[Geocoding] Fresh Nominatim different than nominatim.openstreetmap.org

[Sarah Hoffmann]

On Thu, Apr 10, 2014 at 11:07:09AM +0000, Anders Gunnarsson wrote:
> From: Sarah Hoffmann [mailto:lonvia at denofr.de]
> > On Thu, Apr 10, 2014 at 09:12:32AM +0000, Anders Gunnarsson wrote:
> > > We have made a few fresh installations of Nominatim lately according
> > > to the documentation on
> > > http://wiki.openstreetmap.org/wiki/Nominatim/Installation
> > >
> > > Our installations sometimes give different results than the official
> > on nominatim.openstreetmap.org. So far we have seen the following
> > differences:
> > >
> > >
> > > 1). Hits in Norway and Australia get the country code aq (Antarctica)
> > 
> > That's because people keep annexing Antarctica for their own countries
> > in OSM, see e.g. http://www.openstreetmap.org/relation/2978650
> > Nominatim is rather sensitive to this issue and responds by annexing
> > those countries for Antarctica.
> > 
> > I had an emergency patch installed on osm.org to prevent that from
> > happening but apparently forgot to backport it to the official version.
> > I'll try to get around that as soon as possible, feel free to file a bug
> > report as a reminder. Unfortunately, you will have to reindex the entire
> > country once the patch is installed.
> 
> Ok! Any chance I get a patch so I know more or less what I need to do?

The change is now pushed[1]. You can simply update to the latest version and
install the changes with: 
./utils/setup.php --enable-diff-updates --create-functions --create-partition-functions

You then have to find the place_id of the affected countries. They are in
the URL of the details page. Then log into your database and run:

select place_force_update(<place_id>);

followed by: ./utils/update --index --index-instances 2

This will take a very long time.

[1] https://github.com/twain47/Nominatim/commit/589bad809ef11bd8fe94ffa006e1255efec72480

> > > 2). Differences in importance
> > 
> > Have you installed the wikipedia data, as of
> > http://wiki.openstreetmap.org/wiki/Nominatim/Installation#Wikipedia_rank
> > ings
> 
> Wiki data has been downloaded and hopefully applied. A lot of places have the exact same numbers, but some places like Zagreb differs a lot.
> 
> How can I verify that the import of wiki data was successful?

The table wikipedia_article should be aroung 6GB.

> Is it easy to back trace the calculations for importance, to see what the importance data is based on??

There is an entry in the details page that states which wikipedia page
has been used. If that entry is missing, then importance is simply based
on the rank of the object.


> > > 3). Places not found using structured address search
> > >
> > > Query:
> > > http://nominatim.openstreetmap.org/search.php?format=json&city=tao%20y
> > > uan%20xian&country=taiwan
> > >
> > > 10 places found on nominatim.openstreetmap.org vs 0 on our server.
> > 
> > Check the data of the results on your server on the details page.
> > For example for the first result of the query, you should check:
> > 
> > http://<yourserver>/details.php?osm_type=N&osm_id=60655807
> > 
> > Is the node a city in Taiwan? Also, check that the search for Taiwan
> > works.
> 
> It's not, it's a county. Searching for county works in our setup. This search came from an old unit test, which turned out to be somewhat incorrect. I'm still puzzled why the result differs. Also result on osm.org finds the city the test was looking for.

Do I understand right? node 60655807 is marked as a county on the details page
on your server?

Sarah