BUG: exception was ignored while fetching the attribute __array_interface__

[mwtoews]

This issue is essentially described here.

In a miniconda environment on any OS using a conda-forge channel with geopandas 0.9.0, shapely 1.7.1, pygeos 0.10.1 and numpy 1.21.0, take this example test_numpy_dep.py script:

import geopandas
from random import random
from shapely.geometry import Polygon

geoms = [Polygon((random(), random()) for _ in range(3)) for _ in range(3)]
gdf = geopandas.GeoDataFrame(geometry=geoms)

and run it via python -W all test_numpy_dep.py. The console will be flooded one warning per geometry, each showing:

C:\Users\mtoews\Miniconda3\envs\pyforge\lib\site-packages\geopandas_vectorized.py:106: DeprecationWarning: An exception was ignored while fetching the attribute __array_interface__ from an object of type 'Polygon'. With the exception of AttributeError NumPy will always raise this exception in the future. Raise this deprecation warning to see the original exception. (Warning added NumPy 1.21)

(similar on Linux, I'm just happen to be sitting at Windows desktop now).

The relevant NumPy release note is titled "Exceptions will be raised during array-like creation". I'll admit I don't 100% follow the expected changes.

Note there is a related shapely deprecation warning, see #1659. Also, shapely 1.8 (not released yet) will deprecate the array interface.

Output of geopandas.show_versions()

SYSTEM INFO
-----------
python     : 3.9.6 | packaged by conda-forge | (default, Jul 11 2021, 03:37:25) [MSC v.1916 64 bit (AMD64)]
executable : C:\Users\mtoews\Miniconda3\envs\pyforge\python.exe
machine    : Windows-10-10.0.17134-SP0

GEOS, GDAL, PROJ INFO
---------------------
GEOS       : None
GEOS lib   : None
GDAL       : 3.3.1
GDAL data dir: None
PROJ       : 8.0.1
PROJ data dir: C:\Users\mtoews\Miniconda3\envs\pyforge\Library\share\proj

PYTHON DEPENDENCIES
-------------------
geopandas  : 0.9.0
pandas     : 1.2.5
fiona      : 1.8.20
numpy      : 1.21.0
shapely    : 1.7.1
rtree      : 0.9.7
pyproj     : 3.1.0
matplotlib : 3.4.2
mapclassify: 2.4.2
geopy      : None
psycopg2   : 2.9.1 (dt dec pq3 ext lo64)
geoalchemy2: None
pyarrow    : None
pygeos     : 0.10.1

[mwtoews]

On further inspection, this issue may need to be fixed in Shapely, see shapely/shapely#1173

[martinfleis]

Thanks for the report! This shows up in our CI as well.

The optimal solution would be to fast track shapely 2.0 but I understand that will not happen :).

I guess the best way is to switch to AttributError as you suggest in shapely/shapely#1173. I am not sure we can fix this on geopandas side.

[erikmannerfelt]

This warning is also triggered in geoutils and xdem (which depend on GeoPandas) where we get more warnings than usual 😉

[jorisvandenbossche]

As we are already silencing Shapely deprecation warnings for Shapely 1.8+, we can also silence the numpy warning for older Shapely releases (it can reuse the same context manager that is used throughout our code base). -> #2072

For some more context see numpy/numpy#17965, numpy/numpy#18197 (they initially raise an error instead of a warning)