Mesa NetworkGrid Requires 0-indexed nodes and can't take non-numeric node names

[clarkpetri]

Describe the bug
Mesa fails to run a model, without providing an error, when non-zero indexed network structures are provided as the NetworkGrid object. Additionally, this occurs if a zero indexed networkx object later has a node with a non-numeric name added.

Expected behavior
The model should run and visualize the graph object with arbitrary node names with the same ease in which a NetworkX object can be initiated as such

To Reproduce
This discussion explains everything.

Additional context
Occurred on an M1 MBP running Python 3.10.2

NetworkX can handle a simple example such as:
G = nx.Graph()

edges = [(0,1),(1,2),(2, 0),(2,'a')]

G.add_edges_from(edges)

nx.draw(G, with_labels=True)

But the presence of the (2,'a') edge will cause the bug in Mesa. Removing that edge causes Mesa to initiate properly.

[rht]

The underlying cause is that Python's NetworkX allows node ID to be any hashable object (int, str, tuple, etc), but JavaScript's D3.js expects nodes to be identified by their order in the node list. There is no way around this (see https://stackoverflow.com/questions/23986466/d3-force-layout-linking-nodes-by-name-instead-of-index) than to write a mapper from NetworkX node ID to a node index number (that MUST start from 0).

It might be possible to implement the mapper inside Mesa so that it is automatic. I'm currently looking into it.

But for a temporary solution, the network_portrayal code of the virus on a network example can be modified to

    nodes = []
    node_label_to_node_id = {}
    node_id = 0
    for node_label, agents in G.nodes.data("agent"):
        agent = agents[0]
        nodes.append({
            "size": 6,
            "color": node_color(agent),
            "tooltip": f"id: {agent.unique_id}<br>state: {agent.state.name}",
        })
        node_label_to_node_id[node_label] = node_id
        node_id += 1

    portrayal = dict()
    portrayal["nodes"] = nodes

    portrayal["edges"] = [
        {
            "source": node_label_to_node_id[source],
            "target": node_label_to_node_id[target],
            "color": edge_color(*get_agents(source, target)),
            "width": edge_width(*get_agents(source, target)),
        }
        for (source, target) in G.edges
    ]

[clarkpetri]

Thank as always for such a comprehensive explanation. I hadn't considered the JavaScript limitation. Knowing that NetworkX takes any washable object and that Mesa is using those tools, the limitation seemed elusive.

I'll implement this code soon and report back.

[clarkpetri]

Implemented, works, and clever solution. I can maintain an external dictionary relating the enumerated IDs to original names.