gfdrr / admin-boundaries Goto Github PK

For the Think Hazard project, we are going to make sure world capitals from OSM can be 100% found. For this, we will use a combination of the GAUL dataset to replace the older admin boundaries and will try make use of geonames.org geocoder to help improve the performance of the search.

Improving search has to be done in a way that we don't just fix what we notice is broken. For this, we will implement cross-validation and split the test set 50/50.

• Create a script to download the osm capitals and put it in capitals.json
• Extract the english name and sort it alphabetically in capitals.txt
• Create a curated list of capitals from Wikipedia
• Compare the two capitals list and decide which one to use for testing (We will use the one from WIkipedia, it has more entries).
• Split capitals list in test and validation
• Measure the performance of the current ThinkHazard prototype with the test set (41% of 131 cities were not found).
• Pair geonames records with the current GAUL records to be able to use geonames.org search and a reverse match via the database.
• Measure the performance of the new prototype with the test set
• Tweak the prototype to improve the performance with the test set
• Measure the performance of the current ThinkHazard prototype with the validation test
• Measure the performance of the new prototype with the validation test
• Share the new prototype with the ThinkHazard team based on the validation results