[FEA]: Support binary predicates with GeoSeries - shapely.geometry object
#1316
Comments
GPUtester
added
Needs Triage
|
Hi @vboussange! Thanks for submitting this issue - our team has been notified and we'll get back to you as soon as we can! |
isVoid
added
Python
|
Hi @vboussange , thank you for making this issue! Per the documentation of polys = cuspatial.GeoSeries([polygon] * len(s))
s_cu = cuspatial.from_geopandas(s)However, this doesn't get you there yet. cuSpatial also has a limitation of only supporting binary predicates between two "single geometry type" geoseries for the moment. import cudf
import cuspatial
from cuspatial.core._column.geometa import Feature_Enum
import geopandas
from shapely.geometry import Polygon, LineString, Point
def extract_single_geometry_from_series(
s: cuspatial.GeoSeries,
ty: cuspatial.core._column.geometa.Feature_Enum
):
"""Extract objects of type `ty` from GeoSeries `s`."""
types = s._column._meta.input_types
ilocs = cudf.Series(cudf.core.column.arange(len(types)))
df = cudf.DataFrame(data={"types": types, "ilocs": ilocs})
selected = df[df.types == ty.value].ilocs
return s.iloc[selected]
s = geopandas.GeoSeries(
[
Polygon([(0, 0), (2, 2), (0, 2)]),
Polygon([(0, 0), (1, 2), (0, 2)]),
LineString([(0, 0), (0, 2)]),
Point(0, 1),
],
)
polygon = Polygon([(0, 0), (2, 2), (0, 2)])
s_cu = cuspatial.from_geopandas(s)
tys = [Feature_Enum.POLYGON, Feature_Enum.LINESTRING, Feature_Enum.POINT]
partials = []
for ty in tys:
ss_cu = extract_single_geometry_from_series(s_cu, ty)
# Caching the index of the single type objects
idx = ss_cu.index
ss_cu = ss_cu.reset_index(drop=True)
polys = cuspatial.GeoSeries([polygon] * len(ss_cu))
# Compute within of single type geoseries
partial = ss_cu.within(polys)
# Restore index and cache partial results
partial.index = idx
partials.append(partial)
# Combine all results.
result = cudf.concat(partials).sort_index()
print(result)Output: Please beware that this code will behave significantly slower than GeoPandas due to multiple kernels are launched on a small number of data. The kernel launching overhead dominates the latency. However, if you have a very large dataset, you could benefit from compute speedup. |
isVoid
changed the title
within not working for single polygonGeoSeries - shapely.geometry object
isVoid
added
feature request
|
Hey there, thanks a lot for the explanation and the workaround. For other points = geopandas.GeoSeries([Point(0,1), Point(2,1), Point(4,2)])
polygons = geopandas.GeoSeries([Polygon([(0, 0), (2, 2), (0, 2)]), Polygon([(0, 0), (1, 2), (0, 2)])])
points_cu = cuspatial.from_geopandas(points)
polygons_cu = cuspatial.from_geopandas(polygons)
cuspatial.point_in_polygon(points_cu, polygons_cu)output |
|
Cool! Note that |
|
@isVoid should we close this now, or keep it open? |
|
I would like to keep it open as a feature request issue to support |
|
It should be an easy addition for short-term support - a single shapely geometry object can be placed in a GeoSeries and run as such. |
Version
23.12.01
On which installation method(s) does this occur?
Conda
Describe the issue
Method
withindoes not work as expected when used with a single polygon object.Minimum reproducible example
Relevant log output
`AttributeError: 'Polygon' object has no attribute 'column_type'`Environment details
Other/Misc.
No response
The text was updated successfully, but these errors were encountered: