Add axis reduction function

[cdeil]

We should add a function or functions to reduce the number of axes by slicing one plane or performing some reduction operation over some axis like sum or mean.

Just to illustrate what I mean, in Gammapy we have cube_to_image.
@keflavich Do you have an existing function or suggestion how to implement this in a generic way?

[keflavich]

Yes - and I think this probably does not belong in imageutils. https://github.com/radio-astro-tools/spectral-cube is the place for this, though perhaps we can factor out the spectral part. See in particular https://github.com/radio-astro-tools/spectral-cube/blob/master/spectral_cube/spectral_cube.py#L223

[cdeil]

I'd prefer if we put this somewhere in the Astropy core ... IMO this is general enough. (It has been (partly) re-implemented in an affiliated package for radio- and one for gamma-astronomy and none want to have the other as a dependency, right?)

I think it could go in astropy.image or astropy.wcs or astropy.nddata, not sure which is best!?

[keflavich]

OK, the very simple example you illustrated in cube_to_image is basically already in astropy; e.g. https://github.com/astropy/astropy/pull/2326/files#diff-d4c5b9e74c6c26000684075497033504R1879

slice = cube[:,:,0]
slicewcs = cubewcs.dropaxis(2)

projection = cube.sum(axis=0)
projwcs = cubewcs.dropaxis(0)

[keflavich]

For more sophisticated reductions across axes, I think nddata is the right place. But for now, since it's a 2-liner to get the WCS right for these simple operations, I'd say additional wrappers are probably not necessary.

p.s. wcs.sub also does this. I only learned about it 2 months ago, but it is awesome.