PyGEOS is a C/Python library with vectorized geometry functions. The geometry operations are done in the open-source geometry library GEOS. PyGEOS wraps these operations in NumPy ufuncs providing a performance improvement when operating on arrays of geometries.
Note: PyGEOS is a very young package. While the available functionality should be stable and working correctly, it's still possible that APIs change in upcoming releases. But we would love for you to try it out, give feedback or contribute!
A universal function (or ufunc for short) is a function that operates on n-dimensional arrays in an element-by-element fashion, supporting array broadcasting. The for-loops that are involved are fully implemented in C diminishing the overhead of the Python interpreter.
The Geometry object
The pygeos.Geometry object is a container of the actual GEOSGeometry object. A C pointer to this object is stored in a static attribute of the Geometry object. This keeps the python interpreter out of the ufunc inner loop. The Geometry object keeps track of the underlying GEOSGeometry and allows the python garbage collector to free memory when it is not used anymore.
Geometry objects are immutable. Construct them as follows:
>>> from pygeos import Geometry >>> geometry = Geometry("POINT (5.2 52.1)")
Or using one of the provided (vectorized) functions:
>>> from pygeos import points >>> point = points(5.2, 52.1)
Compare an grid of points with a polygon:
>>> geoms = points(*np.indices((4, 4))) >>> polygon = box(0, 0, 2, 2) >>> contains(polygon, geoms) array([[False, False, False, False], [False, True, False, False], [False, False, False, False], [False, False, False, False]])
Compute the area of all possible intersections of two lists of polygons:
>>> from pygeos import box, area, intersection >>> polygons_x = box(range(5), 0, range(10, 15), 10) >>> polygons_y = box(0, range(5), 10, range(10, 15)) >>> area(intersection(polygons_x[:, np.newaxis], polygons_y[np.newaxis, :])) array([[100., 90., 80., 70., 60.], [ 90., 81., 72., 63., 54.], [ 80., 72., 64., 56., 48.], [ 70., 63., 56., 49., 42.], [ 60., 54., 48., 42., 36.]])
See the documentation for more: https://pygeos.readthedocs.io
Installation using conda
Pygeos requires the presence of NumPy and GEOS >= 3.5. It is recommended to install these using Anaconda from the conda-forge channel (which provides pre-compiled binaries):
$ conda install numpy geos pygeos --channel conda-forge
Installation using system GEOS
$ sudo apt install libgeos-dev
$ brew install geos
Make sure geos-config is available from you shell; append PATH if necessary:
$ export PATH=$PATH:/path/to/dir/having/geos-config $ pip install pygeos
Installation for developers
Ensure you have numpy and GEOS installed (either using conda or using system GEOS, see above).
Clone the package:
$ git clone https://github.com/pygeos/pygeos.git
Install it using pip:
$ pip install -e .[test]
Run the unittests:
If GEOS is installed, normally the
geos-config command line utility
will be available, and
pip install will find GEOS automatically.
But if needed, you can specify where PyGEOS should look for the GEOS library
before installing it:
On Linux / OSX:
$ export GEOS_INCLUDE_PATH=$CONDA_PREFIX/Library/include $ export GEOS_LIBRARY_PATH=$CONDA_PREFIX/Library/lib
On windows (assuming you are in a Visual C++ shell):
$ set GEOS_INCLUDE_PATH=%CONDA_PREFIX%\Library\include $ set GEOS_LIBRARY_PATH=%CONDA_PREFIX%\Library\lib
Relationship to Shapely
Both Shapely and PyGEOS are exposing the functionality of the GEOS C++ library to Python. While Shapely only deals with single geometries, PyGEOS provides vectorized functions to work with arrays of geometries, giving better performance and convenience for such usecases.
There is still discussion of integrating PyGEOS into Shapely (https://github.com/Toblerity/Shapely/issues/782), but for now PyGEOS is developed as a separate project.
- GEOS: http://trac.osgeo.org/geos
- Shapely: https://shapely.readthedocs.io/en/latest/
- Numpy ufuncs: https://docs.scipy.org/doc/numpy/reference/ufuncs.html
- Joris van den Bossche's blogpost: https://jorisvandenbossche.github.io/blog/2017/09/19/geopandas-cython/
- Matthew Rocklin's blogpost: http://matthewrocklin.com/blog/work/2017/09/21/accelerating-geopandas-1
Copyright & License
Copyright (c) 2019, Casper van der Wel. BSD 3-Clause license.