Awesome

Reverse Geocoder

A Python library for offline reverse geocoding. It improves on an existing library called reverse_geocode developed by Richard Penman.

UPDATE (15-Sep-16): v1.5.1 released! See release notes below.

About

Ajay Thampi | @thampiman | opensignal.com | ajaythampi.com

Features

Besides city/town and country code, this library also returns the nearest latitude and longitude and also administrative regions 1 and 2.
This library also uses a parallelised implementation of K-D trees which promises an improved performance especially for large inputs.

By default, the K-D tree is populated with cities that have a population > 1000. The source of the data is GeoNames. You can also load a custom data source so long as it is a comma-separated file with header (like rg_cities1000.csv), containing the following columns:

lat: Latitude
lon: Longitude
name: Name of place
admin1: Admin 1 region
admin2: Admin 2 region
cc: ISO 3166-1 alpha-2 country code

For usage instructions, see below.

Installation

For first time installation,

$ pip install reverse_geocoder

Or upgrade an existing installation using,

$ pip install --upgrade reverse_geocoder

Package can be found on PyPI.

Dependencies

scipy
numpy

Release Notes

v1.0 (27-Mar-15) - First version with support for only Python2
v1.1 (28-Mar-15) - Fix for issue #1 by Brandon
v1.2 (30-Mar-15) - Support for Python 3, conversion of Geodetic coordinates to ECEF for use in K-D trees to find nearest neighbour using the Euclidean distance function. This release fixes issues #2 and #8. Special thanks to David for his help in partly fixing #2.
v1.3 (11-Apr-15) - This release fixes issues #9, #10, #11 and #12. License has been changed from MIT to LGPL (see #12).
v1.4 (08-Jul-16) - Included numpy and scipy as dependencies in setup.
v1.5 (15-Sep-16) - Support for custom data source and fixes for issues #16 and #24. Hat tip to Jason and Gregoire.
v1.5.1 (15-Sep-16) - Fix for #26.

Usage

The library supports two modes:

Mode 1: Single-threaded K-D Tree (similar to reverse_geocode)
Mode 2: Multi-threaded K-D Tree (default)

import reverse_geocoder as rg

coordinates = (51.5214588,-0.1729636),(9.936033, 76.259952),(37.38605,-122.08385)

results = rg.search(coordinates) # default mode = 2

print results

The above code will output the following:

	[{'name': 'Bayswater', 
      'cc': 'GB', 
      'lat': '51.51116',
      'lon': '-0.18426', 
      'admin1': 'England', 
      'admin2': 'Greater London'}, 
     {'name': 'Cochin', 
      'cc': 'IN', 
      'lat': '9.93988',
      'lon': '76.26022', 
      'admin1': 'Kerala', 
      'admin2': 'Ernakulam'},
     {'name': 'Mountain View', 
      'cc': 'US', 
      'lat': '37.38605',
      'lon': '-122.08385', 
      'admin1': 'California', 
      'admin2': 'Santa Clara County'}]

If you'd like to use the single-threaded K-D tree, set mode = 1 as follows:

results = rg.search(coordinates,mode=1)

To use a custom data source for geocoding, you can load the file in-memory and pass it to the library as follows:

import io
import reverse_geocoder as rg

geo = rg.RGeocoder(mode=2, verbose=True, stream=io.StringIO(open('custom_source.csv', encoding='utf-8').read()))
coordinates = (51.5214588,-0.1729636),(9.936033, 76.259952),(37.38605,-122.08385)
results = geo.query(coordinates)

As mentioned above, the custom data source must be comma-separated with a header as rg_cities1000.csv.

Performance

The performance of modes 1 and 2 are plotted below for various input sizes.

Performance Comparison

Mode 2 runs ~2x faster for very large inputs (10M coordinates).

Acknowledgements

Major inspiration is from Richard Penman's reverse_geocode library
Parallelised implementation of K-D Trees is extended from this article by Sturla Molden
Geocoded data is from GeoNames