City Autocomplete

A city autocomplete menu using the GeoNames datasets and the Algolia search engine

Implementation

This repository contains a process to generate a city index based on the GeoNames dataset. When uploaded to the Algolia platform, this index can be used to create a city autocomplete menu. The demo of the autocomplete menu using the generated index can be found here.

Output Format

The exported file ./data/export/city_index.csv contains the following columns:

Column name	Description	GeoNames Table
name	The city name (utf8)	geoname
alternatenames	The alternate names (comma separated)	geoname
country	The country name	countryInfo
postalcodes	The postalcode(s) (comma separated)	postalcodes
population	The city population	geoname
lat	latitude in decimal degrees (wgs84)	geoname
lng	longitude in decimal degrees (wgs84)	geoname

Usage

The data processing pipline is built with Docker. Specifically, it is built with Docker Compose thus allowing to define a multi-container architecture defined in a single file. To run through the city index generation process both need to be installed:

• Install Docker
• Install Docker Compose

In order to generate the city index the following steps need to be followed:

1. Clone the repository and change directory to it:

git clone https://github.com/manuelroth/city-autocomplete.git
cd city-autocomplete

2. Run the download-geonames command:

docker-compose run --rm download-geonames

This downloads the necessary GeoNames datasets, extracts and places it into the ./data/import folder -> Source Code

3. Now start up the database container (as a daemon process):

docker-compose up -d postgres

4. Import the GeoNames datasets into the PostgreSQL database:

docker-compose run --rm import-geonames

This takes the datasets previously placed into the ./data/import folder and imports them into the database -> Source Code

5. Generate and export the city index:

docker-compose run --rm export-city-index

This generates the city index and saves it as CSV-file in the ./data/export folder -> Source Code

(optional) 6. Upload the city-index to the Algolia platform:

docker-compose run --rm upload-city-index

Before uploading the city-index the environment variables APP_ID, API_KEY and INDEX_NAME in the docker-compose.yml file need to be updated -> Source Code

environment:
    APP_ID: <replace with your Algolia app_id>
    API_KEY: <replace with your Algolia app_key>
    INDEX_NAME: <replace with your Algolia index_name>

Used GeoNames datasets and relations between tabels

The city index contains data of different GeoNames datasets (Place data of all countries, Postalcodes of all countries and the Country Info). In order to combine the different datasets, a relation between them needs to be established. Datasets usually use id's for that purpose, but many times there are no id's which help to link the datasets. Therefore other methodes need to be applied.

For both the countryInfo and the postalcodes dataset I decided to link the dataset using one or multiple columns which have the same values. This works very well if the datasets are complete and there is enough information to be able to create a correct match between the tables.

This approach worked very well to create a relation between the geonames and the countryInfo tables. The country code column of both tables is complete. So the strings can be matched and a relation is established.

However the same approach did not work as seamless with the geonames and postalcodes tables. Both tables have columns for the city name, country code, 1. administrative division (state), 2. adm. division (county/province), 3. adm. division (community), latitude and longitude fields. As with every dataset, the geonames dataset is not perfect. There are many rows which do not have data for all the columns. Therefore it was not possible to use all the columns to create a relation between the two tables.

With manual testing I found out that in most cases it is suitable to create a relation using the country code, 1. administrative division (state) and the city name between the tables. These are the minimal required fields. Otherwise postalcodes get related to the wrong citys. The current solution is not perfect yet and should be further improved in the future.