psycopgr: Routing in Python

I am introduing  psycopgr, a Python wrapper of pgRouting written by me.

What is psycopgr

As said in pgRouting docs:

Just considering the different ways that the cost can be calculated, makes it almost impossible to create a general wrapper, that can work on all applications.

Indeed, in many applicatons you may need to modify the database tables and fill some computed values to fit your specific purpose, which is often done in a preprocessing stage by SQL, before real routing starts working. It is not appropriate to be wrapped.

However, after preprocessing things such as database creation, map data import, tables re-calculation and update, you are ready to use psycopgr to do another simple thing:

Computing optimal routes from nodes to nodes on real map, in Python.

Note that psycopgr is never a general purpose wrapper of pgRouting. I am a novice in GIS and what I want from this tool is just routes (with lowest costs) from places to places without writing a single line of SQL.

For preprocessing stage, I have a post “pgRouting notes” for my own reference. Before enjoyable Python coding, you have to prepare a few stuffs.



  1. Prepare a PostgreSQL database, install PostGIS and pgRouting extensions, and import map data to database. “pgRouting notes” is a practical guide.
  2. Update database tables according to your specific requirement.

Install psycopgr

pip install psycopgr

(Yes, it is on PyPI now.)


pipenv install psycopgr

As you may guessed from the name, psycopgr uses psycopg2 as PostgreSQL driver. The above command will install it automatically.



from psycopgr import PgrNode, PGRouting

Create an PGRouting instance with database connection:

pgr = PGRouting(database='mydb', user='user')

Adjust meta datas of tables including the edge table properies if they are different from the default (only the different properties needs to be set), e.g.:

pgr.set_meta_data(cost='cost_s', reverse_cost='reverse_cost_s', directed=true)

This is the default meta datas:

    'table': 'ways',
    'id': 'gid',
    'source': 'source',
    'target': 'target',
    'cost': 'cost_s', # driving time in second
    'reverse_cost': 'reverse_cost_s', # reverse driving time in second
    'x1': 'x1',
    'y1': 'y1',
    'x2': 'x2',
    'y2': 'y2',
    'geometry': 'the_geom',
    'has_reverse_cost': True,
    'directed': True,
    'srid': 4326

Prepare nodes. nodes are represented by PgrNode namedtuple with geographic coordinates rather than vertex id (vid) in the tables. PgrNodes is defined as:

PgrNode = namedtuple('PgrNode', ['id', 'lon', 'lat'])

in which id could be None or self-defined value, and lon and lat are double precision values. Of course nodes could be input from various interfaces such as database or another program.

For example:

nodes = [PgrNode(None, 116.30150, 40.05500),
         PgrNode(None, 116.36577, 40.00253),
         PgrNode(None, 116.30560, 39.95458),
         PgrNode(None, 116.46806, 39.99857)]

Now we can do routings:

# many-to-many
routings = pgr.get_routes(nodes, nodes, end_speed=5.0, pgx_file='r.pgx')
# one-to-one
routings = pgr.get_routes(nodes[0], nodes[1])
# one-to-many
routings = pgr.get_routes(nodes[0], nodes)
# many-to-one
routings = pgr.get_routes(nodes, node[2])
  • end_speed: speed from node to nearest vertices on ways in unit km/h.
  • gpx_file: set it to output paths to a gpx file.

The returned is a dict of dict: {(start_node, end_node): {'path': [PgrNode], 'cost': cost}

By default, cost is traveling time along the path in unit second. It depends on the means of columns of the edge table that you set as cost and reverse_cost. You can assign the relations by set_meta_data function.

We can also get only costs without detailed paths returned:

costs = pgr.get_costs(nodes, nodes)

The returned is also a dict: {(start_node, end_node): cost}

Low-level wrapper of pgRouting functions

psycopgr function pgRouting function
dijkstra pgr_dijkstra
dijkstra_cost pgr_dijkstraCost
astar pgr_astar

These are direct wrappings of pgRouting functions. For example, dijkstra takes vertex ids as input. This list may be extended in the future.

Leave a Reply