Skip to content

test_axes.py::test_csd_freqs failing with numpy 1.15.0 on macOS #11766

Closed

Description

@dstansby

There's a really small change in the generated image, leaving this here to track investigations into why.

Activity

  1. matthew-brett commented on Jul 24, 2018

    @matthew-brett
    Contributor

    We recently changed to using OpenBLAS for the Mac wheels - maybe that's the explanation?

  2. dstansby commented on Jul 24, 2018

    @dstansby
    MemberAuthor

    Here's what I get locally:
    csd_freqs

    And here's the expected image:
    csd_freqs

    There's no noticeable change in the diff, so as @jklymak suggested I think it's safe just to add a small tolerance to this test.

  3. jklymak commented on Jul 24, 2018

    @jklymak
    Member

    Personally I think we should deprecate csd, psd and friends... We shouldn't have our own version of the computation of spectral statistics, and the plots are trivial after the the spectra are calculated.

  4. dstansby commented on Jul 24, 2018

    @dstansby
    MemberAuthor

    I completely agree! I tried to work out a good way to do all the deprecation, but it would require a major breaking API change, so requires quite a bit of thought at work. I'm +1 on getting rid of the rest of mlab, but think keeping ax.csd etc. is okay as long as they use scipy or numpy to do the underlying computations.

  5. jklymak commented on Jul 24, 2018

    @jklymak
    Member

    I'd still deprecate axes.psd, axes.csd, axes.magnitude_spectrum, axes.angle_spectrum, axes.phase_spectrum, axes.cohere, and axes.specgram. I think a plan discussed at some point was to move them into a toolbox to give folks a straight-to-plot method of getting these statistical plots. I think that'd make deprecation pretty straight forward as all the mlab and axes methods could just be moved to there and called from the deprecated methods. Then the maintainer of mpl_spectral could decide how to eventually get rid of the stuff brought over from mlab.

    End-users would call

    import mpl_toolkits.mpl_spectral as mpls
    
    fig, ax = plt.subplots()
    mpls.psd(ax, x, nfft=1024)
  6. tacaswell commented on Jul 25, 2018

    @tacaswell
    Member

    Probably the right way to do this is to talk to the maintainers of the scipy spectral functions and develop a version of these that uses scipy and put them in their own package.

  7. added this to the v3.0 milestone on Jul 25, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions