Anatomical doesn't improve from 30 comps
to 65 comps (The differences are due to different test/train splits in the separate runs of the pipeline)
But SVD does from 30
to 65
This doesn't make sense as anatomical components aren't even ordered by "importance" as opposed to SVD components, therefore SVD with 30 "random" components out of 65 comps should performe worse in comparison. In feature.py max_comps is passed regardless of the used parcellation therefore I don't know how this happens.
Anatomical doesn't improve from 30 comps
to 65 comps (The differences are due to different test/train splits in the separate runs of the pipeline)
But SVD does from 30
to 65
This doesn't make sense as anatomical components aren't even ordered by "importance" as opposed to SVD components, therefore SVD with 30 "random" components out of 65 comps should performe worse in comparison. In feature.py max_comps is passed regardless of the used parcellation therefore I don't know how this happens.