Skip to content

Instantly share code, notes, and snippets.

What would you like to do?
# coding=utf-8
''' Guess at ISO-3166-2 admin 1 codes for Natural Earth data.
Uses subdivision names from Ola Holmström’s iso-3166-2:
ne_10m_admin_1_states_provinces.csv is a direct CSV
export of Natural Earth admin-1 data as of Nov 8, 2012:
Generates out.csv, which can be joined back to Natural Earth on adm1_code.
from unicodecsv import reader, DictReader, writer
from unidecode import unidecode
from operator import itemgetter
from itertools import groupby
from fuzzywuzzy import fuzz
from re import split
def unidecode_lower(string):
return unidecode(string).lower()
with open('olahol-iso-3166-2.csv') as file:
iso_3166_2_rows = list()
for row in reader(file):
# Afghanistan,AF-BDS,Badakhshān,Province,Badaẖšan,AF-BDS*,12267,2,AF,AFG
country, iso_3166_2, name, _, name_variants, _, _, _, iso_a2, _ = row
names = split(r'(?: \[|\])', name) + name_variants.split(', ')
names = set(map(unidecode_lower, filter(None, names)))
iso_3166_2_rows.append(dict(country=country, iso_3166_2=iso_3166_2,
name=name, name_variants=name_variants,
iso_a2=iso_a2, names=names))
with open('ne_10m_admin_1_states_provinces.csv') as file:
ne_admin1_rows = list()
iso_3166_1s = set()
for row in DictReader(file):
# adm1_code,iso_a2,iso_3166_2,name,name_alt,name_local
names = [row['name'], row['name_local']] + row['name_alt'].split('|')
row['names'] = set(map(unidecode_lower, filter(None, names)))
for iso_a2 in iso_3166_1s:
#if iso_a2 not in ('CM', 'RU', 'PL'): continue
# Pick out the Natural Earth entries for this country only missing an ISO-3166-2 code.
_ne_admin1_rows = [R for R in ne_admin1_rows if R['iso_a2'] == iso_a2 and R['iso_3166_2'].endswith('-')]
_iso_3166_2_rows = [R for R in iso_3166_2_rows if R['iso_a2'] == iso_a2]
print iso_a2
# Start with a high minimum cutoff, and work downwards
# so that exact matches are found as early as possible.
for minimum_ratio in range(100, 79, -5):
for ne_admin1_row in _ne_admin1_rows:
best_ratio, best_iso_row = minimum_ratio, None
for iso_3166_2_row in _iso_3166_2_rows:
for ne_name in ne_admin1_row['names']:
for iso_name in iso_3166_2_row['names']:
ratio = fuzz.ratio(ne_name, iso_name)
if ratio > best_ratio:
best_ratio, best_iso_row = ratio, iso_3166_2_row
if best_iso_row:
print best_ratio, ne_admin1_row['names'], best_iso_row['names'], best_iso_row['iso_3166_2']
ne_admin1_row['iso_3166_2'] = best_iso_row['iso_3166_2']
with open('out.csv', 'w') as file:
cols = 'adm1_code,iso_a2,iso_3166_2,name,name_alt,name_local'.split(',')
out = writer(file)
for row in ne_admin1_rows:
out.writerow([row[col] for col in cols])
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment