I fitted some data (20s stimulus) using the SPMG3(20) model and got results like the one shown below (fit in blue):
Are there no constraints on the combinations of the basis functions? I am just wondering how this peculiar shape with the 3 peaks can happen, it seems like an overfitting problem.
My goal is anyway to show that this model performs poorly for this particular data compared to afnis nonlinear regression.
But I think it is known that unconstrained basis sets can produce quite nonsensical shapes, which a reviewer might criticize.
How does the official spm software handle this? maybe there are some hard constraints how much of the temporal/spatial derivative can be added.
Edited 1 time(s). Last edit at 01/21/2014 11:25AM by paranoidandroid.