search

adolphslab / HCP_MRI-behavior

Code for predicting individual differences in behavioral variables (e.g., intelligence, personality) from resting-state fMRI functional connectivity, using data from the Young Adult Human Connectome Project
GNU General Public License v3.0
36 stars 17 forks source link

confound regression: could it be done outside the cross-validation loop? #4

Open jcrdubois opened 5 years ago

jcrdubois commented 5 years ago

we should think whether the confound regression could simply be done on all subjects without introducing any meaningful leakage of information. What we do is very safe, but perhaps overkill and unnecessary

paola-g commented 5 years ago

We could and I think it's a very common approach in literature, even though some advocate a cross-validated approach is the best way to reduce bias. Do you think it impacts substantially the computational demands?