This repo contains the documentation of the python french-cities package, a package aimed at improving the referencing of municipalities in French 🇫🇷 datasets.

A full documentation with usecases is available at https://tgrandje.github.io/french-cities/. Obviously, it is only available in french as yet. Any help is welcome to build a multi-lingual documentation website.

Until then, a basic english documentation will stay available in the present README.

Do you have any data:

• which municipal locations are provided through approximate addresses, or via geographical 🗺️ coordinates?
• which municipalities are referenced by their postal codes and their labels 😮?
• which departments are written in full text 🔡?
• which spelling are dubious (for instance, torturing the Loire Loir-et-Cher) or obsolete (for instance, referencing Templeuve, a city renamed as Templeuve-en-Pévèle since 2015)?
• or compiled over the years and where cities' codes are a patchwork of multiple 🤯 vintages?

Then 'french-cities' is for you 🫵!

pip install french-cities

french-cities uses pynsee under the hood. For it to work, you need to set the credentials up. You can set up to four environment variables:

• insee_key
• insee_secret,
• http_proxy (if accessing web behind a corporate proxy)
• https_proxy (if accessing web behind a corporate proxy)

Please refer to pynsee's documentation to help configure the API's access.

Note that setting environment variable for proxy will set it for both pynsee and geopy.

Note that pynsee and geopy use their own web session. Every Session object you will pass to french-cities will NOT be shared with pynsee or geopy. This explains the possibility to pass a session as an argument to french-cities functions, even if you had to configure the corporate proxy through environment variables for pynsee and geopy.

french-cities can retrieve departement's codes from postal codes, official (COG/INSEE) codes or labels.

Working from postal codes will make use of the BAN (Base Adresse Nationale) and should return correct results. The case of "Cedex" codes is only partially covered by the BAN, so OpenDataSoft's API, constructed upon Christian Quest works. This consumes the freemium API and no authentication is included: the user of the present package should check the current API's legal terms directly on OpenDataSoft's website.

Working from official codes may sometime give empty results (when working on an old dataset and with cities which have changed of departments, which is rarely seen). This is deliberate: it will mostly use the first characters of the cities' codes (which is a fast process and 99% accurate) instead of using an API (which is lengthy though foolproof).

from french_cities import find_departements
import pandas as pd

df = pd.DataFrame(
    {
        "code_postal": ["59800", "97133", "20000"],
        "code_commune": ["59350", "97701", "2A004"],
        "communes": ["Lille", "Saint-Barthélémy", "Ajaccio"],
        "deps": ["59", "977", "2A"],
    }
)
df = find_departements(df, source="code_postal", alias="dep_A", type_field="postcode")
df = find_departements(df, source="code_commune", alias="dep_B", type_field="insee")
df = find_departements(df, source="communes", alias="dep_C", type_field="label")

print(df)

For a complete documentation on find_departements, please type help(find_departements).

french-cities can retrieve cities' codes from multiple fields. It will work out basic mistakes (up to a certain limit).

The columns used by the algorithm can be (in the order of precedence used by the algorithm):

• 'x' and 'y' (in that case, epsg must be explicitly given);
• 'postcode' and 'city'
• 'address', 'postcode' and 'city'
• 'department' and 'city'

Note that the algorithm can (and will) make errors using xy coordinates on a older vintage (ie different from the current one) in the case of historic splitting of cities (the geographic files are not vintaged yet).

The lexical (postcode, city, address, departement) recognition is based on a python fuzzy matching, the BAN API(base adresse nationale) or the Nominatim API of OSM (if activated). The algorithm won't collect underscored results, but failures may still occure.

from french_cities import find_city
import pandas as pd

df = pd.DataFrame(
    [
        {
            "x": 2.294694,
            "y": 48.858093,
            "location": "Tour Eiffel",
            "dep": "75",
            "city": "Paris",
            "address": "5 Avenue Anatole France",
            "postcode": "75007",
            "target": "75056",
        },
        {
            "x": 8.738962,
            "y": 41.919216,
            "location": "mairie",
            "dep": "2A",
            "city": "Ajaccio",
            "address": "Antoine Sérafini",
            "postcode": "20000",
            "target": "2A004",
        },
        {
            "x": -52.334990,
            "y": 4.938194,
            "location": "mairie",
            "dep": "973",
            "city": "Cayenne",
            "address": "1 rue de Rémire",
            "postcode": "97300",
            "target": "97302",
        },
        {
            "x": np.nan,
            "y": np.nan,
            "location": "Erreur code postal Lille/Lyon",
            "dep": "59",
            "city": "Lille",
            "address": "1 rue Faidherbe",
            "postcode": "69000",
            "target": "59350",
        },
    ]
)
df = find_city(df, epsg=4326)

print(df)

For a complete documentation on find_city, please type help(find_city).

Note : to activate geopy (Nominatim API from OpenStreeMap) usage in last resort, you will need to use the argument use_nominatim_backend=True.

french-cities can try to project a given dataframe into a set vintage, starting from an unknown vintage (or even a non-vintaged dataset, which is often the case).

Error may occur for splitted cities as the starting vintage is unknown (or inexistant).

In case of a known starting vintage, you can make use of INSEE's projection API (with pynsee). Note that this might prove slower as each row will have to induce a request to the API (which allows up to 30 requests/minute).

Basically, the algorithm of french-cities will try to see if a given city code exists in the desired vintage:

• if yes, it will be kept (we the aforementionned approximation regarding restored cities);
• if not, it will look in older vintages and make use of INSEE's projection API.

This algorithm will also:

• convert communal districts' into cities' codes;
• convert delegated or associated cities' codes into it's parent's.

from french_cities import set_vintage
import pandas as pd

df = pd.DataFrame(
    [
        ["07180", "Fusion"],
        ["02077", "Commune déléguée"],
        ["02564", "Commune nouvelle"],
        ["75101", "Arrondissement municipal"],
        ["59298", "Commune associée"],
        ["99999", "Code erroné"],
        ["14472", "Oudon"],
    ],
    columns=["A", "Test"],
    index=["A", "B", "C", "D", 1, 2, 3],
)
df = set_vintage(df, 2023, field="A")
print(df)

For a complete documentation on set_vintage, please type help(set_vintage).

french-cities makes use of multiple APIs. Please read :

• documentation (in french) on API Adresse
• documentation (in french) on OpenDataSoft API
• Nominatim Usage Policy

In case of bugs, please open an issue on the repo.

Any help is welcome.

Thomas GRANDJEAN (DREAL Hauts-de-France, service Information, Développement Durable et Évaluation Environnementale, pôle Promotion de la Connaissance).

GPL-3.0-or-later

Stable.