Skip to content
A fast, offline reverse geocoder in Python
Python Jupyter Notebook C++ CMake
Branch: master
Clone or download
Latest commit a81b409 Sep 16, 2016
Type Name Latest commit message Commit time
Failed to load latest commit information.
c++ Cleanup May 18, 2015
dist Fix for #26 Sep 15, 2016
notebook Pydata 2015 demo notebook added Jun 21, 2015
reverse_geocoder Fix for #26 Sep 15, 2016
.gitignore Need rg_cities1000.csv Sep 15, 2016
LICENSE v1.3 update. This update fixes #9, fixes #10, fixes #11 and fixes #12. Apr 11, 2015
MANIFEST Distribution files;cleanup;readme updates Mar 27, 2015 Fix for #26 Sep 15, 2016
README.txt v1.5 Sep 15, 2016
performance.png v1.2 released! This closes #2, #7 and #8. Mar 30, 2015
performance_logscale.png v1.2 released! This closes #2, #7 and #8. Mar 30, 2015
requirements.txt v1.5 Sep 15, 2016 Fix for #26 Sep 15, 2016 v1.2 released! This closes #2, #7 and #8. Mar 30, 2015

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.


Ajay Thampi | @thampiman | |


  1. Besides city/town and country code, this library also returns the nearest latitude and longitude and also administrative regions 1 and 2.
  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.


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.


  1. scipy
  2. numpy

Release Notes

  1. v1.0 (27-Mar-15) - First version with support for only Python2
  2. v1.1 (28-Mar-15) - Fix for issue #1 by Brandon
  3. 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.
  4. v1.3 (11-Apr-15) - This release fixes issues #9, #10, #11 and #12. License has been changed from MIT to LGPL (see #12).
  5. v1.4 (08-Jul-16) - Included numpy and scipy as dependencies in setup.
  6. v1.5 (15-Sep-16) - Support for custom data source and fixes for issues #16 and #24. Hat tip to Jason and Gregoire.
  7. v1.5.1 (15-Sep-16) - Fix for #26.


The library supports two modes:

  1. Mode 1: Single-threaded K-D Tree (similar to reverse_geocode)
  2. 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 = # 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 =,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.


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).


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


Copyright (c) 2015 Ajay Thampi and contributors. This code is licensed under the LGPL License.

You can’t perform that action at this time.