Encountered some issues when using 'voronoi_polygons'

[shiyue-code]

Is this a bug?

Using voronoi_polygons(road) resulted in a memory usage of over 10 GB and then was forcibly terminated by the OOM killer. The road consists of approximately 4.6 million points forming polygons with holes. However, there were no issues encountered when using from scipy.spatial import Voronoi, and the memory usage was minimal.

Steps to reproduce the problem.

First import the geometry from the ’polygon.wkb‘ file. The file's cloud drive link is file: https://drive.google.com/file/d/14pH9wpJRURtokWLc_rpvo2jykLDSLN0-/view?usp=drive_link
This is actually a very complex road
Then use the following code to test if it will result in an OOM.

voronoied = voronoi_polygons(borders,only_edges=True) #equivalent to the scipy.spatial.Voronoi

scipy.spatial .Voronoi is ok

from scipy.spatial import Voronoi
points = []
exterior_points = list(borders.exterior.coords)
interior_points = []
for interior in borders.interiors:
    interior_points.extend(list(interior.coords))
points += exterior_points + interior_points
v = Voronoi(np.array(points))

Operating system

No LSB modules are available.
Distributor ID: Ubuntu
Description: Ubuntu 22.04.3 LTS
Release: 22.04
Codename: jammy

Shapely version and provenance

shapely 2.0.3

[jorisvandenbossche]

To read the file (@shiyue-code it would be helpful to include that in your report, it's not necessarily clear what exactly the file exists of):

import shapely

with open("polygon.wkb", "rb") as f:
    borders = shapely.from_wkb(f.read())

I can reproduce this, causing a lot of memory usage and then (in my case, at the point of taking around 30GB hitting the limit of my system) killing itself.

It is a lot of points though, and GEOS builds up indices on top of this to determine the resulting polygons I assume, so honestly I have no clue if this is "expected" and GEOS simply cannot really handle that much of points given the way the algo works, or if there is actually something GEOS could do to improve memory usage. In either case, it's an upstream GEOS issue, so I would recommend reporting it there (https://github.com/libgeos/geos/issues)

[shiyue-code]

To read the file (@shiyue-code it would be helpful to include that in your report, it's not necessarily clear what exactly the file exists of):

import shapely

with open("polygon.wkb", "rb") as f:
    borders = shapely.from_wkb(f.read())

I can reproduce this, causing a lot of memory usage and then (in my case, at the point of taking around 30GB hitting the limit of my system) killing itself.

It is a lot of points though, and GEOS builds up indices on top of this to determine the resulting polygons I assume, so honestly I have no clue if this is "expected" and GEOS simply cannot really handle that much of points given the way the algo works, or if there is actually something GEOS could do to improve memory usage. In either case, it's an upstream GEOS issue, so I would recommend reporting it there (https://github.com/libgeos/geos/issues)

Thank you for your response, I reported the issue in GEOS.