Re: [Talk-us] Nominatim address lookup

2014-12-06 Thread Sarah Hoffmann
On Sat, Dec 06, 2014 at 01:11:28AM -0600, Paul Johnson wrote:
 Is it just me or did Nominatim stop checking an external database for
 address lookups where in-DB addresses aren't available in the US?

It never checked any external databases, it just uses house numbers
from Tiger where they are missing in OSM. That means that the address
down to street level still needs to be in OSM or the search fails.

Nothing about that has changed in the last two years or so (although
there was a Tiger data outage last weekend but we have been back to normal
since Monday)

Any partucular address you cannot find anymore?

Sarah

___
Talk-us mailing list
Talk-us@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-us


Re: [Talk-us] admin level for US states

2014-11-25 Thread Sarah Hoffmann
On Tue, Nov 25, 2014 at 10:29:25AM +0100, Martin Koppenhoefer wrote:
 2014-11-24 21:18 GMT+01:00 Minh Nguyen m...@nguyen.cincinnati.oh.us:
 
  Assuming this table reflects the actual state of the map, most countries
  have chosen 4 for their state equivalents.
 
 
 
 Actually, many countries do not have something like a state equivalent,
 it is a particularity of the USA because they are a federal republic.

admin_levels have been invented in order that different borders can be
rendered consistently among countries according to the wiki[1]. That's
also what I remember. State eqivalent doesn't mean that they must be
organised exactly in the same way but that they are roughly at the same
level of administrative hierarchies. Under that definition US states are
the same as German bundesländer, French regions, Canadian provinces etc.
even though their political influence and internal organzisation is
wildly different.

There is a lot of software around that works under the assumption that
US states (and the equivalents in other countries) can be found at 
admin_level=4. The current admin level hierarchy is not perfect but
it works for most practical applications. Please don't break it.
If you need to have a more find-grained distinction on how the
administrative units are organised, I suggest introducing a new tag
instead of changing the meaning of a well-established one.

Kind regards

Sarah


[1] http://wiki.openstreetmap.org/wiki/Admin_level#admin_level

___
Talk-us mailing list
Talk-us@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-us


Re: [Talk-us] New I.D Feature

2014-11-12 Thread Sarah Hoffmann
Hi,

On Sun, Nov 09, 2014 at 10:33:58AM -0500, Richard Welty wrote:
 1) the Census Bureau has an area based version of zip codes
 (postal codes to the non-US types) called ZCTA. it is not a
 complete representation, it covers about 30,000 of the
 50,000 unique zip codes, but it covers all the ones that
 can be reasonably envisioned as areas.

Nominatim is already using a version of the ZCTA data
as an additional source. I know there are some bad errors
in there preceisly because it seems that US zip codes
don't really cover areas. The search allows a certain
degree of fuzziness by accepting all nearby zip codes but
that is really not good enough. 

 2) missing from ZCTA is the mapping from zip codes to
 postal city as you call it. there is more to it than just a
 many-to-one mapping, which i'll address in a minute.
 
 3) most of the US mappers are opposed to importing this
 into OSM for a couple of good reasons, i'm one of those
 opposed.

I can see that there are good reasons not to import ZCTA.
It is just an estimate after all.

Until now I was under the impression that the postal cities
are different than postcodes in that they are well defined
areas, i.e. I thought that they are much more similar to
administrative boundaries just created by a different
branch of the government. But from your mail, it sounds
like they are much closer to the zip codes.

 4) however, there is also a movement in the US mapping
 community towards having certain types of data kept out
 of the core OSM database, including various types of
 admin boundaries. there are two projects looking at
 admin boundaries in this manner; i'm working on one of
 them.
 
 5) so if we could 1) come up with the ZCTA-City mappings
 and 2) provide them in a convenient external database for
 use by geocoders, is this something that might reasonably
 be made use of in Nominatim?
 
 i've been considering what a project to crowdsource
 the zip-city mappings for the US might look like. currently
 the data exists in OSM to handle about 4% of the mappings,
 but a maproulette style challenge might get us a lot of
 the rest.
 
 as for that many to one mapping that isn't, basically,
 for each zip code there is a primary city and potentially
 a number of secondary cities. the primary city is the
 city name of the post office that serves the routes; the
 secondary cities are generally traditional place names
 within the delivery area; for example, for years i lived
 in the Lansingburgh neighborhood of Troy NY, and
 the post office would deliver mail for either city name.
 any effort to crowd source this data would need to take
 care of that detail.

If I understand you right then the 'secondary cities' are
the actual cities/towns/villages already mapped as either
place nodes or administrative boundaries. Those are already
used by Nominatim.

So it seems the primary cities are what I was thinking of when
referring to postal cities and they can actually be inferred
from the postcode. If that is right, it should be somehow
possible to add that concept to Nominatim. I'd have to give
it a bit of thought.

Sarah

___
Talk-us mailing list
Talk-us@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-us


Re: [Talk-us] New I.D Feature

2014-11-08 Thread Sarah Hoffmann
Hi,

On Sat, Nov 08, 2014 at 02:05:44AM -0600, Toby Murray wrote:
 Nominatim actually does not correctly use addr:city. You are correct in
 that it does assume a better match between physical border and postal city
 address. What happens is it actually puts city information on *roads* based
 on boundary relations and place=* nodes. Then it links individual address
 points to the roads. The addr:city tags on POIs and building outlines is
 actually completely ignored. This leads to things like this address not
 being found if you search for it in Manhattan, KS:
 https://www.openstreetmap.org/node/2209430502
 
 On the other hand, this one on the next street over is found because I
 added an addr:city tag to the road:
 https://www.openstreetmap.org/node/2209427353
 road:
 https://www.openstreetmap.org/way/13226787
 
 I consider this to be a nominatim bug and haven't attempted to do a mass
 tagging of roads with addr:city because I see this as essentially tagging
 for the renderer.

Well, agreed that Nominatim still misses the feature to mix in addr:* tags.
It is somewhere on the todo list. To be honest, I'm not very motivated to
fix it for the sake of US postal towns because I'm not yet convinced that
the addr:city tag is the right solution for that concept.

Postal towns are by all technical means areas and you should really consider
tagging them as such. Let me just outline why addr:city is difficult from
a point of view of searching. Assume Nominatim would handle addr:city
correctly. That would alow you to find '6001 Stony Brook Drive, Manhattan'.
But what about 'Stone Brook Drive, Manhattan'. I expect you would want
to be able to find that too? That would only work if you added addr:city
to the road as well or Nominatim added some means to infer from the addresses
that the road is postal code-wise in Manhattan. It gets even more difficult
for the addresses that have not yet been mapped. If there are addr:city
tags on addresses in other streets nearby, then some dark magic might still
guess the expected postal town correctly but it still means you need to 
add some addresses first. Any tagging schema that implies that much guessing
really should be reconsidered.

Postal towns is really a concept that wasn't taken into account when the
Karlsruhe address schema was developped and IMHO doesn't really fit. I'd
really like to see some discussion in the US community about how it can
be tagged including considering the possibility to come up with a brand
new way of tagging it. If you then come to the conclusion that addr:city
is still the best way to go, so be it. But I'd be far more happy to
support something in Nominatim that fits exactly the US situation than
hack some European tagging system until it hopefully gives the right
result most of the time.

Kind regards

Sarah
(your friendly nominatim maintainer)



___
Talk-us mailing list
Talk-us@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk-us