GraphLoader

[aMahanna]

Another attempt inspired from #8 to provide a GraphLoader class that returns the following representations of an ArangoDB Graph:

• A Vertex Map dict[str, dict[str, Any]]
• An Adjacency List dict[str, dict[str, dict[str, Any]]]
• A COO Representation Tuple[numpy.ndarray, numpy.ndarray]
• A Vertex ID to Index Map dict[str, int]

This gives us a tool to offer 1) ArangoDB to NetworkX & 2) ArangoDB to NetworkX-cuGraph

Lots of things to iron out, but this is a start

Note: node/edge field specification is disregarded here, i.e we're loading everything

host = "http://localhost:8529"
username = "root"
password = "passwd"
db_name = "_system"
graph_name = "KarateGraph"

client = ArangoClient(hosts=host)
db = client.db(db_name, username=username, password=password)
graph = db.graph(graph_name)

v_cols = graph.vertex_collections()
edge_definitions = graph.edge_definitions()
e_cols = {c["edge_collection"] for c in edge_definitions}

metagraph = {
    "vertexCollections": {col: {} for col in v_cols},
    "edgeCollections": {col: {} for col in e_cols},
}

start_time = time.time()

result = GraphLoader.load(
    db_name,
    metagraph,
    [host],
    username=username,
    password=password,
    parallelism=..., 
    batch_size=...,
    load_node_dict=True,
    load_adj_dict=True,
    load_as_undirected=True,
    load_coo=True,
)

end_time = time.time()

print(end_time - start_time)

assert type(result) == tuple
node_dict, adj_dict, src_indices, dst_indices, vertex_ids_to_indices = result

[aMahanna]

Closing in favour of #21