LinAlgError: SVD did not converge

LinAlgError: SVD did not converge

$ python -c "import numpy as np; import numpy.linalg as la; a=np.random.rand(1000,1000); la.svd(a)"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
numpy.linalg.LinAlgError: SVD did not converge

Why this happens

SVD algorithms can fail to converge on ill-conditioned or extremely large matrices, or when numerical stability limits are reached.

Fix

Precondition the matrix, reduce its size (truncate or use randomized SVD from sklearn or scipy), or try using more stable algorithms (e.g., use scipy.linalg.svd with different options).

Wrong code

import numpy as np
import numpy.linalg as la
a = np.random.rand(1000,1000)
u,s,v = la.svd(a)

Fixed code (use smaller/regularized example)

import numpy as np
import numpy.linalg as la
from scipy.sparse.linalg import svds

# use smaller matrix or randomized SVD
a = np.random.rand(200,200)
u,s,v = la.svd(a)

# or use svds for large sparse-like matrices
# u,s,v = svds(a, k=10)