We have a while before anything leaves, but worth keeping in mind that this will start to fail.
scipy itself recommends moving to using pywavelets. Assuming we are not going to implement our own wavelet code, I think the options we have here are:
add our own version of scipy.signal.morlet and use that
update to use pywavelets, and add it as a dependency (or maybe, as an optional dependency)
drop support for computing wavelets in neurodsp (and suggest people go use pywavelets, or maybe MNE)
I don't have any particularly strong feeling about this one way or another (I don't really use wavelets from ndsp, and don't have much sense of to what extent others do) so if anyone wants to advocate for any particular approach, let us know!
In neurodsp.timefrequency.wavelets, we use the scipy.signal implementation of Morlet wavelets (which we also use for compute_spectrum_wavelet).
scipy is deprecating their morlet wavelet implementation (as of current version 1.12) and announced that they will remove it as of 1.15: https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.morlet.html
We have a while before anything leaves, but worth keeping in mind that this will start to fail.
scipy itself recommends moving to using
pywavelets. Assuming we are not going to implement our own wavelet code, I think the options we have here are:scipy.signal.morletand use thatI don't have any particularly strong feeling about this one way or another (I don't really use wavelets from ndsp, and don't have much sense of to what extent others do) so if anyone wants to advocate for any particular approach, let us know!