A Python CLI script to run travelling salesman problem (tsp) algorithms using brouter routing times.
In addition to the CLI script the tsp_brouter library includes functions to call the algorithms from within Python, of course.
brouter provides routing services with profiles that can be particularly interesting for touring bicyclists. It is highly configurable. tsp_brouter_cli.py is a Python3 script that calls the brouter server and returns (in -createdm mode) a "distance matrix" and route segments for all the points that the user provided (via a point file that is read by the gdal/ogr library).
In the -routes mode tsp_brouter calls up to four algorithms to get traveling salesman problem solutions. Three of them I wrote myself (as an exercise, I suppose):
- a Brute Force algorithm to get an exact solution at such a high cost in processing time that it can't be run for more than about 12 points.
- a Nearest Neighbor algorithm that starts at one point and continues along the route to the end point always going to the next closest available point to a defined end point.
- a Shortest Segment algorithm that looks at all available route segments and always includes the shortest one until there are no available segments left.
- a much more clever and fast algorithm than the ones I wrote, using the ortools library from Google.
The algorithms are available in both one way (the user indicates the start and end points) and round trip versions.
The user selects which parameter from the brouter results to minimize (distance, travel time, energy, or cost (as defined by the profile)). By supplying your own custom profile for brouter you can get time estimates and preferred routes that are tailored to your own riding and prefernces. One custom profile is included in the brouter_profiles folder. It's one that works well for me in mountainous regions of Turkey. The m11n server has many interesting profiles.
The output is a gpx file of the route (to put in one's phone and follow on a bicycle tour, perhaps!).
Required Python3 libraries that may not be installed by default include:
In my testing I was unable to get GDAL and numpy included in the tsp_brouter installation package. Install them separately if you don't already have them.
Numpy should easy:
pip install numpy
GDAL can perhaps be installed like this:
pip install GDAL
Let me know if that doesn't work. There seem to be lots of ways to install GDAL.
Then installing tsp_brouter should be easy:
pip install tsp_brouter
Let's find a bicycle touring route from Demre to Kaş while visiting some Likyan sites.
If you're not familiar with ogr vector datasets, then the easiest way to prepare your data for tsp_brouter may be with simple text files. In sample_data I provide some points in a .csv file (demre_kaş_pt.csv) and an associated .vrt file that enables ogr to read the .csv
The first few lines of the .csv file look like this:
name,desc,longitude,latitude Kyaneai antik kenti (aka Cyaneae),,29.815071,36.245389 Demre,start,29.987891,36.242588 turnoff to Trysa,,29.896623,36.263519 Antiphellos (Likya) rating 2,end,29.638490,36.201590
You'll see that I explicitly specified a start point and an end point in the desc field. Coordinates are in wgs84.
$ tsp_brouter_cli.py tsp_brouter_cli.py -h help -list-servers -list-profiles server_name -verify input_file -l layer -fname name_field -limit number -createdm input_file -l layer -fse se_field -fname name_field out_distance_matrix -routes -rt -ow -bf in_distance_matrix -dp distance_proxy_name out_gpx -server name -profile name
First we'll create our distance matrix, a file we name
$ tsp_brouter_cli.py -createdm demre_kaş_pt.vrt -fse desc -fname name demre_kaş_dm.p server not specified, using default value: brouter profile not specified, using default value: trekking [============================================================] 100.0% ... demre_kaş_dm.p written number of points: 15 script execution time: 0:01:36.033249
Then we create a gpx file of our route, a file we name
$ tsp_brouter_cli.py -routes -ow demre_kaş_dm.p demre_kaş_ln.gpx distance matrix profile not specified, using the default value: time ERROR 1: Empty geometries cannot be constructed NNF: 52992.0 startPT: 0 PTCount: 15 NNR: 48333.0 startPT: 0 PTCount: 15 number of points: 15 script execution time: 0:00:00.773349
Now you can follow the route on your bicycle tour!
- allow user to specify which algorithms to run (currently the only optional one is the brute force algorithm because it is slow)
- cache data from server to avoid needing to call it again for the same from point-to point-server-profile combination
- allow user to override check that their non-local profile already exists (in case the profile exist on the server but not in the tsp_brouter dictionary)
- return dictionaries instead of lists (GetShortestRoute, ReturnTimeGeometryFromGeoJSON, GetTravelTimes, VerifyPTs) (I think this is better coding practice; I'm sure there's a lot I can do toward better coding practice)
- how should I structure the code? It's all just functions (procedural?) now. I suppose I should have a class that runs the algorithms as requested since they pretty much all take the same inputs?