linalg.svd fails with a normal np.ndarray #9620
Description
So I came up with an np.ndarray which fails when I try to factorize it to Singular Value Decomposition. It does not contain NaNs, just float64. It raises LinAlgError: SVD did not converge, but I don't know why.
Reproducing code example:
import numpy as np
from scipy import linalg
path = 'problematic_array.npy'
X = np.load(path)
np.isnan(X).any()
# False
U, s, Vh = linalg.svd(X)
Error message:
LinAlgError Traceback (most recent call last)
<ipython-input-16-119344cda5b9> in <module>
9 # False
10
---> 11 U, s, Vh = linalg.svd(X)
~/.local/lib/python3.6/site-packages/scipy/linalg/decomp_svd.py in svd(a, full_matrices, compute_uv, overwrite_a, check_finite, lapack_driver)
130
131 if info > 0:
--> 132 raise LinAlgError("SVD did not converge")
133 if info < 0:
134 raise ValueError('illegal value in %d-th argument of internal gesdd'
LinAlgError: SVD did not converge
...
Scipy/Numpy/Python version information:
Scipy version: 1.2.0
Numpy version: 1.15.4
Python version: sys.version_info(major=3, minor=6, micro=7, releaselevel='final', serial=0)
I am not sure how to share that ndarray with you. It is inside the following zip in an npy file. The reproducing code example should read it without problems.
problematic_array.zip
If it helps, it comes from a bootstrap of a resampling of MNIST