Braindecode and Moabb dataset

[bruAristimunha]

The following datasets are not working with the braindecode loader:

• <class 'moabb.datasets.epfl.EPFLP300'>
• <class 'moabb.datasets.hinss2021.Hinss2021'>
• <class 'moabb.datasets.huebner_llp.Huebner2017'>
• <class 'moabb.datasets.huebner_llp.Huebner2018'>
• <class 'moabb.datasets.schirrmeister2017.Schirrmeister2017'>

[bruAristimunha]

• <class 'moabb.datasets.epfl.EPFLP300'>
• <class 'moabb.datasets.mpi_mi.GrosseWentrup2009'>
• <class 'moabb.datasets.hinss2021.Hinss2021'>
• <class 'moabb.datasets.huebner_llp.Huebner2017'>
• <class 'moabb.datasets.huebner_llp.Huebner2018'>
• <class 'moabb.datasets.liu2024.Liu2024'>
• <class 'moabb.datasets.mpi_mi.MunichMI'>
• <class 'moabb.datasets.upper_limb.Ofner2017'>
• <class 'moabb.datasets.physionet_mi.PhysionetMI'>
• <class 'moabb.datasets.schirrmeister2017.Schirrmeister2017'>
• <class 'moabb.datasets.sosulski2019.Sosulski2019'>
• <class 'moabb.datasets.Zhou2016.Zhou2016'>

[bruAristimunha]

Less that I expected :)

[bruAristimunha]

The issue with EPFLP300 is the irregular size of events in the raw data in the last events. Initially, I thought it was the case where drop_bad would solve it, but no. In some sessions, we have different values.

I tried to solve it on the braindecode side for half an hour, but I couldn't. I'll come back to this later, but on the moabb side.

Same issue with Hinss2021:

assert np.all(this_trial_sizes == window_size_samples), (

AssertionError: All trial sizes should be the same if you do not supply a window size.

[bruAristimunha]

Huebner2017 has an issue too.

[PierreGtch]

@bruAristimunha

Couls this be related to braindecode/braindecode#563 and #607 ?

Now the annotations should always be set on raw moabb datasets thanks to the SetAnnotations step:

moabb/moabb/datasets/base.py

Lines 357 to 360 in d2edb29

	SetRawAnnotations(
	self.event_id,
	interval=self.interval,
	),

But if the raw already has annotations, this step is skipped:

moabb/moabb/datasets/preprocessing.py

Lines 69 to 70 in d2edb29

	if raw.annotations:
	return raw

I think we should overwrite the annotations' durations, even if they are already set.