[OSM-talk] Zero tolerance on imports

Peter Budny peterb at gatech.edu
Mon Feb 21 20:58:32 GMT 2011


Nathan Edgars II <neroute2 at gmail.com> writes:

> Peter Budny wrote:
>> 
>> I really don't mind whether it's route relations or ref tags.  The
>> problem is that NEITHER is finished.  To get to my house, I have to get
>> on State Route 1966, then 1267.  Neither of these are marked as such on
>> the map, and I'm certainly not going to do it by hand when I could write
>> a tool that will fix those AND all the rest of the U.S. at the same
>> time.
>> 
> The big problem with your proposal to auto-generate relations from the TIGER
> tags is that, while TIGER in many/most areas is pretty good for alignment
> and street names, it's pretty horrible for routings of numbered routes in
> built-up areas and with being up-to-date on route changes. I suggest going
> county-by-county through the official listings at
> http://transportation.ky.gov/planning/reports/SPRS_listings/SPRS_listings.asp
> and marking ref tags on each by hand (rather simple after downloading a xapi
> query into JOSM), with FIXME tags where the routing is unclear. Then you can
> auto-generate relations from the ref tags. (Just remember that some routes,
> I think those numbered 6000+, are not signed.)

That's hardly the only problem with it.  There are lots of roads that
are single-carriageway when they should be dual-carriageway, or vice
verse.  Not all dual-carriageway ways are marked oneway=yes.  And yes,
TIGER doesn't always have the route on the correct roads (although I
would argue it's probably a minority of the cases).

The thing is, whether the work is done by humans or robots, how are you
going to know if it's right?  There are a couple tools for it, and
personally I would consider it easier to have the robot do all the
roads, then I compare them to the documents you linked and correct the
ones that are wrong.  If it's correct, it would probably be a lot less
effort to verify that than to create the route relation in the first
place.

But anyway... how do you know if the data is right?  OSM doesn't have
many tools for assisting humans in much of anything, whether it's
creating new data, or augmenting data with new data sources or with
information gleaned from examining metadata, or verifying data to make
sure it's actually accurate.

If automated edits are problematic, it's not because the robot
apocalypse is coming.  It's because automated edits are hacked together
due to a lack of tools and support in OSM for doing anything other than
manual editing.  This isn't a problem with automated edits; it's a
problem with OSM.
-- 
Peter Budny  \
Georgia Tech  \
CS MS student  \



More information about the talk mailing list