peabiru (Python)

Multimodal routing on OpenStreetMap street networks and GTFS public transport: travel-time matrices, accessibility, and minute-by-minute (expanded) and arrival-time matrices, by walking, cycling, driving and public transport.

The routing engine is written in Rust and runs inside the Python process: no Java is needed, the network is built once in memory and reused by every call, and the results are pandas.DataFrames. The same Rust core powers the R package of the same name.

import datetime as dt
import peabiru

network = peabiru.build_network(peabiru.sample_data_path())
points = peabiru.sample_points()          # id, lon, lat, opportunities

ttm = peabiru.travel_time_matrix(
    network, points, points,
    mode=["WALK", "TRANSIT"],
    departure_datetime=dt.datetime(2019, 5, 13, 14, 0),
    percentiles=[25, 50, 75],
    seed=42,
)

access = peabiru.accessibility(
    network, points, points,
    opportunities_colnames="jobs",
    mode="BICYCLE",
    decay_function="logistic",
    cutoffs=[15, 30],
    decay_value=5,
)

Points are pandas.DataFrames with id, lon and lat (or, with geopandas installed, a GeoDataFrame of points in EPSG:4326); results are pandas.DataFrames in long format. What peabiru does not support yet (transit with bicycle or car access, fares, scenarios, elevation, ...) raises peabiru.UnsupportedError; nothing is silently ignored.

Install

pip install peabiru
pip install "peabiru[geo]"     # with geopandas

Requires Python 3.9 or later and pandas. The release builds wheels for Linux with glibc (x86_64 and aarch64), macOS (arm64 and x86_64) and Windows (x64), one wheel per platform for every Python from 3.9. There are no wheels for Linux with musl (Alpine) or for Windows on Arm: there pip builds the package from the source distribution, which needs a Rust toolchain (rustc 1.81 or later).

From a clone of the repository (needs a Rust toolchain):

cd python
maturin develop --uv --release

Development

cd python
uv venv && uv pip install pandas pytest geopandas
maturin develop --uv
.venv/bin/python -m pytest                         # tests, doc examples, doctests
uv pip install "mkdocs>=1.6" "mkdocs-material>=9.5" "mkdocstrings[python]>=0.26"
.venv/bin/mkdocs build --strict                    # documentation site in site/

Compatibility with r5r

The functions, their arguments, defaults and results follow the R package r5r 2.5.0: the engine is a Rust port of R5 7.6, the engine r5r drives, and is checked against r5r call for call, with identical results, including the last bit of accessibility values. The comparison (needs R with r5r 2.5.0 and a JDK 21) runs from the repository root:

JAVA_HOME=/path/to/jdk21 python/.venv/bin/python bench/compare/py_parity.py poa spo

Sample data

peabiru.sample_data_path() is a small network cut from r5r's Sao Paulo sample data. Its spo_osm.pbf contains map data (c) OpenStreetMap contributors, available under the Open Database License, and spo_gtfs.zip is the SPTrans GTFS feed shipped with r5r, without shapes.txt. Their sources are detailed in LICENSE-THIRD-PARTY.

License

MIT. peabiru ports R5 (MIT, Conveyal) and parts of Apache Commons Math (Apache-2.0); their notices and licence texts are in LICENSE-THIRD-PARTY, shipped with the package. The compiled extension module also contains code from the Rust crates pyo3, miniz_oxide and others (MIT, Apache-2.0 and similar licences); their licences are in LICENSE-RUST-CRATES, also shipped with the package.