Advanced use#

Snap origins and destination to the street network#

Make this Notebook Trusted to load map: File -> Trust Notebook

Sometimes, origin or destination points are far from the walkable, cyclable, or drivable street network. Especially when using a regular grid of points, many origins or destinations of a data set might be in the middle of a swamp (example above), on top of a mountain, or in the deep forest.

While r5py and R⁵ do their best to provide a reasonable route even for these points, at times, you might want to be able to control the situation a bit better.

R5py’s TransportNetwork allows you to snap a GeoSeries of points to points on the network.

Simply load a transport network, have a geo-data frame with origin or destination points, and call the transport network’s snap_to_network() method:

import geopandas

origins = geopandas.GeoDataFrame(
        "id": [1, 2],
        "geometry": [
            shapely.Point(24.841466, 60.208892),
            shapely.Point(24.848001, 60.207177),

origins["snapped_geometry"] = transport_network.snap_to_network(origins["geometry"])

id geometry snapped_geometry
0 1 POINT (24.84147 60.20889) POINT (24.84582 60.20980)
1 2 POINT (24.84800 60.20718) POINT (24.85189 60.20731)
Make this Notebook Trusted to load map: File -> Trust Notebook

By default, snapping takes into consideration all network nodes that support TransportMode.WALK, and that are within search radius of 1600 metres. In other words, points are snapped to the closest path that is accessible on foot, within a maximum of 1.6 kilometres.

Both parameters can be adjusted. For example, to snap to network nodes that are drivable, within 500 m, use the following code:

origins["snapped_geometry"] = transport_network.snap_to_network(
id geometry snapped_geometry
0 1 POINT (24.84147 60.20889) POINT EMPTY
1 2 POINT (24.84800 60.20718) POINT (24.85281 60.20762)

As you can see, one of the points could not be snapped with the tightened requirements: it was returned as an ‘empty’ point.

Convenient short-hands

Both TravelTimeMatrixComputer and DetailedItinerariesComputer support a convenient parameter, snap_to_network, that controls whether the origins and destinations should automatically be snapped to the transport network.

travel_time_matrix_computer = r5py.TravelTimeMatrixComputer(

Limit the maximum Java heap size (memory use)#

A Java Virtual Machine (JVM) typically restricts the memory usage of programs it runs. More specifically, the heap size can be limited (see this stackoverflow discussion for a detailed explanation).

The tasks carried out by R⁵ under the hood of r5py are fairly memory-intensive, which is why, by default, r5py allows the JVM to grant up to 80% of total memory to R⁵ (but ensures to always leave at least 2 GiB to the operating system and other processes).

You may want to lower this limit if you are running other tasks in parallel, or raise it if you have a dedicated computer with large memory and small operating system requirements.

To change the memory limit, you can either create a configuration file and set max-memory from there by specifying the --max-memory or -m command line arguments, or add the same arguments to sys.argv. See detailed explanation on the configuration page.

For instance, to set the maximum heap size to a fixed 12 GiB, you can create a configuration file in the location suitable for your operating system, and add the following line:

max-memory: 12G

Use a custom installation of R⁵#

For some use cases, it can be useful to use a local copy of R⁵, rather than the one downloaded by r5py, for instance, in order to apply custom patches to extend or modify R⁵’s functionality, or to force the use of a certain version for longitudinal comparability.

This can be achieved by passing a configuration option or command line argument to change the class path.

For example, to set a custom classpath inside a Python notebook, you can set sys.argv before importing r5py:

import sys
sys.argv.append(["--r5-classpath", "/opt/r5/"])
import r5py

To use the patched R⁵ version provided by the Digital Geography Lab on their GitHub pages, for example, pass the full URL, instead:

import sys
import r5py