Implicit transformations in patch_function

[d-chambers]

A number of current or planed processing methods (whiten and correlate come to mind) involve processing in the frequency domain. This involves a dft, performing the processing, and an inverse dft. As many of these might be used together, it would be nice if these functions were flexible enough to take a time domain or frequency domain patch, and the output would match the input domain.

Specifically, I am thinking of adding some parameter to the patch_function decorator which would handle the input/output transformations (if applicable) so the processing functions can assume a frequency domain patch as input and return a frequency domain patch. I am not sure what this should look like yet, but as an initial thought:

@patch_function(tranform='dft', inverse_transform='idft')
def some_patch_function(patch, **kwargs):
    # now we can rely on patch having been transformed to the frequency domain with the dim specified by kwargs
    ...

I suppose we might also need a transform_kwargs argument as well to be able to pass arguments to the transform implementation. Can anyone else think of a cleaner way to to do it?

[d-chambers]

We now have some functions (eg spectral whitening, correlation) that can handle inputs in time or frequency domain in a more explicit way so I think this can be closed for now.