dask: dask array failing with numpy 1.17.0rc1 and np.min

Using Python 3.7, dask 2.0, (from conda-forge) and NumPy 1.17.0rc1 (from pip):

import numpy as np
import dask.array as da

image = da.from_array(np.array([[0, 1], [1, 2]]), chunks=(1, 2))
image_min = int(np.min(image))

results in:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<__array_function__ internals>", line 6, in amin
TypeError: no implementation found for 'numpy.amin' on types that implement __array_function__: [<class 'dask.array.core.Array'>]

About this issue

Original URL
State: closed
Created 5 years ago
Comments: 15 (13 by maintainers)

Commits related to this issue

Add amin/amax aliases to min/max in dask array Fixes #5031 — committed to mrocklin/dask by mrocklin 5 years ago
Add regression tests for gh-5031 (__array_function__ min/amin alias) — committed to rgommers/dask by rgommers 5 years ago
Deal with numpy aliases and functions and numpy functions not in dask See gh-5031 for discussion. — committed to rgommers/dask by rgommers 5 years ago
Deal with numpy aliases and functions and numpy functions not in dask See gh-5031 for discussion. Note that decorating all functions is not yet complete; this is enough to show tests in `test_numpy_c... — committed to rgommers/dask by rgommers 5 years ago
Change __array_function__ implementation for backwards compatibility (#5043) * Add regression tests for #5031 (__array_function__ min/amin alias) * Deal with numpy aliases and functions and numpy fu... — committed to dask/dask by rgommers 5 years ago

Most upvoted comments

Closing as Dask 2.1.0 is out which includes the fix from #5043. Thanks all!

jrbourbeau on Jul 8, 2019

Don’t think I like the decorator so much, that messes with stacklevel

That’s not true:

def matches_numpy_api(np_func):
    def decorator(my_func):
        _numpy_to_my_funcs[np_func] = my_func
        return my_func  # no wrapper created, so no arg mangling or stack addition
    return decorator

isn’t less typing than an explicit mapping

It’s not less typing, but it distributes the mapping alongside each function, rather than centralizing it. That means that it’s easier to spot a missing map entry.

The name is lousy, but it’s up to dask to pick that 😃

eric-wieser on Jul 1, 2019

It may not be too late to change TypeError with a warning + old behavior.

rgommers on Jul 1, 2019