fix: isolate required and optional dependencies

[slevang]

This makes the i/o dependencies, which are only required for the .save()/.load() methods, optional. I have an environment where I would really rather not have netcdf4 installed and realized xeofs was the reason it kept getting pulled in.

Along the same lines, numba is a heavyweight dependency and has very limited use here so far (only in GWPCA). numba is probably bigger in size than all other deps combined. statsmodels also has very minimal use (CPCCA). I'm tempted to try and make those optional as well, since probably the majority of people who use this package just stick to EOF. Wanted to get your thoughts first though @nicrie .

EDIT: this does make numba and statsmodels optional as well

[nicrie]

LGTM! I also agree with your suggestion to make Numba and statsmodels optional. What should we call this additional category -- perhaps perf?

Could we add the different installation options to the installation instructions to clarify things for users? Specifically, users of MCA/CCA (which ultimately depend on CPCCA) should be aware that they need the statsmodels dependency.

[slevang]

Ok, I went ahead and moved numba and statsmodels into an optional group as well, with appropriate warnings in the install and contributing docs. I also parameterized the CI over minimal and complete dependencies to make sure the minimal version remains compatible with running the test suite, by skipping certain tests as needed.

I used etc as the miscellaneous group which is what xarray uses for sparse. Could also do accel (which xarray also uses, but this is mostly for things that speed up operations but still work without them) or perf. 🤷

[nicrie]

hey @slevang I was quite packed over the last days, I will have a look now.

👍 for etc

[nicrie]

shouldn't this be a class attribute?__

    self.extra_modules = []

sorry my bad somehow missed that you defined it before the __init__ method.

[nicrie]

LGTM, if you want to address the one point regarding 5hnetcdf great! But even without from my side feel free to merge! :)

[codecov]

Codecov Report

Attention: Patch coverage is 50.00000% with 11 lines in your changes missing coverage. Please review.

Please upload report for BASE (main@a6a32a0). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
xeofs/base_model.py	33.33%	8 Missing ⚠️
xeofs/cross/cpcca.py	50.00%	2 Missing ⚠️
xeofs/single/gwpca.py	50.00%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #237   +/-   ##
=======================================
  Coverage        ?   24.50%           
=======================================
  Files           ?       56           
  Lines           ?     3962           
  Branches        ?        0           
=======================================
  Hits            ?      971           
  Misses          ?     2991           
  Partials        ?        0           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.