Hi Rasmus,
According to Gang, those values are not properly normalized in
order to compare across different models, and what you suggest
is surely part of that. So that is where we start from (saying that
you should not do things like alter the number of stimuli or the
durations). It is generally used to optimize within a given design,
not across designs.
Note that MRT.py does not work to hard to think of what should
be appropriate in any case. It essentially provides a starting point.
If you would like, I could add an option to provide basis functions
on the command line.
- rick