Hi everyone,
I’m really new on this domain and I’m trying to perform RSA on fMRI data. The concept by itself is quite tricky to understand for me so my question will be maybe really really naive (not so stupid, I hope).
So first, I was interested about the crossnobis distance as I understand it as a really robust and fiable measure, but for my own experiment, the correlation distance seems to be the best to reflect the effect (more spatial than linked to activation) in which I’m interesting.
But I asked myself whether it was possible to take noise into account, as with crossnobis, or at least to use crossvalidation.
I looked to see if a cross-validation implementation with the correlation distance was available in the RSA toolbox (2019, Python, Python Representational Similarity Analysis (rsatoolbox) toolbox — rsatoolbox 0.2.0 documentation) and I’ve read quite a few papers in which correlation is used and I’ve never found crossvalidation on correlation distance.
So, here is my question, I’m wondering if something make the crossvalidation incompatible on correlation distance calculation ? Or maybe the correlation distance loose its sense/interpretation ?
Thanks a lot,
Aurore