueError("New channel names are not unique, renaming failed")
ValueError: New channel names are not unique, renaming failed

Additional information

Platform             macOS-13.5.2-arm64-arm-64bit
Python               3.11.3 | packaged by conda-forge | (main, Apr  6 2023, 08:58:31) [Clang 14.0.6 ]
Executable           /Users/moritzgerster/anaconda3/envs/localglobal/bin/python
CPU                  arm (10 cores)
Memory               16.0 GB

Core
├☑ mne               1.4.0
├☑ numpy             1.23.4 (OpenBLAS 0.3.20 with 10 threads)
├☑ scipy             1.10.1
├☑ matplotlib        3.7.1 (backend=MacOSX)
├☑ pooch             1.7.0
└☑ jinja2            3.1.2

Numerical (optional)
├☑ sklearn           1.2.2
├☑ nibabel           5.0.1
├☑ pandas            2.0.1
└☐ unavailable       numba, nilearn, dipy, openmeeg, cupy

[hoechenberger]

The BIDS names and channel order should take precedence, so renaming channels is the correct move. Channel orders in the sidecars and raw file also are allowed to differ IIRC ("should" be the same).

Maybe we should adopt a 2-step procedure to avoid name collisions:

• Assign new unique names to all channels in raw. I'm thinking of something like a hash or random string. It must not be present in raw or the BIDS sidecars.
• Only now rename the channels according to BIDS.

This should prevent the problem you're describing here.

[hoechenberger]

Just to clarify: this issue should only come up if the channel order in channels.tsv differs from the channel order in the raw file.

[larsoner]

I don't understand why any random names should be needed. Why not construct the full mapping from old names to new names and then call raw.rename_channels, then once that's done call raw.reorder_channels using the new names? To me it's a bit weird that rename_channels is called inside a for loop instead of constructing the mapping dict and calling rename_channels once.

[hoechenberger]

Ok sounds good

I don't see why we would need to reorder though?

[larsoner]

I don't see why we would need to reorder though?

I don't fully understand the use case or what MNE-BIDS does in all cases, I just saw "reorder" in the title and "channel order" in your comment so figured it was relevant / I should include for completeness

[hoechenberger]

Ah ok! :)

MNE-BIDS is only supposed to apply the channel names in the order they appear in channels.tsv. There's no explicit "reorder" operation per se (even though it could be implemented by swapping the names of two channels)