Nominatim/Version2
From OpenStreetMap Wiki
Contents |
Changes
Indexing
- Merge admin boundaries and place nodes (where within 2 levels of rank) (working in development version)
- Modify Voronoi algorithm to take account of known fixed boundaries (in progress)
- Native postgresql Voronoi
- Additional per-country index to speed up searches within country - might require better county polygons in OSM?
- Various problems with German postfixes / abbreviations (back ported to v1, being tested live)
- Performance - indexing is far too slow. Test clustering and polygon generation to see effects
- Move to hstore for name storage for compatibility
- Import minute change sets so index fully up to date
- Assign UUID to features. Ideally UUID should be consistent between nominatim systems! (is that even possible?!?)
- Generate indexes in memory
Search
- Re-write search logic in java.
- Use statistics to estimate phrase importance to improve ordering of search graph
- Change all scoring functions to a 0 to 1 scale so scores can be calculated using multiple factors
- Re-score strings after index search (index is lossy, we should recheck - perfect matches should beat transliterations)
- Options for fallback to high level feature when low level features missing (i.e. town when street not yet present)
- 'Did you means' Suggestions. Can't use standard - too many words and phrases (too slow) - so:
- High level admin features only?
- Modified levenshtein btree?
- Geographic indexing of suggestions?
- Word context based indexing?
- Allow detailed / structured search terms in query (i.e. query_country_code=DE&query_town=Berlin) (partial - query_country_code=[countrylist] only)
- Search for amenities by tag pair (partially done)
- Search graph to take account of in / nearest rather than just dropping them (partially done)
- Multilingual support for nearest / in / im (possible list of words in wiki (Name_finder:Translations) (in progress)
- Searching for street intersections
- Provide option to use postgresql full text search module as alternative indexing method
- Better formatting of address output (address format def on wiki? crowdsource?)
- Indexing support for autocomplete text entry (prefix search)
Code
Re-factor and reorganise the code - its a bit of a mess! (in progress)
Gazetteer Output
- Minor index change required to make this efficient, but requires complete re-index of planet (in progress)
- xml interface
- Visual representation of address 'graph' to help people debug
Export
- Provide export files (xml?) at various detail levels
- Structured? (in progress)
- Flat?
- Provide postgresql dump file to get people started?