NeurIPS 2021 BEETL Competition - Benchmarks for EEG Transfer Learning Forum

Go back to competition Back to thread list Post in this thread

> Questions about the Motor Imagery datasets

Hi Beetl team!

We have a question about the motor imagery task. In the starting kit, 3.motor_imagery_kit, a function called MotorImagery is used to preprocess the EEG signals (sampling rate, high pass filter, low pass filter, keeping the same electrodes in all datasets…).

If we understand correctly, this function has been written to be applied to MOABB datasets. So, it works well for the source datasets. My question is the following: can we easily use this function on the leaderboard dataset (assuming for instance, that we can easily transform it into a dataset with the same characteristics as a MOABB dataset) or do we need to recode it if we want to use a similar function?

If we need to recode it, we have another question about the leaderboard dataset. The names of the electrodes are given in the comments. For instance, for dataset B, the names are 'Fp1', 'Fp2', 'F3', 'Fz', 'F4', 'FC5', 'FC1', 'FC2','FC6', 'C5', 'C3', 'C1'... Do these names correspond to the data channels in the same order? For instance, can we access the electrode ‘Fp1’ of the test examples stored in X_MIB_test (as in 1.LeaderboardDataGuide) by writing X_MIB_test[:, 0, :]?

Thank you in advance for your answer.

Posted by: mbonto @ Aug. 7, 2021, 10:50 a.m.

Thanks for the question!
For the function, you could have a look at the code (the files with that function download in your computer in MOABB API) or document ( of MOABB for details. It's designed for MOABB datasets while not directly applied to the competition data. While you could think about reusing some of the code in that function if would like to. For the second question, yes, it's in the correct order.

Xiaoxi Wei

Posted by: BEETLCompetition @ Aug. 9, 2021, 7:47 a.m.

Ok, I understand. Thank you for your quick answer!

Posted by: mbonto @ Aug. 10, 2021, 11:43 a.m.
Post in this thread