Skip to content

jsoma/NICAR20-geocoding-apis

Repository files navigation

Geocoding API hands-on session

Link to notebook or if you're a cheater (or slow typer!) you can skip to the filled-in version

and here are some secrets

Thanks to Joe Fox's geocoding repo

What is geocoding?

You give me an address, I give you a latitude/longitude pair.

Picking your geocoding option

  • Dozens: Do it manually
  • Hondreds-Thousands: Geocod.io if you're US-only
  • Tens of thousands, or international: See below!
  • Millions: do it yourself using Census TIGER data (one, two) if you're in the US

Different tools

There are many many geocoding APIs, all with different cover areas, costs, and APIs. To call out a few specifically:

  • Geocod.io isn't an API - you just upload a spreadsheet - and it's US/Canada only, but it's very easy to use.
  • Census Geocoder is very slow and a little finnicky, but super free.
  • HERE Geocoding and Search is not to be confused with the old HERE Geocoding API, although almost all documentation anywhere is for the old one
  • Google Maps requires you to jump through a hundred hoops to get it to work, but no matter what kind of trash you might feed it it'll usually give you a pretty good answer. Unfortunately you should read their policies, as they're pretty draconian. For example, you can't use Google's geocoder if you display the results on a non-Google map! I'm pretty sure no one follows that rule, but it's a great reason to try alternatives.
  • Once upon a time I used Baidu's geocoder and it was insanely blazing fast, but I definitely can't read the documentation so it isn't terribly useful for me.

The biggest issue is how many free you get. It changes over time, but it's typically 1000-2,5000. What if you have more than that? Usually people just spread it out over multiple days.

Basic use

Census Geocoder

From the docs:

https://geocoding.geo.census.gov/geocoder/locations/onelineaddress?address=4600+Silver+Hil
l+Rd%2C+Suitland%2C+MD+20746&benchmark=9&format=json

For multiple addresses, you need to send it a file! The file needs to be formatted a certain way (house number, street, etc, all split up).

You can test this out with the command-line tool curl if you're exceptionally fancy:

curl --form addressFile=localfile.csv \
  https://geocoding.geo.census.gov/geocoder/locations/addressbatch \
  --output geocoded.csv

Using geocoder:

> import geocoder
> g = geocoder.uscensus('455 Canal St, New Orleans, LA')
> g.ok
True
> g.latlng
[29.951624, -90.06652]

You can also use the Geographies endpoint if instead of just the lat/lon you're interested in city/state/etc.

HERE Geocoder

You'll need to create a REST app to create an API key.

https://geocode.search.hereapi.com/v1/geocode?q=5+Rue+Daunou%2C+75000+Paris%2C+France&apikey=YOUR_API_KEY

Unfortunately the Geocoder library uses the old version of HERE's geocoder.

Google Maps

Using the raw API:

https://maps.googleapis.com/maps/api/geocode/json?address=1600+Amphitheatre+Parkway,
+Mountain+View,+CA&key=YOUR_API_KEY

Using the Geocoder library:

> import geocoder
> g = geocoder.google('455 Canal St, New Orleans, LA', key=YOUR_API_KEY)
> g.ok
True
> g.latlng
[29.951624, -90.06652]

Gotchas

What is an address?

You might want to read falsehoods programmers believe about addresses, which has all sorts of weird addresses that you might encounter. It's more focused on how programmers like addresses to be input into their systems, but there's a chance it'll apply to you as well.

Formatting your data

Questions to ask and tests to run before you dedicate yourself to a geocoding solution:

  • How far does the free tier take you?
  • How well does it handle different abbreviations like N vs North? & vs AND?
  • Does it only geocode addresses, or does it accept intersections? Names of things?
  • Can you geocode an entire location at once, or do you need your data separated into separate fields? (address, stress and city in separate columns)
  • How well does it handle misspellings?
  • Usage restrictions (e.g. Google)

Accuracy

The biggest issue with geocoding is accuracy. And I'm not talking, "oh the geocoder is inaccurate it doesn't know where anything is," I'm talking about you being sure you know what level of accuracy you're getting.

While we think of geocoding and latitude/longitude pairs as representing exact places on the surface of the earth, that isn't always the case with geocoding.

  • If I geocode the United States, where will I end up? It won't say "that's too big of an area," it's probably going to put you somewhere in Kansas!
  • If I geocode New Orleans, where will I end up? It probably won't pick a meaningful political or cultural center, instead you'll get the center of the city's boundaries.
  • If it's a new address that seems to be between two existing addresses, where will the geocoder put you? It'll probably interpolate right in in the middle.
  • If the geocoder can match the zip code, city, state or country, but not the specific address, what does it do? Is the zip code center good enough? The city's center? The state?

For almost every geocoder, along with your latitude/longitude pair you get some measure of accuracy. Pay attention to this!!!!!!!! I can't tell you exactly what you're looking for, though: different geocoders report accuracy differently, and what you're doing with the coordinates might have different requirements (e.g. are you mapping the points individually, or are you just counting the number of locations in each state?).

Handling errors

Add-ons

If you're into Python, tqdm is a fantastic progress bar library.

Reverse geocoding

Tired of geocoding? There's also reverse geocoding! In the way that geocoding takes an address and gives you a lat/lon pair, for reverse geocoding you provide coordinates and receive location information instead.

About

Geocoding APIs repo for NICAR20 session

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published