(can create separate issues for these as needed, just link from here)
Following is an incomplete list of basis functions to include, we can add more as needed:
[ ] Fourier basis
[x] 2d basis
[ ] for position
[ ] for visual input. separate for e.g., retina and V1?
[ ] circular (head position, orientation, etc)
[x] allow for Cartesian product of basis: accept two bases and generically take their product. Could be used to cross position in space and head direction or to generate the simpler 2d bases above
I am coding up the Fourier basis, still working on a PR for it.
A couple of considerations:
Should we enforce equi-spaced sampling when generating the basis? this would make it easy to check for aliasing, and would enforce the evaluated basis to be sinusoidal.
Convolving with this basis is equivalent of using a sliding window FFT as a predictor. would this estimate be robust enough (without multi-tapering or similar robust estimation procedures)?
(can create separate issues for these as needed, just link from here)
Following is an incomplete list of basis functions to include, we can add more as needed: