OpenAddressesUK / sorting_office

A Sinatra app that takes an address string and breaks it into its constituent parts
8 stars 5 forks source link

Cannot process OD Camp's address! #16

Open giacecco opened 9 years ago

giacecco commented 9 years ago

Try process:

Elizabeth II Court East
The Castle
Winchester
Hampshire
SO23 8UJ

"The Castle" will be lost. How come?

Floppy commented 9 years ago

Interesting; is it a street, or a locality do you know?

giacecco commented 9 years ago

Well it's the building where we're hosted for the event. I guess The Castle is supposed to be the street-equivalent, where Elizabeth II Court East is the PAO. @murraydata what do you reckon? Is this one of those examples where the address model we're using shows its limitations?

Floppy commented 9 years ago

Interestingly, 'The Castle' doesn't appear to be a street. That might be what's confusing it. If it can't find a valid street at all, then it probably breaks.

pezholio commented 9 years ago

Yeah, my guess is that The Castle is a non-vehicular street, which aren't in OS Locator. Maybe OS OpenMap will give us better coverage

cblanc commented 9 years ago

Hiya, I did a little digging on that address.

The Castle is the name of the building. The street is called Castle Avenue. I was able to find the street on OpenStreetMap (http://www.openstreetmap.org/way/19791923) although as @pezholio rightly points out - it seems to be a non-vehicular street.

I did a quick lookup (below) to confirm that OS Locator does not have a Castle Avenue for the vicinity inferred from the postcode.

FYI, Google Maps isn't aware of the street's existence either.

SELECT name, ST_Distance(location, ST_GeographyFromText('POINT(-1.31936876967578 \
51.0625511545148)')) AS distance FROM roads WHERE ST_DWithin(location, \
ST_GeographyFromText('POINT(-1.31936876967578 51.0625511545148)'), 500) \
ORDER BY distance;

         name          |   distance
-----------------------+---------------
 TRAFALGAR STREET      |  93.568067083
 PENINSULA SQUARE      | 103.195392587
 GAR STREET            | 110.573084663
 HIGH STREET           | 126.664739952
 ARCHERY LANE          | 135.049990647
 HIGH STREET           | 148.768688425
 SOUTHGATE STREET      | 150.051626693
 UPPER HIGH STREET     | 150.524238815
 BEAUMOND GREEN        | 156.030959329
 ST CLEMENT STREET     | 176.162624093
 ST JAMES' TERRACE     | 199.970965422
 CROSS STREET          | 215.818942909
 CONSTABLE'S GATE      | 225.399124555
 UPPER HIGH STREET     |  228.37868385
 CROWDER TERRACE       |  228.63919758
 ST THOMAS STREET      | 231.605033796
 KINGS HEAD YARD       | 248.659280921
 CLIFTON TERRACE       |  249.55331363
 SUSSEX STREET         | 262.412229053
 STAPLE GARDENS        | 263.302753963
 MINSTER LANE          | 263.713535892
 LITTLE MINSTER STREET | 265.183502288
 TOWER STREET          | 267.308542185
 ALEXANDRA TERRACE     | 269.183113894
 CLIFTON HILL          | 271.248293369
 ...truncated...
pezholio commented 9 years ago

This may be fixed when we can get access to OS Open Map, as I think their data is richer, but for now, there's not a lot we can do

peterkwells commented 9 years ago

We could but it needs the platform to be able to learn address building blocks from data received through APIs, although with lower confidence than address building blocks gathered from authoritative reference data.

That learning is the ultimate goal of the collaborative maintenance model.

pezholio commented 9 years ago

Oh yeah, totally, but that's out of scope, at least for now. It's a nice thing to aim for, but not something that can be achieved in the short term.