Improve test coverage for Steiner Tree & Docs

[vanshika230]

I have increased test coverage for Steinertree.py according to here :-
https://app.codecov.io/gh/networkx/networkx/blob/main/networkx%2Falgorithms%2Fapproximation%2Fsteinertree.py
It is now at 100% as seen below:-

I have also added Raises to docstrings and changed the default from kou to mehlhorn according to recent changes.

[rossbar]

Generally LGTM, and thanks for the cleanup of things I missed in #7337 🚀

Just one note about how to structure the default method test, but other than that LGTM!

[rossbar]

This LGTM, thanks @vanshika230 !

The dispatching tests are failing but I'm not sure why... @eriknw any ideas?

[eriknw]

The dispatching tests are failing but I'm not sure why... @eriknw any ideas?

Thanks for the ping! You can expect me to be around to help with issues like this. This is a good data point, and I suppose we should make the assertion message more informative.

EDIT: this is not the best solution. See my next reply

I debugged this locally, and the solution is to change the _dispatchable decorator of steiner_tree to:

@nx._dispatchable(edge_attrs={"weight": None}, returns_graph=True)

The issue is that "weight" edge attribute was being filled to 1 (the default value according to the docstring) in the converted backend graph, but the correct behavior is to not add edge values if they don't exist. Using edge_attrs={"weight": None} will preserve the weight edge attribute if it exists, but won't add new ones.

[eriknw]

Correction--use this _dispatchble:

@nx._dispatchable(preserve_all_attrs=True, returns_graph=True)

The returned graph preserves edge, node, and graphs attributes from the original graph.

[dschult]

Thanks! This is very helpful!