2.0.0 RC1 Feedback

[rosteen]

Specutils 2.0 RC1 (release candidate 1) has been released on PyPi. You can install this version by calling pip install specutils --upgrade --pre - it would be useful for some folks to start testing this now and raising any concerns you might have before 2.0 is actually released. Please feel free to leave such comments and concerns in this discussion. The major changes to 2.0 are:

New Features

• Spectral axis can now be any axis, rather than being forced to be last. See docs
  for more details. [Allow spectral axis to be anywhere, instead of forcing it to be last #1033]

• Spectrum now properly handles GWCS input for wcs attribute. [Improved GWCS handling in Spectrum1D #1074]

• JWST reader no longer transposes the input data cube for 3D data and retains
  full GWCS information (including spatial). [Improved GWCS handling in Spectrum1D #1074]

• Spectrum1D renamed to Spectrum. [Update spectrum class name from Spectrum1D to Spectrum #1126]

Other Changes and Additions

• Initializing a Spectrum with only a spectral axis (not full WCS) will now
  result in a GWCS matching the dimensionality of the flux array, rather than a
  1D spectral GWCS in all cases. [Refactor gwcs_from_array to provide ND GWCS in ND flux case #1211]

• Spectrum arithmetic now checks whether the spectral axes of the two operand Spectrum
  objects are equal, and fails if they are not. [Refactor gwcs_from_array to provide ND GWCS in ND flux case #1211]

For convenience, I've opened a draft PR from the v2.0-dev branch to main for anyone who would like to view the full changes to the actual code: #1220.

[pllim]

A quick run over at https://github.com/astropy/astropy-integration-testing/ failed for specreduce. Do the tracebacks look familiar to you?

https://github.com/astropy/astropy-integration-testing/actions/runs/13929757409/job/38983646191

Here is a native specreduce run, also failed:

https://github.com/astropy/specreduce/actions/runs/13892729677/job/38983700486

[rosteen]

Yes, basically it looks like specreduce will need to add spectral_axis_index=disp_axis in the Spectrum creation call in _get_data_from_image (and maybe one or two other places, that's just the first one I looked at). The test images are 10x10, so specutils can't automatically determine which axis is the spectral axis, since both flux axes have the same size as the spectral axis array that's input (specutils no longer assumes that the last axis corresponds to the spectral axis).

[pllim]

OK I opened astropy/specreduce#259 and astropy/specreduce#260

[pllim]

Huh, looks like spectral_axis_index=-1 does not work. Naively, I thought that would replace the meaning of "use last axis". I get IndexError.

[rosteen]

FYI I added the ability to specify spectral_axis_index=-1, but it ends up getting stored under the hood as whatever ndim-1 is.

[rosteen]

The window for feedback is over, closing this.