[FIX] Making the MOABB dataset work with BIDS cache

[bruAristimunha]

Code to test:

from braindecode.datasets import MOABBDataset
from braindecode.preprocessing import create_windows_from_events
cache_config = dict(
    use=True,
    path="~/mne_data",
    save_raw=True,
    save_epochs=False,
    save_array=False,
    overwrite_raw=False,
    overwrite_epochs=False,
    overwrite_array=False,
)

dataset = MOABBDataset("AlexMI", subject_ids=[1], dataset_load_kwargs={'cache_config': cache_config})

windows = create_windows_from_events(dataset)

[codecov]

Codecov Report

Attention: Patch coverage is 50.00000% with 7 lines in your changes missing coverage. Please review.

Project coverage is 87.17%. Comparing base (f775002) to head (4463268).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #694      +/-   ##
==========================================
- Coverage   87.25%   87.17%   -0.09%     
==========================================
  Files          76       76              
  Lines        6835     6846      +11     
==========================================
+ Hits         5964     5968       +4     
- Misses        871      878       +7     

[PierreGtch]

Suggested change

	# find events from the stim channel
	stim_channels = mne.utils._get_stim_channel(None, raw.info, raise_error=False)
	if len(stim_channels) > 0:
	# returns an empty array if none found
	events = mne.find_events(raw, shortest_event=0, verbose=False)
	else:
	events, _ = mne.events_from_annotations(raw, verbose=False)
	events, _ = mne.events_from_annotations(raw, verbose=False)

I think we should always get the events using the annotations (with NeuroTechX/moabb#696, the annotations are always set in raw objects obtained from moabb)

[bruAristimunha]

For some strange reason, it does not happen in all the datasets. I can try to run without this and fix this on the moabb side.

[bruAristimunha]

Hey Pierre,

It is falling already with the fake dataset, as I need this to run some stuff. For me, we merge this PR and open one issue on moabb to solve this in the future. It is a problem on the moabb side.

[PierreGtch]

@bruAristimunha I'm not sure I understand why this PR is now needed; what changed on the MOABB/MNE side?
But let's merge if you think it is best.
We should take a deep look at versions compatibility with MOABB before making the next braindecode release

[bruAristimunha]

I will take some time to merge this pr, don't worry