Improve _build_hdbscan_graphs

[paco-barreras]

This method should simply:

1. Build a MST
2. Add self-loops with the core_distance
3. round edges (for efficiency) and sort

Can networkx do this faster and better? Can the core distances not be quickly computed at this stage?

The output: is it the same graph in two formats? can't networkx handle returning (and operating) on a single graph object?

[paco-barreras]

• Core distances can be quickly computed from networkx weighted graph G (already truncated by time)
• HBDSCAN graph is G with core distances as self-loops, then we modify the edges with something like

def MRD(G):
    H = G.copy()
    nx.set_edge_attributes(
        H,
        {
            (u, v): min(data["weight"], G[u][u]["weight"], G[v][v]["weight"])
            for u, v, data in G.edges(data=True)
            if u != v
        },
        name="weight",
    )
    return H

• Minimum spanning tree is very easy with MST = nx.minimum_spanning_tree(G).

@andresmondragont