Degree Centrality Calculation Differs From Networkx about graphblas-algorithms

[eriknw]

Thanks for the report and sharing the issue on StackOverflow @davidsilveiro!

In the example above G is directed (as is required for in_degree_centrality), and GG is undirected. So, to fix the example, change GG = ga.Graph.from_networkx(G) to GG = ga.DiGraph.from_networkx(G):

- GG = ga.Graph.from_networkx(G)
+ GG = ga.DiGraph.from_networkx(G)

I would also suggest considering using the networkx API for dispatching:

nx.in_degree_centrality(GG)

or

nx.in_degree_centrality(G, backend="graphblas")

The latter converts the input graph G to a graphblas graph. This conversion can be cached in NetworkX 3.3 via first setting a configuration:

nx.config.cache_converted_graphs = True

Finally, ga.in_degree_centrality should probably raise if the input is an undirected graph, so this is a valid bug report. Using ga.nxapi.in_degree_centrality(GG) raises as expected if GG is undirected. I think having functions appear in both ga and ga.nxapi namespaces may be a little unclear, and it would be better if instead we moved the networkx backend API to e.g. nx_graphblas.