Comment by richardw | Hacker News Reader

Comment by richardw | Hacker News Reader

What do you use for Geocoding? (for use in a web app that isn't open, so can't use Google etc.)

MapBox does a great job of the maps but the very next thing I want is to search for an address. Back to square one.

OSM uses MapQuest but I'm unsure how licensing and quality affects real-world apps. Is the open data good enough? Is there a sub-$2500 commercial offering that doesn't suck?

E.g. MapQuest's FAQ: http://developer.mapquest.com/web/tools/getting-started/plat...

How can I use geocodes?

Geocoding usage in the Community Edition on Licensed Data is limited to only the locations entered by an end user or that you have uploaded through our Data Manager tool. Geocodes under the Community Edition on Licensed Data cannot be stored for use outside the MapQuest service. If you would like more flexible options for geocode use, please contact MapQuest Platform Services to learn more about our extended rights geocoding product. Geocoding on Open Data has different terms, please check the Terms of Use for details.

Wikipedia has many:

http://en.wikipedia.org/wiki/Geocoding#List_of_geocoding_sys...

What do you use?

Slight confusion here:

OSM uses its own service, Nominatim, as tommorris has posted. It's OSM's own code using OSM data. The data and code are both free, but the osm.org servers have limited capacity; you can download the code and data and run it yourself if you like.

Mapquest Open runs an instance of this. MQ's servers are beefier so they have more liberal usage than osm.org's instance.

'Mainstream' Mapquest uses commercial data and therefore has pretty strict ToUs. Those are the ones you've quoted above: they don't apply to Mapquest Open.

I'm heading towards using the MapQuest Open myself, which seems to be a good default.

I think a heatmap of OSM quality would help a lot. Does such a thing exist?

There's not really a good metric of "OSM quality" yet.

The only things I can suggest are:

- issues: number reported, number closed, diversity of reporters and closers - on the basis that if lots ofpeople are reporting issues, that means "many eyes make all bugs shallow" might be working. If lots of people are closing them, then someone is actually paying some attention.

- tag diversity. Human mappers tend to actually use a fairly wide variety of tags for points of interest.

- some measure of how much the data that has been imported from, say, government public domain GIS datasets, has been modified.

- last update

- presence of house/building names and numbers

There's a bunch of similar stuff up at http://wiki.openstreetmap.org/wiki/Quality_assurance

In the U.S., there are various efforts to compare the OSM data to TIGER data from after Census started their big quality improvement project. They go a long ways towards showing where people are actually working on OSM (which has a rough correlation with quality).

> you can download the code and data and run it yourself if you like.

Check out Gisgraphy:

http://www.gisgraphy.com/

You install it and basically get a local Nominatim.

OpenStreetMap provides a geocoding service: Nominatim.

https://wiki.openstreetmap.org/wiki/Nominatim

It's a long way from perfect but will get better the more data people plonk into OSM. There's a big focus in a lot of local OSM communities on getting street numbering data.

Nominatim data is the same license as OSM, and if you really need to, you can run your own Nominatim instance.

They do, but it seems a poor cousin (rate limiting, no service guarantees, only open data) of the MapQuest Nominatim service that it's named for (or is that the other way around?), which leads back to the same issues of quality vs price.

I'd love to use OSM for all the usual reasons, just asking what it's like in real life. What do the zillions of startups out there use, what are the tradeoffs you've experienced? If it were that easy an answer there would only be OSM and the commercial providers would only have value-added data.

Google still has by far and away the best (read: most forgiving) geocoder. It'll understand "W4" as "West 4th Street" in New York and as Chiswick in London. Stuff like that is incredibly hard to get right as every country/city/street has so many permutations.

I've love to see an open-source alternative that rivals it, but haven't found one yet.

Its geocoder still doesn't handle leading zeroes to the best of my knowledge, which makes Southwest Portland a bit tough.

A description for the uninitiated: Portland has five quadrants in street names, NW/NE/N/SW/SE. SW has "negative block numbers" east of a certain line, which are represented by a significant leading zero in the building number.

To be fair, I don't think anyone handles this. Thankfully I'm not in SW pretty much ever.

Reply on news.ycombinator.com