Serialization of Spectrum1D object (How to make Spectrum1D pickle-able?)

[MichalSteiner]

Hello everyone,

I struggle with serialization (through the pickle or dill library) of the sp.Spectrum1D object. In particular, the default wcs is not pickle-able. I use serialization to both save the data as pickle file and for multiprocessing of several functions, so this functionality would be helpful, especially since I actually don't really care about the sp.Spectrum1D.wcs attribute.

I know it's possible to bypass this by temporarily removing the wcs with spectrum.wcs = None, or using the high-level API WCS class. However, the first solution is a patchwork solution that doesn't solve the fundamental issue (the default wcs is re-added at some point in my pipeline). I think it is possible to setup the high-level WCS for spectral data, but I am unsure how to do it properly. Right now, I am using the spam CTYPE of WCS to do this, but this resets the spectrum.spectral_axis to pixel array in several functions, which is something I would like to avoid.

Is there a workaround that would fix this issue altogether? I think the easiest way is to create the WCS correctly, but unfortunately, the WCS documentation is not super helpful for the spectral data. As far as I gathered, since my spectral_axis is not linear, I should use the tabulated option, based on this paper, Greisen et al. (2006), section 6. Is there some convenience function in specutils that creates high-level WCS given a spectral_axis?

Minimal example:

import dill # Pickle library
import numpy as np
import specutils as sp
# Mock spectrum
spectrum = sp.Spectrum1D(spectral_axis = np.arange(10)*u.AA,
                              flux = np.arange(10)*u.dimensionless_unscaled
                              )
print('Pickleable :' , dill.pickles(spectrum))

# Remove wcs
spectrum.wcs = None
print('Pickleable :' , dill.pickles(spectrum))

# Mock spectrum with high-level API WCS of ctype = 'SPAM'
wcs = WCS(naxis=1)
wcs.wcs.ctype = ['SPAM']
spectrum = sp.Spectrum1D(spectral_axis = np.arange(10)*u.AA,
                         flux = np.arange(10)*u.dimensionless_unscaled,
                         wcs = wcs
                         )
print('Pickleable :' , dill.pickles(spectrum))