[OSM-talk] Skip geographical (redundant) address tags

Pierre-Alain Dorange pdorange at mac.com
Fri May 6 19:44:36 BST 2011


Maarten Deen <mdeen at xs4all.nl> wrote:

> > I was unaware I still had the country wrong for some places, I 
> > thought I'd
> > found and fixed all these.  Recalculating the street now produces the
> > right
> > result (as you can see if you re-do your search) so I'll do another
> > forced
> > update and try and get the last of these problems fixed.
> 
>  I now get two results, one is still "Jacob van Marisring, België" 
>  (maybe just a residual result?), the other is now "Jacob van Marisring,
>  Peel en Maas, Limburg, 5988KJ, Nederland".
>  While the location arrow is correct, the name is not. It displays the
>  municipality and not the town (Peel en Maas should be Helden) which I
>  think is caused by the missing admin_level=10 boundaries. It also 
>  displays the wrong zipcode. 5988KJ is the zipcode for Willem van 
>  Heukelomstraat [1], the correct zipcode would be 5988KG.
>  Where do you get the zipcode from? I have a few houses tagged with 
>  addr: keys, all of which I believe to have correct addr:postcode and
>  addr:street.

I could not really help you (i don't know your region, but  as i can see
(using JOSM) there is no admin relation in this area, so few clues to
help nomatim to find location.

An admin relation groups all boundary ways into a single object that
form a closed area. From that it's very easy for nominatim to find exact
adresses.
For example in my area (Cognac, France) adresses are correct aven on
borders.

http://open.mapquestapi.com/nominatim/v1/search.php?q=rue+du+buisson+mor
eau&viewbox=5.98%2C51.33%2C6%2C51.31

Their is 2 results (because just write the street name without the city)
: one in Corea and the other in France.
the second one (the good one of course) has all the correct admin info :
Champ de Foire (the suburb), Cognac (city defined by a relation),
Communauté de Communes de Cognac (city group defined by a relation),
Charentes (the county, defined by a relation), Poitou-Charentes (region,
defined by a relation) and of course France and the zip code.
It just add a strange "Les Ormeaux" (a place located 150 km away)

OpenMapQuest API let you examine the different areas and objects :
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=49125331

For example, Cognac city defined by an admin  relation is very precise
for Nomatim :
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=79399357

The suburb, defined by a node place is approximative and nomatim use an
estimation :
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=79399357

More precision, as Cognac street has adresses on building you can even
located a precise adress in a street : "28 buisson moreau, cognac"
return the exact place :
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=8938235

If we examine you request in Nominatim's OpenMapquest we got :
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=14534301

It show that Peel en Maas is an admin relation
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=79449231
It's topologicaly right, the street is in this area.

and that Helden (rejected by algo) is just a place village (so Nomatim
has to estimate the area with a baloon) :
http://open.mapquestapi.com/nominatim/v1/details.php?place_id=241938

And unfortunnally the ballon-estimation place the street half in, half
ou the ballon, so finilly Nomatim decide to reject... A boundary
relation will solve the problem.

Nominatim used admin relation, it's far more precise than any other
method.
-- 
Pierre-Alain Dorange
OSM experiences : <http://www.leretourdelautruche.com/map/>




More information about the talk mailing list