[Geocoding] New geocoder test system
openstreetmap at brian.quinion.co.uk
Wed Sep 30 17:06:02 BST 2009
After several false starts the goecoding tool I've been working on
seems to have succeeded in indexing the full planet file and it is now
available for people to experiment with here:
Data is from 2 weeks ago (approx.) and will remain as is for a week or
so so people can try things after which I'll be using it as a test
system to try importing the daily diff files.
There are still several known problems:
1) No suggestion for miss-spelt words
I have code that produces suggestions but have not yet written the
interface for it.
2) Ordering of citys / towns by size
This is currently disabled because it is taking too long to index for
the planet. I hope to bring it back online shortly.
3) Filtering by viewbox
Currently broken and disabled (again)
4) Ordering of POI by distance
It finds the 10 nearest points (pub, station, whatever) but then
re-sorts them into an arbitrary order. It's an apples and oranges
problem caused by trying to order the points by a combination of
distance and importance and I'm still working on it.
At the moment the code only shows up to 10 results and there is no way
to get more.
6) wrong name translation shown
Where the feature has names in other languages but not in the local
language (i.e. 'name:cy', but no 'name:en' only 'name') it can end up
showing the wrong version of the name. This is because I've not been
able to find a list of default languages for each country - if anyone
knows of such a list this is easily fixed!
7) Repetition in search results
Hopefully this is mostly fixed however some repetition is still
present where a road is in multiple areas / postcodes or where the
data is actually duplicated in OSM (I've avoided hiding the OSM
duplication so that problems can be found and fixed rather than just
Please report other bugs to me either using the reporting tool built
into the site, by email or on IRC in #osm-geocoding. Before reporting
problems with addresses place check how the address was created using
the 'details' link next to each search result. The best way to
improve address information is to either create administrative
boundaries or convert places points into place areas (otherwise it has
to resort to 'nearest' which is always going to have problems).
In addition to HTML view results can also be retrieved in xml or json
format by adding 'format=xml' or format=json paramater to the url,
Feedback and bug reports, particularly regarding problems with
non-english searching appreciated.
More information about the Geocoding