Cannot import name '_centered' from 'scipy.signal.signaltools'

Question

Unable to import functions from scipy module.

Gives error :

from scipy.signal.signaltools import _centered
Cannot import name '_centered' from 'scipy.signal.signaltools'

scipy.__version__
1.8.0

Answer 1

I encountered the same problem while using statsmodels~=0.12.x. Increasing the statsmodels package to version 0.13.2, this import issue is resolved.

UPDATE with more notes:

• before:
  • installation of fixed version of statsmodels==0.12.2 which is dependent on scipy
  • there was newly released scipy==1.8.0 - 2022-02-05
    • when installing it, got this problem:

    from statsmodels.tsa.seasonal import seasonal_decompose
  File "/usr/local/lib/python3.8/site-packages/statsmodels/tsa/seasonal.py", line 12, in <module>
    from statsmodels.tsa.filters.filtertools import convolution_filter
  File "/usr/local/lib/python3.8/site-packages/statsmodels/tsa/filters/filtertools.py", line 18, in <module>
    from scipy.signal.signaltools import _centered as trim_centered
ImportError: cannot import name '_centered' from 'scipy.signal.signaltools' (/usr/local/lib/python3.8/site-packages/scipy/signal/signaltools.py)

• after:

  • when bumping up statsmodels to the latest version available 0.13.2 release 2022-02-08, it works

• If you are not using statsmodels but other package which is dependent on scipy, have a look if there is newer version available (after the release of scipy to v.1.8.0)

Answer 2

If you need to use that specific version of statsmodels 0.12.x with scipy 1.8.0 I have the following hack. Basically it just re-publishes the existing (but private) _centered function as a public attribute to the module already imported in RAM.

It is a workaround, and if you can simply upgrade your dependencies to the latest versions. Only use this if you are forced to use those specific versions.

import  scipy.signal.signaltools

def _centered(arr, newsize):
    # Return the center newsize portion of the array.
    newsize = np.asarray(newsize)
    currsize = np.array(arr.shape)
    startind = (currsize - newsize) // 2
    endind = startind + newsize
    myslice = [slice(startind[k], endind[k]) for k in range(len(endind))]
    return arr[tuple(myslice)]

scipy.signal.signaltools._centered = _centered