Alejandro Candioti

I'm +1 on removing it from the dictionary. Since the start is a parameter the user won't fully lose the information

Nice version @yxiaoaz ! The interesting thing that I see here is that you basically implemented a shortest path cache data structure. That is a very useful one with plenty...

> > The new keyword parameter seems to be orthogonal to the existing `most_valuable_edge` function: They are both functions, only one can be specified, and they return different length tuples....

> > Are these methods published anywhere? Do we have references or citations? This is very interesting. There are plenty of papers that talk about optimizing betweeness centrality on dynamic...

What about the following interface: - We leave Girman-Newman function to be only for edge betweenness. We deprecate the custom top selection. - We add a new method called something...

So are we good at least doing the cache optimization? That one is still Girvan-Newman and seems to improve performance significantly

Regarding spanning tree iterations: another option could be adding a parameter "minimum_only" that only yields MSPs

I'm not able to reproduce this issue. This test passes: ```python def test_repro(): G = nx.Graph([(1, 2)]) assert nx.density(G) == 1 assert list(nx.all_node_cuts(G)) == [] ``` `all_node_cuts` should exit early...

I put up a property-based test: ```python def _random_connected_graph(n, prob, seed): return nx.compose( nx.random_labeled_tree(n, seed=seed), nx.erdos_renyi_graph(n, prob, seed=seed), ) @settings(print_blob=True) @given( graph_st( _random_connected_graph, n=st.integers(5, 10), prob=st.floats(0, 1) ) ) def...

Does it make sense to have this method discoverable from the tree module too?