SciPy SVD vs Numpy SVD

Question

Both SciPy and Numpy have built-in functions for decomposing singular values ​​(SVDs). The teams are mainly scipy.linalg.svd and numpy.linalg.svd . What is the difference between the two? Are any of them better than the other?

Answer 1

Besides checking errors, the actual work seems to be done in lapack both with numpy and scipy .

Without benchmarking, I think that performance should be the same.

Answer 2

The FAQ page says that the scipy.linalg submodule provides a more complete wrapper for the Fortran LAPACK library, while numpy.linalg tries to build independently of LAPACK.

I did some tests for various implementations of svd functions and found that scipy.linalg.svd is faster than a simple example:

However, jax is wrapped in NumPy, otherwise jax.numpy.linalg.svd even faster:

A full test pad is available here .