Bug with EEG decoding

Dear François,
I tried to use decoding for EEG data : SVM plug is installed, Brainstorm is updated
This is an oddball task with two condition with 1345 trials (275 vs 1070).
Parameters are Sensor types : EEG, Low-pass = 30 Hz, Number of permutations = 100 and folds =5, Decoding = pairwise
The process does not execute fully with the error
Any idea to solve this ? Thanks a lot !
Julien

Can you please share a small example dataset?
Zip together some file that can allow us to reproduce exactly this error:

  • A few data files that if placed in FilesA/FilesB produce this error (try with a small number of files)
  • The channel file(s) associated with these data files

Thanks

@pantazis

Ok fine.
Where I send you a link for get the data ?
Thanks
Julien

Thank you for the example files.
I tested them on my computer and it works:

It looks like you work on MacOS, and I suspect that you haven't compiled the library. I made the instructions more explicit in the tutorial:
https://neuroimage.usc.edu/brainstorm/Tutorials/Decoding#Download_and_installation

I added a test to make the error message more explicit than what you got:
https://github.com/brainstorm-tools/brainstorm3/commit/aab58620804c95da9e31f327f3b8b3b04b8aced1

Perfect, you were fully right ! After having compiled the library, I had a nice result decoding result.
Thanks a lot for your help and this very nice process.
Last question : is there a significant value for decoding accuracy ? 80 - 90 %? I guess this is user's "choice" ?
Best

@pantazis ?

Hi Jungju,

To establish statistical significance in decoding results, you still need to do statistics because high classification values can still arise by chance. See for example https://www.sciencedirect.com/science/article/pii/S0165027015000114

I use permutation tests in my work, when having multiple subjects and each subject contributes one decoding time course.

Best,
Dimitrios