Greg Troxel wrote:
> 
> The real difficulty is that different cyclists have different tolerances
> for riding in traffic.  What I consider ok is very different than what
> some others I cycle with consider ok.  The essence of tagging for
> cycle-friendliness of roads is to capture this judgement in a way that
> is consistent.
> [..]
> The real goal is that when I get to a -1 road then my mental projection
> of comfort from knowing -1 matches my experience.  Obviously impossible,
> but the closer the better.

After fifteen years of riding across Paris, I have developed a pretty 
good mental model of the city. I would not be capable of describing my 
routing algorithm offhand, but it features (in decreasing order of 
objectivity) distances, elevations, surfaces, waviness, congestion, 
shopping opportunities, motorists behavior and the likelihood of pretty 
girls walking by. The first items can be encoded easily, the middle ones 
could probably be encoded... But are you sure you want my highly 
subjective sexual preferences in the main database ? Subjective data is 
by essence not consensual. Including it can only lead to discord.

But wait, there is a way : stuff as much objective data in the base and 
let algorithms decide what to do with it and user preferences. Including 
shopping opportunities in the routing algorithm ? Just feed you shopping 
preferences in the algorithm and let it compute the route considering 
some gravitational attraction toward the relevant points of interest.

To me, the question about whether or not to include subjective data 
appears to be answered by enforcing separation between consensual 
objectivity in the common database and personal subjectivity in the 
algorithms that exploit it - and it sounds like what I understand of the 
current OSM doctrine.

Felix, don't whine about the database being motorist-centric : feed 
cycling relevant data into it and push innovative algorithms ! Others 
are doing it already.

_______________________________________________
talk mailing list
talk@openstreetmap.org
http://lists.openstreetmap.org/listinfo/talk

Reply via email to