MAMEM dataset evaluation fails due to cv_split 5 in eval

[Div12345]

Within Session Evaluation fails while running for MAMEM dataset with the following error -

/usr/local/lib/python3.7/dist-packages/sklearn/model_selection/_split.py:672: UserWarning: The least populated class in y has only 4 members, which is less than n_splits=5.
% (min_groups, self.n_splits)), UserWarning)

The CV has been hard coded as 5 in the Evaluation -

moabb/moabb/evaluations/evaluations.py

Line 114 in aa4274f

cv = StratifiedKFold(5, shuffle=True, random_state=self.random_state)

[Div12345]

As per discussion with @sylvchev, a possible cause of this might be from the Rest condition being included in the data. Need to check the data in further detail.

[sylvchev]

It seems that this is not the reason: there is a resting period but is it is not extracted in the actual dataset class. The only classes available are the flash frequencies : 6.66, 7.5, 8.57, 10 and 12 Hz.
For MAMEM1, there are between 12 and 20 trials per class, for MAMEM2 and MAMEM3, between 20 to 30 trials per class. I'm re-checking the trials ID, there are different (and conflicting) indications in the documentation on the website, in the data description (PDF) and in the matlab codebase.

[sylvchev]

After some investigation, the problem is an incorrect description: MAMEM sessions are instead 'runs'. The errors come from the low number of stimulus in each runs. PR #256 correct this issue.