SSD on reduced rank data

Hi all,

I was wondering if anyone has any experience on using Spatio-Spectral Decomposition (SSD) on data that does not have full rank (e.g. after removing artifactual components with ICA).
Does the method still provide meaningful results on low-rank data?

Thanks in advance.


Yes, it’s completely fine to apply SSD to data that does not have full rank, the results will still be meaningful. Just keep in mind that the number of components returned would be limited to the rank of the data.

Internally there are some additional steps which must be taken for SSD to work with this type of data, but this is all done automatically in MNE after computing the data’s rank.

In case your data has some small but non-zero singular values and this is not caught (should raise a LinAlgError), you can always manually specify the rank of the data using the rank parameter.

Hope this helps!

1 Like