Loading your files

Your download page lists the files your purchase includes. Find each file's format below. Every example runs as written with DuckDB (the command-line tool, or the duckdb Python package); replace the file name with yours. Download links last 24 hours; re-issue one from your account page. Column meanings are in the data dictionary.

CSV (gzip)

UTF-8, a header row, RFC 4180 quoting, gzip-compressed. DuckDB and most tools read the .gz directly; there is no need to unzip it first.

Count rows and look at a few (DuckDB CLI)

SELECT count(*) AS rows FROM read_csv('brand.csv.gz', header = true);
SELECT poi_id, name FROM read_csv('brand.csv.gz', header = true) LIMIT 5;

Load into Python with the duckdb package

import duckdb
con = duckdb.connect()
rows = con.sql("SELECT poi_id, name FROM read_csv('brand.csv.gz', header = true)").fetchall()
print(len(rows))

CSV in several parts

Very large datasets arrive as ordered parts of one file. Keep every part in one folder and read them together with a wildcard; each part carries the header row.

Read every part as one table (DuckDB CLI)

SELECT count(*) AS rows FROM read_csv('full-us-master-part-*.csv.gz', header = true, union_by_name = true);

Parquet

Columnar and typed. The fastest way into a warehouse or a dataframe; column types come with the file.

Count rows and look at a few (DuckDB CLI)

SELECT count(*) AS rows FROM read_parquet('brand.parquet');
SELECT poi_id, name FROM read_parquet('brand.parquet') LIMIT 5;

Load into Python with the duckdb package

import duckdb
con = duckdb.connect()
rows = con.sql("SELECT poi_id, name FROM read_parquet('brand.parquet')").fetchall()
print(len(rows))

Parquet in several parts

Ordered parts of one dataset. Read them together with a wildcard.

Read every part as one table (DuckDB CLI)

SELECT count(*) AS rows FROM read_parquet('full-us-master-part-*.parquet');

Building geometry (.geo.parquet)

The optional building-footprint add-on ships as its own Parquet file beside the data. Read it like any Parquet file.

Look at the geometry file (DuckDB CLI)

SELECT count(*) AS rows FROM read_parquet('brand.geo.parquet');

NOTICE.txt

The attribution notice for your licence tier. Keep it with the data wherever the data goes; your licence requires it.