Skip to content

The country converter (coco) - a Python package for converting country names between different classification schemes.

License

Notifications You must be signed in to change notification settings

IndEcol/country_converter

Repository files navigation

country converter

The country converter (coco) is a Python package to convert and match country names between different classifications and between different naming versions. Internally it uses regular expressions to match country names. Coco can also be used to build aggregation concordance matrices between different classification schemes.

image image image image

image image image image

Motivation

To date, there is no single standard of how to name or specify individual countries in a (meta) data description. While some data sources follow ISO 3166, this standard defines a two and a three letter code in addition to a numerical classification. To further complicate the matter, instead of using one of the existing standards, many databases use unstandardised country names to classify countries.

The country converter (coco) automates the conversion from different standards and version of country names. Internally, coco is based on a table specifying the different ISO and UN standards per country together with the official name and a regular expression which aim to match all English versions of a specific country name. In addition, coco includes classification based on UN-, EU-, OECD-membership, UN regions specifications, continents and various MRIO and IAM databases (see Classification schemes below).

Installation

Country_converter is registered at PyPI. From the command line:

pip install country_converter --upgrade

The country converter is also available from the conda forge and can be installed using conda with (if you don't have the conda_forge channel added to your conda config add "-c conda-forge", see the install instructions here):

conda install country_converter

Alternatively, the source code is available on GitHub.

The package depends on Pandas; for testing pytest is required. For further information on running the tests see CONTRIBUTING.md.

Usage

Basic usage

Use within Python

Convert various country names to some standard names:

import country_converter as coco
some_names = ['United Rep. of Tanzania', 'DE', 'Cape Verde', '788', 'Burma', 'COG',
              'Iran (Islamic Republic of)', 'Korea, Republic of',
              "Dem. People's Rep. of Korea"]
standard_names = coco.convert(names=some_names, to='name_short')
print(standard_names)

Which results in ['Tanzania', 'Germany', 'Cabo Verde', 'Tunisia', 'Myanmar', 'Congo Republic', 'Iran', 'South Korea', 'North Korea']. The input format is determined automatically, based on ISO two letter, ISO three letter, ISO numeric or regular expression matching. In case of any ambiguity, the source format can be specified with the parameter 'src'.

In case of multiple conversion, better performance can be achieved by instantiating a single CountryConverter object for all conversions:

import country_converter as coco
cc = coco.CountryConverter()

some_names = ['United Rep. of Tanzania', 'Cape Verde', 'Burma',
              'Iran (Islamic Republic of)', 'Korea, Republic of',
              "Dem. People's Rep. of Korea"]

standard_names = cc.convert(names = some_names, to = 'name_short')
UNmembership = cc.convert(names = some_names, to = 'UNmember')
print(standard_names)
print(UNmembership)

In order to more efficiently convert Pandas Series, the pandas_convert() method can be used. The performance gain is especially significant for large Series. For a series containing 1 million rows a 4000x speedup can be achieved, compared to convert().

import country_converter as coco
import pandas as pd
cc = coco.CountryConverter()

some_countries = pd.Series(['Australia', 'Belgium', 'Brazil', 'Bulgaria', 'Cyprus', 'Czech Republic',
                  'Guatemala', 'Mexico', 'Honduras', 'Costa Rica', 'Colombia', 'Greece', 'Hungary',
                  'India', 'Indonesia', 'Ireland', 'Italy', 'Japan', 'Latvia', 'Lithuania',
                  'Luxembourg', 'Malta', 'Jamaica', 'Ireland', 'Turkey', 'United Kingdom',
                  'United States'], name='country')
 
iso3_codes = cc.pandas_convert(series=some_countries, to='ISO3')                  

Convert between classification schemes:

iso3_codes = ['USA', 'VUT', 'TKL', 'AUT', 'XXX' ]
iso2_codes = coco.convert(names=iso3_codes, to='ISO2')
print(iso2_codes)

Which results in ['US', 'VU', 'TK', 'AT', 'not found']

The not found indication can be specified (e.g. not_found = 'not there'), if None is passed for 'not_found', the original entry gets passed through:

iso2_codes = coco.convert(names=iso3_codes, to='ISO2', not_found=None)
print(iso2_codes)

results in ['US', 'VU', 'TK', 'AT', 'XXX']

Internally the data is stored in a Pandas DataFrame, which can be accessed directly. For example, this can be used to filter countries for membership organisations (per year). Note: for this, an instance of CountryConverter is required.

import country_converter as coco
cc = coco.CountryConverter()

some_countries = ['Australia', 'Belgium', 'Brazil', 'Bulgaria', 'Cyprus', 'Czech Republic',
                  'Denmark', 'Estonia', 'Finland', 'France', 'Germany', 'Greece', 'Hungary',
                  'India', 'Indonesia', 'Ireland', 'Italy', 'Japan', 'Latvia', 'Lithuania',
                  'Luxembourg', 'Malta', 'Romania', 'Russia', 'Turkey', 'United Kingdom',
                  'United States']

oecd_since_1995 = cc.data[(cc.data.OECD >= 1995) & cc.data.name_short.isin(some_countries)].name_short
eu_until_1980 = cc.data[(cc.data.EU <= 1980) & cc.data.name_short.isin(some_countries)].name_short
print(oecd_since_1995)
print(eu_until_1980)

All classifications can be directly accessed by:

cc.EU28
cc.OECD

cc.EU27as('ISO3')

and the classification schemes available:

cc.valid_class

There is also a method for only getting country classifications (thus omitting any grouping of countries):

cc.valid_country_classifications

If you rather need a dictionary describing the classification/membership use:

import country_converter as coco
cc = coco.CountryConverter()
cc.get_correspondence_dict('EXIO3', 'ISO3')

to also include countries not assigned within a specific classification use:

cc.get_correspondence_dict('EU27', 'ISO2', replace_nan='NonEU')

The regular expressions can also be used to match any list of countries to any other. For example:

match_these = ['norway', 'united_states', 'china', 'taiwan']
master_list = ['USA', 'The Swedish Kingdom', 'Norway is a Kingdom too',
               'Peoples Republic of China', 'Republic of China' ]

matching_dict = coco.match(match_these, master_list)

Country converter by default provides a warning to the python logging logger if no match is found. The following example demonstrates how to configure the coco logging behaviour.

import logging
import country_converter as coco
logging.basicConfig(level=logging.INFO)
coco.convert("asdf")
# WARNING:country_converter.country_converter:asdf not found in regex
# Out: 'not found'

coco_logger = coco.logging.getLogger()
coco_logger.setLevel(logging.CRITICAL)
coco.convert("asdf")
# Out: 'not found'

See the IPython Notebook (country_converter_examples.ipynb) for more information.

Command line usage

The country converter package also provides a command line interface called coco.

Minimal example:

coco Cyprus DE Denmark Estonia 4 'United Kingdom' AUT

Converts the given names to ISO3 codes based on matching the input to ISO2, ISO3, ISOnumeric or regular expression matching. The list of names must be separated by spaces, country names consisting of multiple words must be put in quotes ('').

The input classification can be specified with '--src' or '-s' (or will be determined automatically), the target classification with '--to' or '-t'.

The default output is a space separated list, this can be changed by passing a separator by '--output_sep' or '-o' (e.g -o '|').

Thus, to convert from ISO3 to UN number codes and receive the output as comma separated list use:

coco AUT DEU VAT AUS -s ISO3 -t UNcode -o ', '

The command line tool also allows to specify the output for none found entries, including passing them through to the output by passing None:

coco CAN Peru US Mexico Venezuela UK Arendelle --not_found=None

and to specify an additional data file which will overwrite existing country matching

coco Congo --additional_data path/to/datafile.csv

See https://github.com/IndEcol/country_converter/tree/master/tests/custom_data_example.txt for an example of an additional datafile.

The flags --UNmember_only (-u) and --include_obsolete (-i) restrict the search to UN member states only or extend it to also include currently obsolete countries. For example, the Netherlands Antilles were dissolved in 2010.

Thus:

coco "Netherlands Antilles"

results in "not found". The search, however, can be extended to recently dissolved countries by:

coco "Netherlands Antilles" -i

which results in 'ANT'.

In addition to the countries, the coco command line tool also accepts various country classifications (EXIO1, EXIO2, EXIO3, WIOD, Eora, MESSAGE, OECD, EU27, EU28, UN, obsolete, Cecilia2050, BRIC, APEC, BASIC, CIS, G7, G20). One of these can be passed by

coco G20

which lists all countries in that classification.

For the classifications covering almost all countries (MRIO and IAM classifications)

coco EXIO3

lists the unique classification names. When passing a --to parameter, a simplified correspondence of the chosen classification is printed:

coco EXIO3 --to ISO3

For further information call the help by

coco -h

Use in Matlab

Newer (tested in 2016a) versions of Matlab allow to directly call Python functions and libraries. This requires a Python version >= 3.4 installed in the system path (e.g. through Anaconda).

To test, try this in Matlab:

py.print(py.sys.version)

If this works, you can also use coco after installing it through pip (at the windows commandline - see the installing instruction above):

pip install country_converter --upgrade

And in matlab:

coco = py.country_converter.CountryConverter()
countries = {'The Swedish Kingdom', 'Norway is a Kingdom too', 'Peoples Republic of China', 'Republic of China'};
ISO2_pythontype = coco.convert(countries, pyargs('to', 'ISO2'));
ISO2_cellarray = cellfun(@char,cell(ISO2_pythontype),'UniformOutput',false);

Alternatively, as a long oneliner:

short_names = cellfun(@char, cell(py.country_converter.convert({56, 276}, pyargs('src', 'UNcode', 'to', 'name_short'))), 'UniformOutput',false);

All properties of coco as explained above are also available in Matlab:

coco = py.country_converter.CountryConverter();
coco.EU27
EU27ISO3 = coco.EU27as('ISO3');

These functions return a Pandas DataFrame. The underlying values can be access with .values (e.g.

EU27ISO3.values

I leave it to professional Matlab users to figure out how to further process them.

See also IPython Notebook (country_converter_examples.ipynb) for more information - all functions available in Python (for example passing additional data files, specifying the output in case of missing data) work also in Matlab by passing arguments through the pyargs function.

Building concordances for country aggregation

Coco provides a function for building concordance vectors, matrices and dictionaries between different classifications. This can be used in python as well as in matlab. For further information see (country_converter_aggregation_helper.ipynb)

Classification schemes

Currently the following classification schemes are available (see also Data sources below for further information):

  1. ISO2 (ISO 3166-1 alpha-2) - including UK/EL for Britain/Greece (but always convert to GB/GR)
  2. ISO3 (ISO 3166-1 alpha-3)
  3. ISO - numeric (ISO 3166-1 numeric)
  4. UN numeric code (M.49 - follows to a large extend ISO-numeric)
  5. A standard or short name
  6. The "official" name
  7. Continent: 6 continent classification with Africa, Antarctica, Asia, Europe, America, Oceania
  8. Continent_7 classification - 7 continent classification spliting North/South America
  9. UN region
  10. EXIOBASE 1 classification (2 and 3 letters)
  11. EXIOBASE 2 classification (2 and 3 letters)
  12. EXIOBASE 3 classification (2 and 3 letters)
  13. WIOD classification
  14. Eora
  15. OECD membership (per year)
  16. MESSAGE 11-region classification
  17. IMAGE
  18. REMIND
  19. UN membership (per year)
  20. EU membership (including EU12, EU15, EU25, EU27, EU27_2007, EU28)
  21. EEA membership
  22. Schengen region
  23. Cecilia 2050 classification
  24. APEC
  25. BRIC
  26. BASIC
  27. CIS (as by 2019, excl. Turkmenistan)
  28. G7
  29. G20 (listing all EU member states as individual members)
  30. FAOcode (numeric)
  31. GBDcode (numeric - Global Burden of Disease country codes)
  32. IEA (World Energy Balances 2021)
  33. DACcode (numeric - OECD Development Assistance Committee)
  34. ccTLD - country code top-level domains
  35. GWcode - Gledisch & Ward numerical codes as published in https://www.andybeger.com/states/articles/statelists.html
  36. CC41 - common classification for MRIOs (list of countries found in all public MRIOs)
  37. IOC - International Olympic Committee (IOC) country codes

Coco contains official recognised codes as well as non-standard codes for disputed or dissolved countries. To restrict the set to only the official recognized UN members or include obsolete countries, pass

import country_converter as coco
cc = coco.CountryConverter()
cc_UN = coco.CountryConverter(only_UNmember=True)
cc_all = coco.CountryConverter(include_obsolete=True)

cc.convert(['PSE', 'XKX', 'EAZ', 'FRA'], to='name_short')
cc_UN.convert(['PSE', 'XKX', 'EAZ', 'FRA'], to='name_short')
cc_all.convert(['PSE', 'XKX', 'EAZ', 'FRA'], to='name_short')

cc results in ['Palestine', 'Kosovo', 'not found', 'France'], whereas cc_UN converts to ['not found', 'not found', 'not found', 'France'] and cc_all converts to ['Palestine', 'Kosovo', 'Zanzibar', 'France'] Note that the underlying dataframe is available at the attribute .data (e.g. cc_all.data).

Data sources and further reading

Most of the underlying data can be found in Wikipedia, the page describing ISO 3166-1 is a good starting point. The page on the ISO2 codes includes a section "Imperfect Implementations" explaining the GB/UK and EL/GR issue. UN regions/codes are given on the United Nation Statistical Division (unstats) webpage. The differences between the ISO numeric and UN (M.49) codes are also explained at wikipedia. EXIOBASE, WIOD and Eora classification were extracted from the respective databases. For Eora, the names are based on the 'Country names' csv file provided on the webpage, but updated for different names used in the Eora26 database. The MESSAGE classification follows the 11-region aggregation given in the MESSAGE model regions description. The IMAGE classification is based on the "region classification map", for REMIND we received a country mapping from the model developers.

The membership of OECD and UN can be found at the membership organisations' webpages, information about obsolete country codes on the Statoids webpage.

The situation for the EU got complicated due to the Brexit process. For the naming, coco follows the Eurostat glossary, thus EU27 refers to the EU without UK, whereas EU27_2007 refers to the EU without Croatia (the status after the 2007 enlargement). The shortcut EU always links to the most recent classification. The EEA agreements for the UK ended by 2021-01-01 (which also affects Guernsey, Isle of Man, Jersey and Gibraltar). Switzerland is not part of the EEA but member of the single market.

The Global Burden of Disease country codes were extracted form the GBD code book available here.

Communication, issues, bugs and enhancements

Please use the issue tracker for documenting bugs, proposing enhancements and all other communication related to coco.

You can follow me on mastodon - @[email protected] and twitter to get the latest news about all my open-source and research projects (and occasionally some random retweets/toots).

Contributing

Want to contribute? Great! Please check CONTRIBUTING.md if you want to help to improve coco and for some pointer for how to add classifications.

Related software

The package pycountry provides access to the official ISO databases for historic countries, country subdivisions, languages and currencies. In case you need to convert non-English country names, countrynames includes an extensive database of country names in different languages and functions to convert them to the different ISO 3166 standards. Python-iso3166 focuses on conversion between the two-letter, three-letter and three-digit codes defined in the ISO 3166 standard.

If you are using R, you should have a look at countrycode.

Citing the country converter

Version 0.5 of the country converter was published in the Journal of Open Source Software. To cite the country converter in publication please use:

Stadler, K. (2017). The country converter coco - a Python package for converting country names between different classification schemes. The Journal of Open Source Software. doi: 10.21105/joss.00332

For the full bibtex key see CITATION

Acknowledgements

This package was inspired by (and the regular expression are mostly based on) the R-package countrycode by Vincent Arel-Bundock and his (defunct) port to Python (pycountrycode). Many thanks to Robert Gieseke for the review of the source code and paper for the publication in the Journal of Open Source Software.