This is the code of an accepted conference paper submitted to EUSIPCO 2018. The preprint is available on this arXiv link. If you are using this code please cite our paper.

First, download the source code. Then, download the dataset "Four class motor imagery (001-2014)" of the BCI competition IV-2a. Put all files of the dataset (A01T.mat-A09E.mat) into a subfolder within the project called 'dataset' or change self.data_path in main_csp and main_riemannian.

• python3
• numpy
• sklearn
• pyriemann
• scipy

The packages can be installed easily with conda and the _config.yml file:

$ conda env create -f _config.yml -n msenv
$ source activate msenv 

For the recreation of the CSP results run main_csp.py. Change self.svm_kernel for testing different kernels:

• self.svm_kernel='linear' -> self.svm_c = 0.05
• self.svm_kernel='rbf' -> self.svm_c = 20
• self.svm_kernel='poly' -> self.svm_c = 0.1

$ python3 main_csp.py

For the recreation of the Riemannian results run main_riemannian.py. Change self.svm_kernel for testing different kernels:

• self.svm_kernel='linear' -> self.svm_c = 0.1
• self.svm_kernel='rbf' -> self.svm_c = 20

Change self.riem_opt for testing different means:

• self.riem_opt = "Riemann"
• self.riem_opt = "Riemann_Euclid"
• self.riem_opt = "Whitened_Euclid"
• self.riem_opt = "No_Adaptation"

$ python3 main_riemannian.py

• Michael Hersche - Initial work - MHersche
• Tino Rellstab - Initial work - tinorellstab