shading: use morphology.footprint_rectangle((1, 3)) for 1x3 structuring element

[Omesh37]

Fixes a TypeError / FutureWarning caused by differing scikit-image APIs. The structuring element creation in pvanalytics/features/shading.py now uses the modern morphology.footprint_rectangle((1,3)) call.

Testing

• Closes Tests emit FutureWarning from skimage #226
• Non-API functions clearly documented with docstrings or comments as necessary.
• Adds description and name entries in the appropriate "what's new" file
  in docs/whatsnew
  for all changes. Includes link to the GitHub Issue with :issue:`num`
  or this Pull Request with :pull:`num`. Includes contributor name
  and/or GitHub username (link with :ghuser:`user`).
• Pull request is nearly complete and ready for detailed review.
• Maintainer: Appropriate GitHub Labels and Milestone are assigned to the Pull Request and linked Issue.

[cwhanse]

@Omesh37 did you test this locally? pvanalytics isn't getting much work and this fix has uncovered another unrelated problem.

[Omesh37]

@cwhanse Yes, I have tested using command pytest --cov=pvanalytics --cov-config=.coveragerc --cov-report term-missing pvanalytics --runslow locally, which results 224 passed, 29 skipped (require ruptures, not installed locally), 0 failed.
After appending .copy() at the end of ghi_image[...] = np.nan

[cwhanse]

@Omesh37 there's also an error related to converting a timedelta. If you don't have time/patience to track that down, I'll see if I can push to your PR.

[Omesh37]

@cwhanse I suggest a possible fix to this that is:

offset = frequencies.to_offset(freq)
try:
   return pd.to_timedelta(offset)
except ValueError:
   return pd.Timedelta(offset.nanos, unit='ns')

Storing pandas offset object and then returning the result. I have tested using:
pytest pvanalytics/tests/quality/test_gaps.py::test_completeness_score_no_data -v and it was successful locally.
But, by mistake i have pushed the code in wrong PR (#235)

[cwhanse]

@Omesh37 let's not fix that timedelta here. I'll submit a PR to fix it, then we can move this forward.

[Omesh37]

@cwhanse Sorry about that! I've reverted the timedelta change from PR #235. It now only contains the divide-by-zero fix in _fit.py.

[cwhanse]

@Omesh37 can you rebase this to main, so we can see if the Timedelta fix + this PR exposes any new errors? Thanks.

[Omesh37]

@cwhanse I have rebased to main :)
Let's see what errors we get now!

[Omesh37]

@cwhanse The import fallback needed to be at module level rather than inside the function, otherwise footprint_rectangle was never bound before the call. All tests pass locally :)

[cwhanse]

Le sigh. The fix I pushed in #236 requires pandas > 2.1. Since pvanalytics stops tests when the first one fails (I should change this), the error went undetected.

If you can find a way to fix that statement (series. index.diff()) without advancing pandas, that's fine. Or, I can bump up the minimum pandas version.

[Omesh37]

@cwhanse I tried using plain index subtraction which works across pandas versions.
seconds_per_sample = (series.index[1] - series.index[0]).total_seconds()
All the test was passed successfully :), could be a option

[cwhanse]

Winner. Thanks for sticking this one out.

[Omesh37]

Finally it worked. Glad to help you out :)