Construct `PolygonArray` or `MultiPolygonArray` from Exterior Coordinates

philipc2 commented 1 year ago

Closes #122

Overview

Introduces the PolygonArray.from_exterior_coords() and MultiPolygonArray.from_exterior_coords() class methods, which take-in an list/array of exterior coordinates equivalent to the following which are derived from Shapely Polygons/MultiPolygons

https://github.com/holoviz/spatialpandas/blob/dd941f2f5f6994c6f09996e9b8af315e16d00051/spatialpandas/geometry/polygon.py#L40-L42

https://github.com/holoviz/spatialpandas/blob/dd941f2f5f6994c6f09996e9b8af315e16d00051/spatialpandas/geometry/multipolygon.py#L40-L48

Usage

In UXarray, we use SpatialPandas paired with Datashader to visualize geoscience data that resides on unstructured grids as Polygons. Our unstructured grid is represented as a set of connectivity and coordinate arrays, which we would like to avoid needing to convert to Shapely, because we can derive the exterior of each Polygon quickly using the connectivity and coordinate information.

Performance Comparison

_Initial benchmarks, comparing across n_polygon & n_vertices_

All benchmarks are run on an NCAR Casper HPC Node, with 4 CPU Threads and 128gb of Memory

Test 1

(40,000,000 Closed Polygons with 7 total verticies)

Method / # of Polygons	400,000	4,000,000	40,000,000
Construct with Exterior Coord Array	1.02s	9.74s	97.0s
Create Shapely Polygons + Construct with Shapely Polygons	16.2s	159s	1560s
Only Construct with Shapely Polygons	15.2s	149s	1500s

Roughly a 16x speedup compared to using Shapely

philipc2 commented 1 year ago

Depending on any initial feedback from the developers here, I'd be willing to continue implementing this to support Polygons with holes, MultiPolygons, etc.

While this PR focusses on Polygons specifically, I believe other geometries could benefit from this type of construction support.

hoxbro commented 1 year ago

Thank you for the PR! :+1:

holoviz / spatialpandas