dask: Broken FFT unit tests with mkl-2018.0.2

After upgrading anaconda to the latest version, dask.array.tests.test_fft.test_nd_fft_axes (git head) started failing, with the message:

Expected :dtype('complex128')
Actual   :dtype('complex64')

Tests OK: conda.ok.txt Tests fail: conda.broken.txt

Diff: < mkl=2018.0.1=h19d6760_4 < numpy=1.14.2=py36hdbf6ddf_0 < scipy=1.0.0=py36hbf646e7_0

> mkl=2018.0.2=1 > mkl_fft=1.0.1=py36h3010b51_0 > mkl_random=1.0.1=py36h629b387_0 > numpy=1.14.2=py36hdbf6ddf_1 > scipy=1.0.1=py36hfc37229_0

About this issue

Original URL
State: closed
Created 6 years ago
Comments: 15 (10 by maintainers)

Most upvoted comments

Is this still an issue? If so can anyone recommend a fix?

mrocklin on Oct 24, 2018

If I recall correctly, the numpy.fft functions always use double precision. Maybe mkl_fft supports single precision FFTs, leading to the unexpected dtype?

grlee77 on Apr 16, 2018