Grasshopper Address Parser

Simple microservice for parsing US address strings into their component parts. This project is used by the Grasshopper geocoding service, but can also be used for general-purpose address parsing.

Requirements

• Flask - Python HTTP microframework
• US Address parsing library - Python library for parsing US addresses using natural language processing (NLP) methods.

Build and Deploy

grasshopper-parser is a standard Python app with few dependencies, making it quite simple to build and launch.

Locally

1. Build it.

   pip install -r requirements.txt

2. Run it.

   • Flask Dev Server

     python app.py

   • ...or Gunicorn

     gunicorn -c conf/gunicorn.py -b localhost:5000 app:app

Docker

The service can also be run as a Docker container. See the Dockerfile for details on the Docker image.

1. Build it.

   docker build --rm --tag=hmda/grasshopper-parser .

2. Run it.

   docker run -ti -p 5000:5000 hmda/grasshopper-parser

Configuration

Much of the parser's logic is maintained within rules.yaml. This file is read once at startup. If you wish to make changes, you will need to restart the API for the changes to take affect.

rules.yaml is composed of the following sections:

1. address_parts: Maps all address parts to their underlying data source.

   1. standard: Maps usaddress's "components" to the parser's address "parts". This is currently a one-to-one mapping, but may diverge from usaddress in the future. These are the default set of "parts" returned if no "profile" is given.

   2. derived: Adds additional composite parts, made by joining multiple "parts" from the standard mapping. These "parts" are only available when mapped to a given "profile".

2. profiles: Provides additional address part processing, such as returning "derived" address parts, and validating that a given address string can be parsed into the minimum "required" parts.

API Usage

The following resources are available. All examples assume running on localhost, port 5000.

/ (root resource)

Displays the current state of the API.

Request

GET http://localhost:5000/

Response

{
    "host": "yourhost.local",
    "service": "grasshopper-parser",
    "status": "OK",
    "time": "2015-05-06T19:14:19.304850+00:00",
    "upSince": "2015-05-06T19:08:26.568966+00:00"
}

/parse

The /parse resource is the heart the API. It parses free-text address strings into their component parts. It supports GET requests for individual with the following query parameters:

• address: Free-text address string to be parsed.

• profile: Additional parsing logic based on pre-defined "profiles"

  The "grasshopper" profile is included by default. It is geared towards the grasshopper geocoder's parsing requirement. Additional profiles can be added in rules.yaml.

Single address

Request

GET http://localhost:5000/parse?address=1600+Pennsylvania+Ave+NW+Washington+DC+20006

Response

The input value will always be the address string provided in the request.

{
  "input": "1600 Pennsylvania Ave NW Washington DC 20006",
  "parts": [
    { "code": "address_number", "value": "1600" }, 
    { "code": "street_name", "value": "Pennsylvania" }, 
    { "code": "street_name_post_type", "value": "Ave" }, 
    { "code": "street_name_post_directional", "value": "NW" }, 
    { "code": "city_name", "value": "Washington" }, 
    { "code": "state_name", "value": "DC" }, 
    { "code": "zip_code", "value": "20006" }
  ]
}

Single address with optional profile

Request

GET http://localhost:5000/parse?address=1600+Pennsylvania+Ave+NW+Washington+DC+20006?profile=grasshopper

Response

Notice there are two additional "derived" parts at the end of the list.

{
  "input": "1600 Pennsylvania Ave NW Washington DC 20006",
  "parts": [
    { "code": "address_number", "value": "1600" }, 
    { "code": "street_name", "value": "Pennsylvania" }, 
    { "code": "street_name_post_type", "value": "Ave" }, 
    { "code": "street_name_post_directional", "value": "NW" }, 
    { "code": "city_name", "value": "Washington" }, 
    { "code": "state_name", "value": "DC" }, 
    { "code": "zip_code", "value": "20006" }, 
    { "code": "address_number_full", "value": "1600" }, 
    { "code": "street_name_full", "value": "Pennsylvania Ave NW" }
  ]
}

Single incomplete address with profile

Request

GET http://localhost:5000/parse?address=1234+main+st&profile=grasshopper

Response

This is considered an error, and returned with a 400 HTTP status.

{
  "error": "Could not parse out required address parts: ['state_name', 'zip_code']",
  "statusCode": 400
}

Batch address parsing

Batch parsing of up to 5000 addresses is also supported via HTTP POST. The format is very similar to the URL structure using in parsing individual addresses.

Request

The following example contains a batch of 3 address strings, 2 good, 1 incomplete.

POST -H 'Content-Type: application/json' http://localhost:5000/parse

{
  "profile": "grasshopper", 
  "addresses": [
    "1600 Pennsylvania Ave NW Washington DC 20006", 
    "1315 10th St Sacramento CA 95814", 
    "1234 Main St"
  ]
}

Response

{
  "failed": [
    "1234 Main St"
  ],
  "parsed": [
    {
      "input": "1600 Pennsylvania Ave NW Washington DC 20006",
      "parts": [
        { "code": "address_number", "value": "1600" },
        { "code": "street_name", "value": "Pennsylvania" },
        { "code": "street_name_post_type", "value": "Ave" },
        { "code": "street_name_post_directional", "value": "NW" },
        { "code": "city_name", "value": "Washington" },
        { "code": "state_name", "value": "DC" },
        { "code": "zip_code", "value": "20006" },
        { "code": "address_number_full", "value": "1600" },
        { "code": "street_name_full", "value": "Pennsylvania Ave NW" }
      ]
    },
    {
      "input": "1315 10th St Sacramento CA 95814",
      "parts": [
        { "code": "address_number", "value": "1315" },
        { "code": "street_name", "value": "10th" },
        { "code": "street_name_post_type", "value": "St" },
        { "code": "city_name", "value": "Sacramento" },
        { "code": "state_name", "value": "CA" },
        { "code": "zip_code", "value": "95814" },
        { "code": "address_number_full", "value": "1315" },
        { "code": "street_name_full", "value": "10th St" }
      ]
    }
  ]
}

Testing

This project uses the Flask Testing tools, which really uses a Werkzeug Test Client under the hood. In addition, we're using nose as the unittesting framework, coverage for test coverage, flake8 for code quality, and tox to glue it all together.

Prerequisite

In order to use these tools, you will first need to install:

pip install -r tests/requirements.txt

Unit Testing

All unit tests are in the [tests]() directory. To run them:

$ nosetests -vs

...which should get you something similar to:

/explode - 500 on internal error ... ok
/parse - 400 with 'address' param is not present ... ok
/parse - 200 with just address, and parses correctly ... ok
...
/status - simple request ... ok

----------------------------------------------------------------------
Ran 11 tests in 0.089s

Test Coverage

If you'd like to include test coverage with your tests:

nosetests -vs --with-coverage --cover-package=app

...resulting in a report like:

Name    Stmts   Miss  Cover   Missing
-------------------------------------
app        67      2    97%   96, 127
----------------------------------------------------------------------
Ran 11 tests in 0.115s

Code Quality

flake8 is a wrapper code quality tools pep8, pyflakes, and mccabe. To run flake8:

flake8

To adjust the flake8 settings, edit the [flake8] section of tox.ini.

[flake8]
exclude = .git,.tox
format = pylint
max-line-length = 160
max-complexity = 10
show-source = 1

Do-it-all

To execute all of the above mentioned tools in one fell swoop, and simulate the full test suite executed by our Travis CI cfpb/grasshopper-parser, you can also use tox:

tox

All tox settings can be found in tox.ini.

Getting involved

For details on how to get involved, please first read out CONTRIBUTING guidelines.

Open source licensing info

1. TERMS
2. LICENSE
3. CFPB Source Code Policy