[Talk-us] US Addressing

[andrzej zaborowski]

On 29 November 2012 21:12, Toby Murray <toby.murray at gmail.com> wrote:
> Now for the hard part. Converting and conflating the information with
> the non-trivial number of addresses I have already collected on the
> ground.

Compared to conflating names or geometries, addresses are not a
problem because the street and the housenumber form a unique id.
Before comparing them it is worth splitting the street name into
words, reordering the words within the name to order them lexically,
and abbreviating them using a not-necessarily-perfect word list (such
as that from tiger or nominatim), to account for variations.  We've
had to do that for one city recently but it turned out to be a simple
check.

The (150 loc) conversion script for that data would read 4 files:
* the new addresses in a particular format,
* an output of an overpass/xapi query for elements tagged with
adds:housenumber=*
* an output of an overpass/xapi query for ways tagged with building=*
* an output of an overpass/xapi query for named highways.

it would output:
* an .osm file adding the addresses missing from OSM, either attached
to existing buildings or added as nodes
* a list of potentially missing streets who's names appeared in the addresses.

Cheers