searxng/utils/fetch_currencies.py
2016-07-08 23:43:28 +07:00

161 lines
4.2 KiB
Python

# -*- coding: utf-8 -*-
import json
import re
import unicodedata
import string
from urllib import urlencode
from requests import get
languages = {'de', 'en', 'es', 'fr', 'hu', 'it', 'nl', 'jp'}
url_template = 'https://www.wikidata.org/w/api.php?action=wbgetentities&format=json&{query}&props=labels%7Cdatatype%7Cclaims%7Caliases&languages=' + '|'.join(languages)
url_wmflabs_template = 'http://wdq.wmflabs.org/api?q='
url_wikidata_search_template='http://www.wikidata.org/w/api.php?action=query&list=search&format=json&srnamespace=0&srprop=sectiontitle&{query}'
wmflabs_queries = [
'CLAIM[31:8142]', # all devise
]
db = {
'iso4217' : {
},
'names' : {
}
}
def remove_accents(data):
return unicodedata.normalize('NFKD', data).lower()
def normalize_name(name):
return re.sub(' +',' ', remove_accents(name.lower()).replace('-', ' '))
def add_currency_name(name, iso4217):
global db
db_names = db['names']
if not isinstance(iso4217, basestring):
print "problem", name, iso4217
return
name = normalize_name(name)
if name == '':
print "name empty", iso4217
return
iso4217_set = db_names.get(name, None)
if iso4217_set is not None and iso4217 not in iso4217_set:
db_names[name].append(iso4217)
else:
db_names[name] = [ iso4217 ]
def add_currency_label(label, iso4217, language):
global db
db['iso4217'][iso4217] = db['iso4217'].get(iso4217, {})
db['iso4217'][iso4217][language] = label
def get_property_value(data, name):
prop = data.get('claims', {}).get(name, {})
if len(prop) == 0:
return None
value = prop[0].get('mainsnak', {}).get('datavalue', {}).get('value', '')
if value == '':
return None
return value
def parse_currency(data):
iso4217 = get_property_value(data, 'P498')
if iso4217 is not None:
unit = get_property_value(data, 'P558')
if unit is not None:
add_currency_name(unit, iso4217)
labels = data.get('labels', {})
for language in languages:
name = labels.get(language, {}).get('value', None)
if name != None:
add_currency_name(name, iso4217)
add_currency_label(name, iso4217, language)
aliases = data.get('aliases', {})
for language in aliases:
for i in range(0, len(aliases[language])):
alias = aliases[language][i].get('value', None)
add_currency_name(alias, iso4217)
def fetch_data(wikidata_ids):
url = url_template.format(query=urlencode({'ids' : '|'.join(wikidata_ids)}))
htmlresponse = get(url)
jsonresponse = json.loads(htmlresponse.content)
entities = jsonresponse.get('entities', {})
for pname in entities:
pvalue = entities.get(pname)
parse_currency(pvalue)
def add_q(i):
return "Q" + str(i)
def fetch_data_batch(wikidata_ids):
while len(wikidata_ids) > 0:
if len(wikidata_ids) > 50:
fetch_data(wikidata_ids[0:49])
wikidata_ids = wikidata_ids[50:]
else:
fetch_data(wikidata_ids)
wikidata_ids = []
def wdq_query(query):
url = url_wmflabs_template + query
htmlresponse = get(url)
jsonresponse = json.loads(htmlresponse.content)
qlist = map(add_q, jsonresponse.get('items', {}))
error = jsonresponse.get('status', {}).get('error', None)
if error != None and error != 'OK':
print "error for query '" + query + "' :" + error
fetch_data_batch(qlist)
def wd_query(query, offset=0):
qlist = []
url = url_wikidata_search_template.format(query=urlencode({'srsearch': query, 'srlimit': 50, 'sroffset': offset}))
htmlresponse = get(url)
jsonresponse = json.loads(htmlresponse.content)
for r in jsonresponse.get('query', {}).get('search', {}):
qlist.append(r.get('title', ''))
fetch_data_batch(qlist)
## fetch ##
for q in wmflabs_queries:
wdq_query(q)
# static
add_currency_name(u"euro", 'EUR')
add_currency_name(u"euros", 'EUR')
add_currency_name(u"dollar", 'USD')
add_currency_name(u"dollars", 'USD')
add_currency_name(u"peso", 'MXN')
add_currency_name(u"pesos", 'MXN')
# write
f = open("currencies.json", "wb")
json.dump(db, f, indent=4, encoding="utf-8")
f.close()