libpostal integration

[klokan]

We may consider for future version to integrate libpostal into indexing step as well as into the search interface - to work with the normalised addresses and accept human input in various languages better.

https://github.com/geometalab/OSMNames/wiki/libpostal-overview

https://github.com/openvenues/libpostal

in our case integrated via python binding:
https://github.com/openvenues/pypostal

[klokan]

Target use: remove from query the house number.

[MartinMikita]

Libpostal is C library, which requires 2.1 GB data for parsing address.
Loading python wrapper (import libpostal) uses ~1.4 GB RAM with this data.

We have decided to not use this library directly, but only remove numbers from the query, if original query will not succeed.

libpostal changes are in the separate branch libpostal

[klokan]

Reopening - for implementation of the search in postal addresses.

[klokan]

There is libpostal 1.0:
https://medium.com/@albarrentine/statistical-nlp-on-openstreetmap-part-2-80405b988718
https://github.com/openvenues/libpostal

It has many new features and very good coverage now.
They claim it can run on 1GB or RAM.