search

dask / dask-image

Distributed image processing
http://image.dask.org/en/latest/
BSD 3-Clause "New" or "Revised" License
210 stars 47 forks source link

`import dask_image.imread` fails with `scipy.__version__`=1.14.1 #388

Open habi opened 3 weeks ago

habi commented 3 weeks ago

I'm working in a new conda environment, with the following environment.yaml file (mine contains more, but this is boiled down to nearly the minimum).

name: newenvironment
channels:
  - conda-forge
dependencies:
  - scipy
  - dask-image

In 'old' notebooks I was able to import dask-image with the call below

import dask_image.imread

(I know, this is probably not the way to do it, but it used to work for a long time :)

Now this fails with

AttributeError: `np.round_` was removed in the NumPy 2.0 release. Use `np.round` instead.

Environment:

habi commented 3 weeks ago 
from dask.distributed import Client
client = Client()

also does not work, with AttributeError: module 'numpy' has no attribute 'bool8'. Did you mean: 'bool'?

m-albert commented 3 weeks ago

Hi @habi, thanks for reporting the issue here.

Unfortunately, I can't reproduce the error using the provided environment file.

AttributeError: np.round_ was removed in the NumPy 2.0 release. Use np.round instead. AttributeError: module 'numpy' has no attribute 'bool8'. Did you mean: 'bool'?

These two error messages hint towards some package in your environment (which dask-image depends on) not being compatible with numpy >= 2.

For debugging it, it'd be useful to have

habi commented 3 weeks ago

a larger part of the error traceback you get (for seeing which package contains the np.round_)

This is the full traceback:

(newenvironment) [habi@anaklin25|dask-issue]:python
Python 3.13.0 | packaged by conda-forge | (main, Oct  8 2024, 20:04:32) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import dask_image.imread
Traceback (most recent call last):
  File "<python-input-0>", line 1, in <module>
    import dask_image.imread
  File "/home/habi/miniconda3/envs/newenvironment/lib/python3.13/site-packages/dask_image/imread/__init__.py", line 6, in <module>
    import dask.array as da
  File "/home/habi/miniconda3/envs/newenvironment/lib/python3.13/site-packages/dask/array/__init__.py", line 2, in <module>
    from dask.array import backends, fft, lib, linalg, ma, overlap, random
  File "/home/habi/miniconda3/envs/newenvironment/lib/python3.13/site-packages/dask/array/ma.py", line 8, in <module>
    from dask.array.routines import _average
  File "/home/habi/miniconda3/envs/newenvironment/lib/python3.13/site-packages/dask/array/routines.py", line 1569, in <module>
    @implements(np.round, np.round_)
                          ^^^^^^^^^
  File "/home/habi/miniconda3/envs/newenvironment/lib/python3.13/site-packages/numpy/__init__.py", line 400, in __getattr__
    raise AttributeError(
    ...<3 lines>...
    )
AttributeError: `np.round_` was removed in the NumPy 2.0 release. Use `np.round` instead.
>>>

(I saw the problem in a Jupyter notebook, but the error also pops up in Python proper)

the full environment.yml

(newenvironment) [habi@anaklin25|dask-issue]:cat environment.yaml 
name: newenvironment
channels:
  - conda-forge
dependencies:
  - scipy
  - dask-image
(newenvironment) [habi@anaklin25|dask-issue]:

the output of pip list

(newenvironment) [habi@anaklin25|dask-issue]:pip list
Package           Version
----------------- ---------
bokeh             2.4.3
Brotli            1.1.0
cffi              1.17.1
click             8.1.7
cloudpickle       3.1.0
cytoolz           1.0.0
dask              2023.3.0
dask-image        0.0.0
distributed       2023.3.0
fsspec            2024.9.0
h2                4.1.0
hpack             4.0.0
hyperframe        6.0.1
imagecodecs       2024.6.1
imageio           2.36.0
Jinja2            3.1.4
locket            1.0.0
lz4               4.3.3
MarkupSafe        3.0.1
msgpack           1.1.0
numpy             2.1.2
packaging         24.1
pandas            2.2.3
partd             1.4.2
pillow            11.0.0
PIMS              0.7
pip               24.2
psutil            6.0.0
pycparser         2.22
PySocks           1.7.1
python-dateutil   2.9.0
pytz              2024.1
PyYAML            6.0.2
scipy             1.14.1
six               1.16.0
slicerator        1.1.0
sortedcontainers  2.4.0
tblib             3.0.0
tifffile          2024.9.20
toolz             1.0.0
tornado           6.4.1
typing_extensions 4.12.2
tzdata            2024.2
urllib3           2.2.3
zict              3.0.0
zstandard         0.23.0
jakirkham commented 3 weeks ago

Can you please update dask? Looks like it and associated packages are coming from 2023. Expect those are not compatible with NumPy 2. Would update to something much more recent

habi commented 2 weeks ago

That (obviously) fixed the issue. In this case I seem to misunderstand conda environments, as I expected them to be self-contained. E.g. when I have dask-image in it, this also "pulls" a recent version of dask.

(I don't understand why this environment file installs dask=2023.3.0, as the machine I'm installing this on physically did not exist prior to July 2024 :) )

m-albert commented 2 weeks ago

Great that at least you have a working environment again :)

However what's weird is that dask_image is listed with version 0.0.0 in pip list. I wonder whether there's a problem with the conda feedstock, e.g. related to this.