Closed thomas-fred closed 1 year ago
Thanks for this @thomas-fred - clearly a useful speed-up, that I think we should merge in, even if it's a lone introduction of multiprocessing within the codebase for now.
Let's catch up to think through what a more performance-oriented design could look like.
When running
link_nodes_to_nearest_edge
for a large network (gridfinder, with 1.6M nodes, 3.6M edges), the majority of processing time is spent insplit_edges_at_nodes
. This PR is how I accelerated that task from ~days to an hour or so with process parallelism.