Numpy masked operations failing due to comparison functions for __eq__

[scottstanie]

Using the Reader's .read method on a geotiff gives back ImageData. I expected to be able to run np.nansum on it, but I get ValueError: The truth value of an array with more than one element is ambiguous. .
Running np.sum works fine.

Reproducing script:

$ uv run --with numpy,rio-tiler  python demo-rio-tiler-error.py

# demo-rio-tiler-error.py
import rasterio
import numpy as np
from rasterio.transform import Affine

from rio_tiler.io import Reader

# Create a 32x32 numpy array with some data
data = np.ones((32, 32), dtype=np.float32)

# Add some random values
data = data * np.random.rand(32, 32) * 100

# Add some NaN values in a pattern
data[5:10, 5:10] = np.nan  # Small square of NaNs

# Define the transformation (here using a simple identity transform)
transform = Affine(1.0, 0.0, 0.0,
                  0.0, 1.0, 0.0)

# Create a new GeoTIFF file
output_path = "test_with_nans.tiff"

# Open a new file for writing
with rasterio.open(
    output_path,
    'w',
    driver='GTiff',
    height=data.shape[0],
    width=data.shape[1],
    count=1,
    dtype=data.dtype,
    crs="epsg:4326",
    transform=transform,
    nodata=np.nan
) as dst:
    dst.write(data, 1)

with Reader(output_path) as src:
    data = src.read(1)
    np.nansum(data)

Traceback

File ~/.cache/uv/archive-v0/0bGl3mVbNZ8ByG2ZVow-P/lib/python3.13/site-packages/numpy/lib/_nanfunctions_impl.py:726, in nansum(a, axis, dtype, out, keepdims, initial, where)
    634 @array_function_dispatch(_nansum_dispatcher)
    635 def nansum(a, axis=None, dtype=None, out=None, keepdims=np._NoValue,
    636            initial=np._NoValue, where=np._NoValue):
    637     """
    638     Return the sum of array elements over a given axis treating Not a
    639     Numbers (NaNs) as zero.
   (...)    724
    725     """
--> 726     a, mask = _replace_nan(a, 0)
    727     return np.sum(a, axis=axis, dtype=dtype, out=out, keepdims=keepdims,
    728                   initial=initial, where=where)

File ~/.cache/uv/archive-v0/0bGl3mVbNZ8ByG2ZVow-P/lib/python3.13/site-packages/numpy/lib/_nanfunctions_impl.py:102, in _replace_nan(a, val)
     98 a = np.asanyarray(a)
    100 if a.dtype == np.object_:
    101     # object arrays do not support `isnan` (gh-9009), so make a guess
--> 102     mask = np.not_equal(a, a, dtype=bool)
    103 elif issubclass(a.dtype.type, np.inexact):
    104     mask = np.isnan(a)

File ~/.cache/uv/archive-v0/0bGl3mVbNZ8ByG2ZVow-P/lib/python3.13/site-packages/attr/_make.py:1647, in __ne__(self, other)
   1642 def __ne__(self, other):
   1643     """
   1644     Check equality and either forward a NotImplemented or
   1645     return the result negated.
   1646     """
-> 1647     result = self.__eq__(other)
   1648     if result is NotImplemented:
   1649         return NotImplemented

File <attrs generated methods rio_tiler.models.ImageData>:20, in __eq__(self, other)
     17 if other.__class__ is not self.__class__:
     18     return NotImplemented
     19 return  (
---> 20     self.array == other.array and
     21     self.assets == other.assets and
     22     self.bounds == other.bounds and
     23     self.crs == other.crs and
     24     self.metadata == other.metadata and
     25     self.band_names == other.band_names and
     26     self.dataset_statistics == other.dataset_statistics and
     27     self.cutline_mask == other.cutline_mask
     28 )

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

[vincentsarago]

well ImageData is not a numpy object so I would expect np.nansum(data: ImageData) to fail.

if you do np.nansum(data.array) then it should work. I'm not sure why/how np.sum work for it as well

[scottstanie]

oh interesting, maybe it's some combination of numpy being open in what it tries to evaluate (due to ArrayLike) and ImageData having some methods which make it look "array like".

Something I noticed in the traceback: are there cases where self.array == other.array ever works? I would think that if they are indeed np.ndarrays, that statement would always trigger The truth value of an array with more than one element is ambiguous