facility link snapping

[Theodore-Chatziioannou]

Adds CLI method for snapping activity facilities to a network geometry.

It can be invoked as: pam snap-facilities <path_population_in> <path_population_out> <path_network_geometry> -f <link_id_field>

pam snap-facilities --help

Usage: pam snap-facilities [OPTIONS] PATH_POPULATION_IN PATH_POPULATION_OUT
                           PATH_NETWORK_GEOMETRY

  Snap facilities to a network geometry.

Options:
  -f, --link_id_field TEXT  The link ID field to use in the network shapefile.
                            Defaults to 'id'.
  --help                    Show this message and exit.

For each activity, the method will identify the closest link in the shapefile, and add it as a link=.. attribute to the activity element. Finally, the updated plans file will be saved to a new xml file.

Checklist

Any checks which are not relevant to the PR can be pre-checked by the PR creator.
All others should be checked by the reviewer(s).
You can add extra checklist items here if required by the PR.

• CHANGELOG updated
• Tests added to cover contribution
• Documentation updated

[brynpickering]

Nice and clean, just a few suggested minor changes.

[Theodore-Chatziioannou]

thanks @brynpickering - I have made some updates in response to your comments. Before merging, @gac55 let me know if it works for your use case.

[gac55]

Thanks @Theodore-Chatziioannou , picking this up now

[syhwawa]

Hi @Theodore-Chatziioannou,

I've tested the new pam snap-facilities function with 100 agents' plan for Londinium, and it worked quite well!

When I tried it on TE pop, I noticed it perform quite slow when scaling up to larger populations. It took about 30 minutes of runtime for 0.01% of the population for TE pop.

I feel the primary slowdown seems to come from the way distances are calculated for each activity to all network links. Maybe some vectorized operations or something similar would help to potential speed-ups to process?

[syhwawa]

Previously, the snap_facilities_to_network performed slowly when it needed to work with big population inputs.
Now it utilizes cKDTree for efficient nearest neighbour search and batch processing of activity locations. This update improves performance in snapping facilities to the network geometry now. @Theodore-Chatziioannou

[Theodore-Chatziioannou]

is a def missing here?

[syhwawa]

I think it just calls the functionsnap_facilities_to_network here.

[Theodore-Chatziioannou]

thanks @syhwawa

• your change has broken the CI - can you fix the linting? I believe this due to using single quotes
• the code can be simpler. There is no need to two separate loops, one collecting the data and another applying it, saving the intermediate data in activity_points and activities_info. The updated locations can be queried and assigned directly in the first loop.

[Theodore-Chatziioannou]

is this needed? Loc should always be a point with x and y coordinates (or None).
See here: https://github.com/arup-group/pam/blob/3f817a662317d44fca2126295f07d89317db707c/src/pam/location.py#L7-L18.
If that is not the case in the input data, then the inputs should be corrected rather than fixing them silently here.

[syhwawa]

I have removed this.

[syhwawa]

Thanks for the comments @Theodore-Chatziioannou

The function has been updated to remove the separate loops for collecting and applying data. Now, the updated locations are queried and assigned directly within the first loop. And the lint error has been fixed.