Bugs with Numpy

[tmorzade]

Some functions no longer work because of Numpy convention
For example : np.sum(np.abs(A * Omega))
It does not return a scalar but a vector

It could be solved using np.linalg.norm(A * Omega, 1)

[JulienRoussel77]

With which numpy version do you observe this issue? Due to numba dependency, numpy is limited to <2.1 and I did not observe this issue. I was not able to reproduce with more recent numpy version.

Moreover the norm used in RPCA is the L_1,1 (element-wise sum of coefficients absolute values), which is not implemented by numpy.linalg.norm (L_1 != L_1,1). Cf the Matrix norm wikipedia.

pyenv virtualenv 3.12.9 env_test_np
pyenv activate env_test_np
python -m venv .venv
source .venv/bin/activate
pip install --upgrade
pip install --upgrade pip
pip install numpy==2.3.2
python
Python 3.12.9 (main, Feb 11 2025, 00:29:11) [Clang 14.0.0 (clang-1400.0.29.202)] on darwin
Type "help", "copyright", "credits" or "license" for more information.

import numpy as np
A = np.eye(3,3)
A
array([[1., 0., 0.],
[0., 1., 0.],
[0., 0., 1.]])
np.sum(np.abs(A))
np.float64(3.0)
np.version
'2.3.2'