Shortest Path as Time instead of Distance

[cojow]

Hey,

I'm trying to create a network skim from a gmns network using path4GMNS, specifically using the shortest path between two nodes function.
Currently this function gives the distance and the path sequence. You also have a function that computes time-dependent link travel times. Is there a way that the shortest path function can give out the travel time in minutes instead of the distance?

I've already made a branch of the code where I compute a network skim using a modified find_shortest_path, but I have to convert the length values into fftt before I read it into the network, and then change them back afterwards. This is able to get me a a network skim in minutes.
But when I do the same conversion to the network after it is read into path4gmns the engine gives our travel times in miles, even though the lengths in the network have been changed to the same values as fftt.

Any guidance or assistance would be appreciated. Thanks.

Here's a link to my branch for reference.
https://github.com/cojow/Path4GMNS_network_skimming.git

[jdlph]

Thank you for opening this ticket!

1. G.links[i].length = G.links[i].fftt at line 437 would not work as the shortest path calculation utilizes link_cost_array of the network. The initialization is done in allocate_for_CAPI() in class Network. As shown in the code snippet below, link_cost_array is already initialized using the free flow travel time of each link.

link_cost_array = [link.fftt for link in self.links]

2. This actually indicates a bug in printing out the shortest path information (find_shortest_path()), where the unit shall be in time rather than in distance. We will fix this bug and include it in the next release.

    if seq_type.startswith('node'):
        return f'distance: {path_cost:.2f} mi | node path: {path}'
    else:
        return f'distance: {path_cost:.2f} mi | link path: {path}

3. As single_source_shortest_path(G, from_node_id, engine_type='c') already establishes the shortest path tree from the source node (i.e., from_node_id), you do not have to repeatedly invoke find_shortest_path_distance() in your compute_row_distances(). In other words, you can simply retrieve the shortest path distance/time from this node to any other node using _get_path_cost(G, to_node_id), which is illustrated in the following code snippet. It shall boost the performance of your implementation.

    single_source_shortest_path(G, from_node_id, engine_type='c')
    return [_get_path_cost(G, col_node) for col_node in nodes]

Please let me know if you have any further questions.

[jdlph]

The bug on printing out the proper shortest path information has been fixed under commit 0034598. We also introduce a new API get_shortest_path_tree() under commit 141bcab.

You may want to pull out the latest dev branch (dada26d) to see how we approach this-that is, finding the shortest path according to different cost types (i.e., distance or cost). We hope it can give you some insight on accomplishing your implementation.

Thank you again for opening this tickect, which helped us identify the above error and improve the source code!!

The ticket is closed now. You can reopen it if you have any follow-up questions or comments.