[Tagging] Lyft and nameless sectioning in OSM

[Evan Carroll]

> I do not understand 'automatically generated'.  Landuse is about the
primary human use of the land, and that's something that has to be
obeserved, or come from another dataset (as an import) where it was
observed.

This is true if the landuse conveys _additional_ information, like a
name. But for unnamed landuse on developed land this is not true: the
dataset is OSM. If you an area with 100% detached residences inside,
it's a residential. Right? Always. No exceptions, it follows from the
detached residences inside. The same is true is commercial if it has
commercial buildings. Or Retail if it's shops. Computers can group
this just fine without further human observation. In fact, humans
can't really see "landuse" either. They infer it from whatever is on
the land. Some neighborhoods have signs with names, which is great
because you can add value with the name. But if you find yourself on a
cul de sac and can only see detached homes, you know the landuse is
residential.

> That's not landuse in the traditional geography sense; you can have
> hi-rise residential and hi-rise office and one is residential and the
> other commercial.

That's why it's a hypothetical, and not the actual thing I'm talking
about in the OT. The point is to show you another way to group
properties which can be entirely inferred.

> but for landuse it is perfectly reasonable for a
> mapper to observe that an area is filled with shops and add a
> landuse=retail polygon.  It may be that a building or so within that is
> commercial not retail, but that's ok.

Right, but they can have one building, two buildings or 49% of
buildings. You'll never know. Computers can also see the majority of
buildings are of a specific, they can also generate a label for the
group, and they can tell you to the degree the label is accurate! As
in: tag:landuse=residential; tag:computed_average=0.85.

> Part of the issue is that landuse should more or less follow property
> lines, unless there is some reason why not.  a several-acre parcel with
> a house and some trees is still landuse=residential on all of it, absent
> farming or some side industrial business.

Property lines isn't mentioned anywhere in the wiki. Land Use crosses
properties lines and groups buildings on different pieces of property.
If anything, ideally it's delineated by roads (which is explicitly
mentioned in the wiki). That's how the algorithm I provided works, it
looks for polygons that remain after you subtract out named landuse,
and roads.

--
Evan Carroll - me at evancarroll.com