raphaellaude / us-addrs

Rust implementation of usaddress, package for parsing unstructured US address strings into address components.

us-addrs

US Addrs is a rust crate for parsing unstructured United States address strings into address components. It is a rust implementation of the awesome usaddress library. Thank you to the folks at datamade for releasing such a cool tool.

US Addrs is currently 79% (~5x) faster than usaddress, though additional optimizations should be possible. Accuracy is close to usaddress but not quite matching yet. The goal of this implementation is to faciliate use cases requiring better performance, such as geocoding large batches of addresses.

⚠️ This crate is under active development and may not match usaddress's accuracy. US Addrs will be better tested / documented shortly.

Examples

US Addrs can be run from the command line to parse an address

cargo run -- parse --address '33 Nassau Avenue, Brooklyn, NY'
    Finished dev [unoptimized + debuginfo] target(s) in 0.12s
     Running `target/debug/us_addrs parse --address '33 Nassau Avenue, Brooklyn, NY'`
[("33", "AddressNumber"), ("Nassau", "StreetName"), ("Avenue", "StreetNamePostType"), ("Brooklyn", "PlaceName"), ("NY", "StateName")]

or export a list of addresses to CSV

cargo run -- parse-file --file-path tests/test_data/test_addrs.txt test.csv

or by importing the crate and using the parse function

let addresses = read_to_string("tests/test_data/test_addrs.txt").expect("Could not read file");
let addresses: Vec<&str> = addresses.lines().collect();
for (i, addr) in addresses.iter().enumerate() {
    let parsed = parse(addr);
    println!("Test {}: Parsed address is {:?}", i + 1, parsed);
}

Test 1: Parsed address is [("123", "AddressNumber"), ("Elm", "StreetName"), ("St.", "StreetNamePostType"), ("Apt.", "OccupancyType"), ("4", "OccupancyIdentifier"), ("New", "PlaceName"), ("York", "PlaceName"), ("NY", "StateName"), ("10001", "ZipCode")]

etc...

Accuracy

Currently, the tool is very accurate for simple addresses, but mediocre at complex addresses. See test results:

running 4 tests
test test_simple_address_patterns ... ok
There were 34 mistagged address components of 782 (4.3%). 22 partially failed addresses of 103 (21.4%)
test test_labeled ... ok
There were 61 mistagged address components of 4627 (1.3%). 34 partially failed addresses of 687 (4.9%)
test test_us50_tagged ... ok
There were 151 mistagged address components of 20977 (0.7%). 139 partially failed addresses of 4120 (3.4%)
test test_synthetic_clean_osm_data ... ok