search

openvenues / libpostal

A C library for parsing/normalizing street addresses around the world. Powered by statistical NLP and open geo data.
MIT License
4.08k stars 419 forks source link

Numbered buildings in German addresses #247

Open Aknilam opened 7 years ago

Aknilam commented 7 years ago

For query: infanteriestr. Gebäude C the result is:

"house": "infanteriestr. gebäude c"

but the expected result is

"road": "infanteriestr.",
"house": "gebäude c"

For similar query: bahnhofstr. Gebäude C the result is proper:

"road": "bahnhofstr.",
"house": "gebäude c"

Checked on v1.1 and master using src/address_parser.

albarrentine commented 7 years ago

Hi @Aknilam, the v1.1 branch is work-in-progress and should not be used. Our releases take a little more time than in other software libraries because we have to first generate the planet-wide training data so the parser has sufficient examples of the patterns we want, then train the model on a billion examples, then upload to S3 and release it so people can use it. It's not just a simple code change to change behavior (or even if it is, switching branches doesn't help unless there's a new model). So v1.1 is not functional yet, and in fact performance will likely be worse when compiling on that branch because it will be using different code with a model which was trained on and expects 1.0. As such, only master should be used. When I have new models available on S3, I'll merge the new branch and release 1.1.

The 1.1 release will include this case (however, since things like "Gebäude C" are relatively rare in OSM, we will have to generate them like we do unit/floor numbers, and instead of "house" it will be labeled either as "house_number" or possibly a new tag for "building").