Confidence levels in the geocoder

[Fred R]

I used the dstk geocoder to bulk geocode a lot of data and I'm wondering how accurate the result is. Can someone explain what the confidence level number that is being returned mean in this case? How reliable is the OSM data? If the confidence level is high (let's say over 0.645) would you say that the lat/lng should be fairly accurate. 

Thanks,

Fred

[Pete Warden]

I don't have a great quantitative answer on the accuracy of the TIGER and OSM data I'm using. Anecdotally it's usually within 100m or so for urban US areas, but I've actually set up a new project to produce a more scientific answers on the accuracy:

http://petewarden.typepad.com/searchbrowser/2013/02/how-good-are-our-geocoders.html

That will help me assign some accuracy ranges to the confidence levels eventually, but right now that number's somewhat arbitrary. In the US it's based on Schuyler Erle/GeoIQ's geocoder gem, from the score value documented here:

https://github.com/geocommons/geocoder/blob/2500432a06d34ea726ec8b86c703430c8224b5d0/REST.rdoc

"The percentage of text match between the given address and the geocoded

    result, expressed as a float between 0 and 1. A higher score indicates

    a closer match. Results with a score below 0.5 should be regarded with

    care."

--
You received this message because you are subscribed to the Google Groups "dstk-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dstk-users+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

--

Jetpac

You should get the app for your iPad!

[Pete Warden]

ps, I'm looking for contributions to the test addresses, so if you want to add a few representative addresses you care about for your application, I'd love to get a push request with additions to https://github.com/petewarden/geocodetest/blob/master/testinput.txt 

Currently-unsupported, international, messy and spelling mistakes welcome, as long as they're real user-entered strings!

cheers,

           Pete

[Fred R]

Thanks for the reply. Clears things up.

FR