[FIXED] Scipy sosfilt ValueError ndarray is not C-contiguous

Issue

sosfilt from the scipy=1.9.3 library is giving me a strange ValueError when inputting a numpy=1.23.4 array.

MWE:

import numpy as np
from scipy.signal import sosfilt

fs=48000

rng = np.random.default_rng()
signal = rng.normal(size=(fs))

b_0k = [1.01589602025559, 0.958943219304445, 0.961371976333197,
        2.22580350360974, 0.471735128494163, 0.115267139824401,
        0.988029297230954, 1.95223768730136]
b_1k = [-1.92529887777608, -1.80608801184949, -1.76363215433825,
        -1.43465048479216, -0.366091796830044, 0.0, -1.91243380293387,
        0.162319983017519]
b_2k = [0.922118060364679, 0.876438777856084, 0.821787991845146,
        -0.498204282194628, 0.244144703885020, -0.115267139824401,
        0.926131550180785, -0.667994113035186]
a_0k = np.ones(len(b_0k))
a_1k = [-1.92529887777608, -1.80608801184949, -1.76363215433825,
        -1.43465048479216, -0.366091796830044, -1.79600256669201,
        -1.91243380293387, 0.162319983017519]
a_2k = [0.938014080620272, 0.835381997160530, 0.783159968178343,
        0.727599221415107, -0.284120167620817, 0.805837815618546,
        0.914160847411739, 0.284243574266175]

sos = np.array([b_0k, b_1k, b_2k, a_0k, a_1k, a_2k]).T

signalFiltered = sosfilt(sos, signal, axis=0)

gives me

signalFiltered = sosfilt(sos, signal, axis=0) Traceback (most recent call last):

Cell In[102], line 1 signalFiltered = sosfilt(sos, signal, axis=0)

File C:\ProgramData\Miniconda3\envs\devenv\lib\site-packages\scipy\signal_signaltools.py:4247 in sosfilt _sosfilt(sos, x, zi)

File _sosfilt.pyx:81 in scipy.signal._sosfilt._sosfilt

File stringsource:660 in View.MemoryView.memoryview_cwrapper

File stringsource:350 in View.MemoryView.memoryview.cinit

ValueError: ndarray is not C-contiguous

However

signal.flags

shows

C_CONTIGUOUS : True

F_CONTIGUOUS : True

OWNDATA : True

WRITEABLE : True

ALIGNED : True

WRITEBACKIFCOPY : False

I have also tried this on other signals imported from wav files, with the same error. Having looked around, there does not seem to be any known issue or other examples of this error using this function (although there are some machine learning package bugs discussed elsewhere)

What is the problem with Scipy?

Solution

It's the transpose that's giving problems, making it F contiguous. Add a .copy().

Look at the array flags, before and after transpose and copy.

transpose works by reversing the strides and shape. It's a view, a cheap way of doing the job, but messes with the continuity.

Answered By - hpaulj