Integration of CalcDeltaImpact class of unsequa into develop

[simonameiler]

Changes proposed in this PR:

• Integration of CalcDeltaImpact class of unsequa into develop
• Update unsequa tutorial
• Modification of plotting functionalities of the uncertainty output in unc_output to skip columns of the uncertainty output dataframe which are zeros or nan.
• addition of a small function called save_divide in util which handles division by zero and NaN values in the numerator or denominator. This is needed for the relative impact change calculation (delta impact). Tests for the function are also added.

PR Author Checklist

• Read the Contribution Guide
• Correct target branch selected (if unsure, select develop)
• Descriptive pull request title added
• Source branch up-to-date with target branch
• Documentation updated
• Tests updated
• Tests passing
• No new linter issues
• Changelog updated

PR Reviewer Checklist

• Read the Contribution Guide
• CLIMADA Reviewer Checklist passed
• Tests passing
• No new linter issues

[emanuel-schmid]

seems counter intuitive to me. why not safe_divide([nan, nan, 1.], 4) == [nan, nan, 0.25] ?
I don't understand why division by zero is dealt with element wise but if any of the elements is nan all should turn to nan?
Since safe_divide(4, 0) is equal to safe_divide(4, nan), i.m.o., safe_divide([4, 4], [1, 0]) should also be equal to safe_divide([4, 4], [1, nan]).

[simonameiler]

good point. the choice in the safe_divide function to treat NaN values in array inputs differently than division by zero is indeed not necessary. I updated the function:

` with np.errstate(divide="ignore", invalid="ignore"):
result = np.true_divide(numerator, denominator)

    # Check if the result is a scalar
    if np.isscalar(result):
        if not np.isfinite(result):
            return replace_with
    else:
        # Replace infinities, NaNs, and division by zeros in np.ndarray
        result[~np.isfinite(result)] = replace_with

return result`

[emanuel-schmid]

There is no test for this case yet, or is there?

[simonameiler]

With the revised safe_divide function, the specific code section is no longer present. However, I still added tests for NaN handling.

[emanuel-schmid]

there are only tests for scalar and np.ndarray, but it could be a just a list - right?

however - instead of adding tests I'd rather pin the input data to np.ndarray.

[simonameiler]

yes, indeed. I added a test for lists too.

[emanuel-schmid]

the return type for array-like input is np.ndarray

[peanutfun]

This looks good and useful! There's not much left to do:

• Please resolve all linter issues (except too-many-arguments and too-many-local-variables). This should be rather simple
• The module documentation is missing. Please add an appropriate section to doc/climada/climada.engine.unsequa.rst and double-check the rendered docs.
• The tutorial section on the new class should probably start with an explanation what the class does (current paragraphs 2 and 3), and then mention that it works analogously to the other class (paragraph 1)
• The table of contents in the tutorial was not updated, and the links don't work anyway. Maybe just remove it for now?

[peanutfun]

Good idea, but it looks a bit weird that the subplots are missing completely. Instead of removing the axes, you could make an empty plot and write a short text like "No data to plot"?

[simonameiler]

This looks good and useful! There's not much left to do:

• Please resolve all linter issues (except too-many-arguments and too-many-local-variables). This should be rather simple
• The module documentation is missing. Please add an appropriate section to doc/climada/climada.engine.unsequa.rst and double-check the rendered docs.
• The tutorial section on the new class should probably start with an explanation what the class does (current paragraphs 2 and 3), and then mention that it works analogously to the other class (paragraph 1)
• The table of contents in the tutorial was not updated, and the links don't work anyway. Maybe just remove it for now?

I addressed all your points. In my eyes, it's good to go.

[peanutfun]

Thank you @simonameiler! 🙌 I fixed the remaning whitespace issues and will merge as soon as the pipeline succeeds (which it should 🤞)