Feat: Vectorized orient function

[kylebarron]

Expected behavior and actual behavior.

It looks like the shapely.geometry.polygon.orient function is not yet vectorized and only works with a single polygon object

Steps to reproduce the problem.

import shapely
xmin = [0, 1]
ymin = [3, 4]
xmax = [5, 6]
ymax = [7, 8]
geoms = shapely.box(xmin, ymin, xmax, ymax, ccw=False)
shapely.geometry.polygon.orient(geoms)
# AttributeError: 'numpy.ndarray' object has no attribute 'exterior'

Operating system

MacOS

Shapely version and provenance

2.0.0 from pip

[idanmiara]

Hi @kylebarron
Please use the new vectorized function shaply.ops.orient added in #1690

[kylebarron]

It looks like shapely.ops.orient is already vectorized as of 2.0, and additionally is already much faster than a python for loop (as we'd expect, so not sure if #1690 is necessary?):

import shapely
xmin = [0, 1]
ymin = [3, 4]
xmax = [5, 6]
ymax = [7, 8]
geoms = shapely.box(xmin, ymin, xmax, ymax, ccw=False)
big_geoms = list(geoms) * 1000

%timeit shapely.ops.orient(big_geoms)
# 313 ns ± 7.58 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

%timeit [shapely.ops.orient(geom) for geom in big_geoms]
# 244 ms ± 4.66 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

I didn't know shapely.ops.orient existed; I'll move this to a doc request instead of a feature request. When I search for orient in the doc website, the only API doc result is for shapely.geometry.polygon.orient. Is there an API doc page in the website for everything in shapely.ops? Or is it laid out by intention, not by python module

[idanmiara]

It looks like shapely.ops.orient is already vectorized as of 2.0

Nope. It's just returns the original geom since it's not supported. just try your original test on it.

Is there an API doc page in the website for everything in shapely.ops? Or is it laid out by intention

There are many duplicated functions from in shapely 2 with fragmented documentation.
I'm working on improving the docs #1688

[kylebarron]

It's just returns the original geom since it's not supported

Oh that seems really bad: it's silently returning results with no indication that the orient didn't happen

[idanmiara]

It's just returns the original geom since it's not supported

Oh that seems really bad: it's silently returning results with no indication that the orient didn't happen

I think that the idea was that if the input is not BaseMultipartGeometry and not Polygon then it's a geometry that is oriented (not sure if that's true).
Now if you pass something that's not a Geometry, like a geometry array, it could be considered as an invalid input thus the output is unexpected.

With #1690 it works also for geometry array.

[jorisvandenbossche]

Yes, the existing shapely.ops.orient was indeed assuming you always pass a single geometry (and passing through anything that has no polygonal parts). But it is certainly unfortunate that it didn't at least check that it receives a geometry, to avoid such confusion.

Existing issue about providing a vectorized version of this: #1366

[jorisvandenbossche]

Duplicate of #1366

[idanmiara]

@kylebarron
Please see new implementation for force_ccw in #1690
I didn't time it, but it is probably a bit faster than the old orient.