fast-geodist 0.4.0

An implementation of Haversine distance calculations using Rust and PyO3

fast-geodist

An implementation of the Haversine formula for calculating Great Circle distance using Rust and PyO3.

This package was started as an experiment for learning how to implement functions in Rust and wrap them for use in python. The Haversine Formula is implemented following georust, with an array-wise implementation written using ndarray::parallel. It is wrapped for python using Maturin and integrates rust-numpy for array operations.

Installation

pip install fast-geodist

Quick start

The package provides two main entry points:

import numpy as np
from fast_geodist import haversine

# input either a tuple of (lat/lon, lat/lon)
result = haversine((1, 1, 0, 0))

# or a numpy array of coordinates pairs
# useful if lots of distances need to be calculated
# will be computed in parallel
result = haversine(np.array([(1, 1, 0, 0), (2, 2, 0, 0)]))

Benchmarks

The results of benchmarking show the rust implementation is 14x faster than the python implementation. This benchmark computes distances on an array containing 1,000,000 pairs of coordinates (see bench.py):

--------------------------------------------- benchmark: 2 tests ---------------------------------------------
Name (time in ms)              Min                   Max                  Mean                Median
--------------------------------------------------------------------------------------------------------------
test_benchmark_fast       164.0635 (1.0)        171.7663 (1.0)        168.0218 (1.0)        168.4129 (1.0)
test_benchmark_slow     2,335.4281 (14.23)    2,439.7850 (14.20)    2,395.6077 (14.26)    2,406.6356 (14.29)
--------------------------------------------------------------------------------------------------------------

Computed on an Intel i7-1165G7.

Development

Prequisites:

• Python (>=3.7) and make
• Rust toolchain

After cloning the repository, the Makefile includes helpful commands for setting up a development environment, linting, formatting, testing, and benchmarking. Get started as follows:

# setup a new virtual environment
python -m venv .venv
source .venv/bin/activate

# install the development dependencies
make install

# check other available commands
make help

Tooling:

• Cargo and Pytest are used for testing the Rust and Python code respectively (see /tests).
• Python code is linted using flake8 and formatted using Black; rust code with cargo fmt and cargo clippy.
• pre-commit is used to run these checks locally before files are pushed to git
• The Github Actions pipeline runs these checks and tests
• Semantic-release is used with conventional commits for automated releasing to PyPI

References

As well as the excellent PyO3 documentation, the following posts helped with creation of this package:

• https://blog.yossarian.net/2020/08/02/Writing-and-publishing-a-python-module-in-rust
• https://depth-first.com/articles/2020/08/10/python-extensions-in-pure-rust-with-pyo3/
• http://saidvandeklundert.net/learn/2021-11-18-calling-rust-from-python-using-pyo3/
• https://itnext.io/how-to-bind-python-numpy-with-rust-ndarray-2efa5717ed21