[Geocoding] Using Nominatim to geocode Toronto addresses

[Sarah Hoffmann]

Hi Stewart,

On Tue, May 28, 2013 at 03:33:32PM -0400, Stewart C. Russell wrote:
> I was trying to use Nominatim (Mapquest's server) to geocode the event
> locations for last weekend's Doors Open Toronto(XML feed:
> http://wx.toronto.ca/inter/culture/doorsopen.nsf/DOT_open_data.xml)
> using Geopy's OpenMapQuest geocoder. The addresses are clearly
> human-entered; here are some examples:
> 
>   * 255 Spadina Rd
>   * 1 King St W Floor : Lobby, 2nd, 12th
>   * 350 King St W Floor : 4
>   * 95 Front Street East St E
> 
> Even after adding ", Toronto, Ontario, Canada" to the query, I'd still
> get unreliable results. Some of them returned no results. Some returned
> obviously wrong results 

> (such as just under 1/5 of the addresses were
> returned as "King's College Circle", even simple ones like '1451 Avenue
> Rd').

Not sure what you mean here. Searching for '1451 Avenue Rd, Toronto'
looks good for me. It returns:
"Havergal College, 1451, Avenue Road, Lawrence Park, Toronto"

(osm.org and Mapquest both return the same results)

> Toronto's got some pretty good address data in OSM. Am I naïve expecting
> Nominatim to be able to find locations for hand-keyed addresses? How can
> I improve the addressing so that Nominatim finds things better?

It looks like you are running into problems because Nominatim always
insists to use everything from the address that you ask for. If things
like floor or room number are added, it gets confused and finds nothing.
For above examples, the following queries return results that look sensible
(at least for somebody far away from Toronto):

   * 255 Spadina Rd, Toronto
   * 1 King St W, Toronto
   * 350 King St W, Toronto
   * 95 Front Street East, Toronto

You could try to somehow truncate the addresses a bit if you don't get a
decent result. It helps sometimes. There is little on the server side
that can be done in a short term but the issue is known.

Sarah